The NEXT LAST R600 Rumours & Speculation Thread

[neliz]

A15 was me funnin'. Somebody or other was claiming that a few months back. :smile: I see what you see on that chip shot. I suspect the Hexus piece was closer to when the order for A13 went to the fab than when it came back. . .

http://www.theinquirer.net/default.aspx?article=37718

Inq said in feb. that A13 revision production would begin in the last week of february.
Since the article is from feb 19th and the dieshots we've seen so far are from Week 7 it looks like he was only three weeks off...

Dailytech had an article in January about A13 being production silicon .. they also said 610 and 630 would be released in June at its earliest..
http://www.dailytech.com/article.aspx?newsid=5630
They were also stating that level505 used the A12 core

[Jawed]

Forget about it Jawed. You don't get it. This isn't about % ALU utilization on R300 vs. R600. It's about the importance of compiler co-issue for competitiveness.
And what's annoying me is that you think ATI has never written a co-issuing compiler ever before. R300 is a 4-issue ALU. R600 is 5-issue. R300's four instructions are all different (either in component count or capability or both). 4 of R600's ALUs are identical. The sky isn't falling in.

The extra ADD doesn't use much die space on R300, nor does its usage make much difference in most shader code. Neither applies to R600's coissue. NVidia said somewhere they get nearly 2x speed boost by going scalar, and looking at PS benchmarks I believe them.
NVidia also moved texture address calculation out of the PS pipeline. And on top of that, the residual asymmetry in the pipeline (look at SF and the pp_NRM) and the register fetch bandwidth limitation means that it is a dog's dinner. As I said earlier, NV4x/G7x's badness intensifies G80's goodness.

Why do we need more sequencers? You're really twisting my Xenos example around with the whole smaller Xenos quadrupled interpretation.

Forget about that example for now. Take R600, and assume the batch size is 64 pixels. Now make the single ALU array per quarter act on one channel of all pixels instead of all channels of 16 pixels. That's pretty much the sum of it (ignoring the SF details).
1-clock instructions are extremely costly...

----------------

HOWEVER, I realized that I forgot something. Latency requires you to cycle between batches. An ALU array in R600 takes 4 cycles to get through a batch, so a 12-cycle instruction latency, for example, requires you to have three batches in a cycle that switches every 4 clocks. My design would need 12 batches in the the cycle that switches every clock. That's a lot more data shuffling.
I'm glad the penny's dropped. This is just one way that your sequencer complexity increases.

If instead of combining a 64-pixel ALU with a single-clock instruction pipeline your sequential scalar GPU has 16-pixel ALUs and four-clock instructions, you've still got increased sequencer complexity compared against R600, because you've just multiplied the number of batches in flight 4x in order to retain the same batch size (this is the "four Xenoses glued together" scenario).

So, now muse on how much of G80 is batch sequencing logic, since it has 16 batches in flight and compare that against the 4 batches in R600.

I guess there is indeed a legitimate reason for going the co-issue route.
And, seemingly, it isn't starkly different from R5xx, either. Except that operand fetching is rather more interesting. (As is constant fetching, vertex fetching and texture fetching :smile: ).

Jawed

[Sound_Card]

http://www.imagebanana.com/img/b11ska11/xh29.jpg



:shock:

:drools:

[nAo]

:drools:
You're leaking the same liquid that was used to cool that thing :)

[nAo]

So, now muse on how much of G80 is batch sequencing logic, since it has 16 batches in flight and compare that against the 4 batches in R600.
That's one of the reasons why, imho, G80 has not completely decoupled TMUs, while R600 is likely to have TMUs that can serve any ALU( you want to limit your on chip traffic ).

[Domell]

waouuu impressive ! kudos ATI/AMD for bringing better product than 6 months old G80 with better process :roll:
well if it's true, I hope ATI/AMD will enjoy these coming 3 months of performance lead because when G92 will arrive, R600/650 will look like a toy, same way as R580 was a toy compared to G80...

What is G92??

[Geo]

http://www.theinquirer.net/default.aspx?article=37718

Inq said in feb. that A13 revision production would begin in the last week of february.
Since the article is from feb 19th and the dieshots we've seen so far are from Week 7 it looks like he was only three weeks off...

Dailytech had an article in January about A13 being production silicon .. they also said 610 and 630 would be released in June at its earliest..
http://www.dailytech.com/article.aspx?newsid=5630
They were also stating that level505 used the A12 core


I surrender already! Can't a fellow engage in a little VR Zone bashing anymore? :cool:

http://www.vr-zone.com/?i=4510

[Rangers]

The SM3.0 are HDR tests and that's the scenario where you'd expect superior bandwidth of XT to come through. It should be particularly evident in texture-light, HDR-heavy Deep Freeze.

It should be more like that drastically texture limited R600 can process HDR textures much faster according to that slide.

Textures is again probably the only reason R600 can pull well in 3Dmark but not real games, as I was informed that bench only stresses shaders/vertices mostly. Beyond that it looks like Nvidia has trumped them in several other possible areas as well.

Bandwidth isn't R600's problem.. everything else is, such that it doesn't even need the BW it has. It's very unbalanced in more ways than one. Unbalanced is how you get the consistent die-size/performance defecit ATI has been creating the last two gens. Last gen they were hugely bigger for similar performance, this time they look to be more similar in size (if not still bigger) while performing much lower. The unbalancing really shows up in what a hot power monster R600 is for subpar games performance. It's not a poor performer for any lack of transistors firing, that's for sure.

This architecture looks pretty much toast. AMD's engineering still looks ace from the Barcelona tidbits, so either by next gen they'll slap ATI into shape or they're pretty much done for.

We dont have final benches but on all accounts R600 is going to be a real disappointment. ATI has never been in this bad of a situation before, with clear performance inferiority to go along with their worsening lateness. Inexcusable. ATI's engineering is simply very poor.

Oh and to add insult to all the other injury, at least last time ATI even had R580 waiting in the wings on a non-delayed schedule. This time they have nothing of the sort, as even their delays got delayed apparently.

[Jawed]

That's one of the reasons why, imho, G80 has not completely decoupled TMUs, while R600 is likely to have TMUs that can serve any ALU( you want to limit your on chip traffic ).
In what sense "not completely decoupled"? G80's TMUs have to serve 2 independent ALU arrays within a cluster. Something like the Xenos model of 3 independent ALU arrays taking it in turns to use the single TMU array (or the single VF array). Confused what you mean...

I expect R600 is like R5xx, in that a TU is only usable within the "quarter" of R600 that it finds itself in. e.g. the batches that are localised to shader unit 1 can execute on ALU pipes 1-16 and TU 1 (8TAs, 4TFs, 20 samplers). Shader unit 2 has ALU pipes 17-32 and TU 2 exclusively for its own batches, etc.

Apart from screen space tiling you also have register file connectivity to account for in submitting batches to the TUs and receiving results back from them.

It seems unlikely to me that a TU in shader unit 1 could serve texels to batches running in shader unit 3. The cache connectivity, the L2/L3 thing (if I'm right about the patent application) should provide for that kind of distribution - but obviously that costs extra processing within each local TU.

Jawed

[Sound_Card]

http://img66.imageshack.us/img66/782/cooledr6001qc1.jpg

The system used to hit that 1ghz clock and that 16k 3dmark o6 score.:wink:

[Sound_Card]

cooling solution looks rather primitive if you ask me. A water kit + volt mod should get me up to 1.2ghz +:wink:

Hmm, how neat it would be if I can get the core up to the same freq of G80's shaders?

(I love overclocking)

[Razor1]

cooling solution looks rather primitive if you ask me. A water kit + volt mod should get me up to 1.2ghz +:wink:

Hmm, how neat it would be if I can get the core up to the same freq of G80's shaders?

(I love overclocking)

Primative? Thats a cpu cooler and a 12cm fan on the other side, ya know how much more air those things move?

[ninjaspa]

:shock:

:drools:

yeah 16k on 06 is about sick as it gets...

but sound card, what is this quote in your sig about? something that can't be disclosed til next week? is there any better hint?

[SugarCoat]

i dont mean to rant or anything....well actually i do

3DMARK SCORES ARE MEANINGLESS STOP PRAISING THEM.


One of the main reasons i liked these forums in the past was the lack of beating 3Dmark scores to death everytime a random number was leaked in conjunction to a new peice of hardware. Is this a new trend? Maybe its just me but it looks ignorant!

[Kaotik]

Primative? Thats a cpu cooler and a 12cm fan on the other side, ya know how much more air those things move?

Primitive doesn't mean it couldn't be powerful cooling solution too :wink:
But you gotta admit it looks primitive in a "cavemen threw a heatsink here, fan there" kinda way?

[Razor1]

Primitive doesn't mean it couldn't be powerful cooling solution too :wink:
But you gotta admit it looks primitive in a "cavemen threw a heatsink here, fan there" kinda way?


True, actually that heatsink looks like stock. At least one variation of one.

[Sound_Card]

Primative? Thats a cpu cooler and a 12cm fan on the other side, ya know how much more air those things move?

I'm aware how much air that thing moves. Im left wondering how good that does blowing on the back of the board. How is that CPU cooler better than the default cooler? The heat sink and heat pipes look much smaller and as well as the fan.

I don't know, maybe he has a special method.

[Geo]

Primitive doesn't mean it couldn't be powerful cooling solution too :wink:
But you gotta admit it looks primitive in a "cavemen threw a heatsink here, fan there" kinda way?

I'm claiming IP on this idea, but clearly they should sell "designer classic hot-rod sound" high-end models to mask the cooler. Get your R600 with a '64 Pontiac GTO exhaust note. . . it's not the loud, it's the loud and whiny. :lol: The Euro boys can have F1 team sounds.

[Sound_Card]

yeah 16k on 06 is about sick as it gets...

but sound card, what is this quote in your sig about? something that can't be disclosed til next week? is there any better hint?

You know I don't even know for sure!!!:razz:

My best guess has been a wide card launch. We won't know for sure untill the 14th. Their has been some other intresting quotes from Kombatant as well if you want the links to them.:cool:

[Razor1]

I'm aware how much air that thing moves. Im left wondering how good that does blowing on the back of the board. How is that CPU cooler better than the default cooler? The heat sink and heat pipes look much smaller and as well as the fan.

I don't know, maybe he has a special method.


That CPU fan is bigger then the stock fan, also the fins of the CPU fan have more surface area it can push more air in the same rotation. Its an open air unit, it would be very hard to do the same in a closed box. Also remember the shim Kinc was talking about, probably that and no custom air coolers yet.

[DemoCoder]

And what's annoying me is that you think ATI has never written a co-issuing compiler ever before. R300 is a 4-issue ALU. R600 is 5-issue. R300's four instructions are all different (either in component count or capability or both).


Umm, maybe I'm confused, but I thought the R300 could only co-issue a vec3+scalar. I hope your not trying to claim that any vector ops == co-issue, that's rather twisted semantics that would make any DX8 card, or even DX7 register combiners "co-issue" GPUs.

[Rangers]

That CPU fan is bigger then the stock fan, also the fins of the CPU fan have more surface area it can push more air in the same rotation. Its an open air unit, it would be very hard to do the same in a closed box. Also remember the shim Kinc was talking about, probably that and no custom air coolers yet.

It's still air cooling. Period.

[Farhan]

In what sense "not completely decoupled"? G80's TMUs have to serve 2 independent ALU arrays within a cluster. Something like the Xenos model of 3 independent ALU arrays taking it in turns to use the single TMU array (or the single VF array). Confused what you mean...

I expect R600 is like R5xx, in that a TU is only usable within the "quarter" of R600 that it finds itself in. e.g. the batches that are localised to shader unit 1 can execute on ALU pipes 1-16 and TU 1 (8TAs, 4TFs, 20 samplers). Shader unit 2 has ALU pipes 17-32 and TU 2 exclusively for its own batches, etc.

Apart from screen space tiling you also have register file connectivity to account for in submitting batches to the TUs and receiving results back from them.

It seems unlikely to me that a TU in shader unit 1 could serve texels to batches running in shader unit 3. The cache connectivity, the L2/L3 thing (if I'm right about the patent application) should provide for that kind of distribution - but obviously that costs extra processing within each local TU.

Jawed
Then how do you explain RV630 with 3 shader units and 2 TUs? Or RV610 with 2 shader units and 1 TU? Seems to me like the TUs can serve any of the SUs.

[Bob]

I'm claiming IP on this idea, but clearly they should sell "designer classic hot-rod sound" high-end models to mask the cooler.
Someone already beat you to it (http://www.youtube.com/watch?v=lWuSKXo98kE).

[Geo]

Someone already beat you to it (http://www.youtube.com/watch?v=lWuSKXo98kE).

Curses, foiled again! If only they'd made it sound like a Harley. . .

[nAo]

I expect R600 is like R5xx, in that a TU is only usable within the "quarter" of R600 that it finds itself in. e.g. the batches that are localised to shader unit 1 can execute on ALU pipes 1-16 and TU 1 (8TAs, 4TFs, 20 samplers). Shader unit 2 has ALU pipes 17-32 and TU 2 exclusively for its own batches, etc.
Maybe you're right, who knows.. but they must use that very wide ring bus for something, right? :)

[Moloch]

Maybe you're right, who knows.. but they must use that very wide ring bus for something, right? :)
Other than 512bit huge pee-pee bragging rights?

Guess we'll see week.

[ninjaspa]

i dont mean to rant or anything....well actually i do

3DMARK SCORES ARE MEANINGLESS STOP PRAISING THEM.


One of the main reasons i liked these forums in the past was the lack of beating 3Dmark scores to death everytime a random number was leaked in conjunction to a new peice of hardware. Is this a new trend? Maybe its just me but it looks ignorant!

just because people are interested does not mean they consider these revelations the be-all-end-all of gpu testaments. no matter how you slice it, it IS relevant. if you don't think so, then feel free to ignore them and keep to yourself. in an info-obsessed community, calling other people ignorant for considering these synthetic results as any indication of some new tech's capabilities when there is a nda in effect and very little official info to go on is pretty judgmental.

i mean the title of this thread is 'Rumours & Speculation', so cut people some slack. if you want concrete thorough analysis based on a gauntlet of the latest pc game fraps, then what are you doing in this thread anyway?

besides, we already get it: real world > synthetic. obviously. but pissing on 3dmark as 'meaningless' is pretty 'ignorant' in itself, if that is the term you wish to use. i'm no hardware professor, but when i looked at the latest vga charts on tomshardware last, it took only a cursory glance to discover that the 'overall games fps' average of 7 unique 3d intensive games scaled almost identically with the '3dmark 06 [v 1.0.2.]' results. reckon this is only coincidental?

[Geo]

Personally, it's mostly in top-end parts that I think default 3DM scores are worse than useless. I can see where they'd have some thumbnail usefullness in midrange and lowend parts. I mean, clearly they *do*, as the OEMs and IHVs use them that way. But default for high-end? No thanks.

[Jawed]

Umm, maybe I'm confused, but I thought the R300 could only co-issue a vec3+scalar.
I've looked at the CTM documentation, sections 3.2 and 3.3 indicate 3 instruction co-issue (ignoring flow control and texturing):

RGB
A
RGBA "presubtract"Three operands are allowed. The presubtract operates on operand 0 or operands 0 and 1:

SRCP_OP_BIAS - 1-2*src0
SRCP_OP_SUB - src1-src0
SRCP_OP_ADD - src1+src0
SRCP_OP_INV - 1-src0and the result can be used as one of the three operands for either or both of the "main" instructions.

I hope your not trying to claim that any vector ops == co-issue, that's rather twisted semantics that would make any DX8 card, or even DX7 register combiners "co-issue" GPUs.
No, I was claiming that a vector instruction devolves into a trivial co-issue in R600.

Jawed

[Sobek]

Out of curiosity, did anyone manage to get all of the images from that thread at OCworkbench?

http://my.ocworkbench.com/bbs/showthread.php?p=412421#post412421

i dont mean to rant or anything....well actually i do

3DMARK SCORES ARE MEANINGLESS STOP PRAISING THEM.

One of the main reasons i liked these forums in the past was the lack of beating 3Dmark scores to death everytime a random number was leaked in conjunction to a new peice of hardware. Is this a new trend? Maybe its just me but it looks ignorant!

It's a rumors and speculation thread...going ballistic over 3dmark06 scores supposedly pertaining to R600 is what we have to do here! To do otherwise would be blasphemy! BLAS-PHE-MYYYYYY! :twisted:

[DemoCoder]

I've looked at the CTM documentation, sections 3.2 and 3.3 indicate 3 instruction co-issue (ignoring flow control and texturing):
...


My reading of it seems to imply that presubtract cannot freely choose the two operands, so it's not like you can arbitrarily co-issue a X+Y or X-Y for any two X and Y. Instead, it just seems to be a DX8 source modifier unit with 2 extra ops.

Should I call _bias, _sat, _x2, _d2, etc in DX8 "co-issue"?

[Jawed]

Then how do you explain RV630 with 3 shader units and 2 TUs? Or RV610 with 2 shader units and 1 TU? Seems to me like the TUs can serve any of the SUs.
If you read horizontally, everything comes out just fine. These diagrams aren't consistent in the way they orientate the ALUs (if you go back to R5xx versions of these diagrams you'll see what I mean):

http://www.digit-life.com/articles2/video/r520-part1.html

e.g. X1600 should show the ALUs organised vertically if it follows the X1800 convention - which we know is 1:1 ALU:TEX. Or maybe X1800 is the odd one out, lol - except I've got an X1900 diagram here that agrees with the X1800 orientation.

:lol:

Jawed

[Jawed]

Maybe you're right, who knows.. but they must use that very wide ring bus for something, right? :)
I wonder if there'll be "cache coherency" traffic too.

One of the things that's got me puzzled is the separate cache that's shown at the left, labelled "memory read/write cache". If you interpret that as streamout for written data, then that could imply it's the source of vertex data when its read back in again. But that cache isn't connected to the samplers...

Also, you'd expect a cache to be too small to hold a streamed-out buffer - unless R600 is consuming the vertex streams at the same time as they're being written (dunno if D3D10 allows that). So if the cache is too small, when's it used as a read cache? Is it really distinct from the L2 cache for texture/vertex data?

It could be functioning as post-transform cache, I suppose.

Jawed

[Geeforcer]

Might R600 finally (for an ATI part) having PCF support (as required in the DX10 spec) play in there somehow?

If the rumored pricing is correct, then it is likely very indicative. Even if it's an excellent price/performance part, history is pretty conclusive that it's unlikely to be an absolute performance king at that price. AMD has a fiduciary responsibility to their shareholders even to not give away the store. What was the last performance king that launched at the price point the rumor sites are publishing? R360? Edit: Nope, 9800 XT was $499. . . .

Well, I ran down all the performance kings (as in, the fastest card when launched) from the last few years, in cronologically discending order

8800 GTX - $650
7950 G2 - $599
X1950 XTX -$450
X1900 XTX - $650
7800 GTX 512- $650 (Heh.... I don't think any of the 43 people who got it paid less than $700)
X1800 XT - $550
7800 GTX - $600
X850 XT PE - $550
6800 Ultra - $500
9800 XT - $500
9700 Pro - $400

So, to answer Geo's question.. ;) ... August 2002.

[Silent_Buddha]

Well, I ran down all the performance kings (as in, the fastest card when launched) from the last few years, in cronologically discending order

8800 GTX - $650
7950 G2 - $599
X1950 XTX -$450
X1900 XTX - $650
7800 GTX 512- $650 (Heh.... I don't think any of the 43 people who got it paid less than $700)
X1800 XT - $550
7800 GTX - $600
X850 XT PE - $550
6800 Ultra - $500
9800 XT - $500
9700 Pro - $400

So, to answer Geo's question.. ;) ... August 2002.

It'll be interesting that after the middle of the month presumably you can add 8800 GTX Ultra 849(?) and HD 2900 XT 399 to that list. :P

From just a casual glance one would almost think that the companies are going in opposite directions with one going for higher prices and one going for lower prices. :D

It has absolutely no bearing on actual price trends as I'd imagine an XTX type card will go for 499 or 599. But it's still amusing to see either way.

Regards,
SB

[Julidz]

http://http://img523.imageshack.us/img523/6871/tdpvfr600td6.jpg

what's the difference between GPU TDP and Board TDP ?


so the R600's(GDDR3) power consumption is 160 or 225 ??


and 80's Board TDP is 225W too , no ?

[Jawed]

My reading of it seems to imply that presubtract cannot freely choose the two operands, so it's not like you can arbitrarily co-issue a X+Y or X-Y for any two X and Y.
No - it definitely makes compilation more interesting within the constraint of the 3 source operands available :lol: The documentation also suggests there's a pipeline hazard there.

The independence of the alpha channel in the main ALU allows things like this in one instruction:

MAD r0,rgb, r1, r2, r3
ADD r0.a, r1, r2
MUL r0.a, r0.a, r0.a

Instead, it just seems to be a DX8 source modifier unit with 2 extra ops.
That ADD being quite useful, according to Mike Houston when talking about folding@home.

Should I call _bias, _sat, _x2, _d2, etc in DX8 "co-issue"?
As far as the ALU pipeline is concerned that's an implementation detail. We're talking about low-level co-issue here, aren't we, at the pipeline level? We're talking about a compiler using the units fully? How did NVidia implement the mini-ALUs for these modifiers? Was there an additional payload in NV4x or G7x?...

I wonder if those DX8 modifiers consume additional instruction slots in a D3D10 GPU. Sort of similar to the way that the fog unit has disappeared (thus requiring shader code)?

I don't see why that ADD raises your hackles. It's there for the compiler to use.

Makes R600 co-issue seem less fraught, actually :lol: Isn't that where we came in?

Jawed

[Fox5]

Well, I ran down all the performance kings (as in, the fastest card when launched) from the last few years, in cronologically discending order

8800 GTX - $650
7950 G2 - $599
X1950 XTX -$450
X1900 XTX - $650
7800 GTX 512- $650 (Heh.... I don't think any of the 43 people who got it paid less than $700)
X1800 XT - $550
7800 GTX - $600
X850 XT PE - $550
6800 Ultra - $500
9800 XT - $500
9700 Pro - $400

So, to answer Geo's question.. ;) ... August 2002.

Enter crackpot theory mode: (which can otherwise be completely disregarded because the 9700 pro was an all time high price point when it came out)

ATI was rather down and out (in terms of sales and debatably performance/features) when their performance champ came out, and had no choice but to price it low in order to get sales! Similar to the current situation with the X2900xtx!

[Julidz]

http://img523.imageshack.us/img523/6871/tdpvfr600td6.jpg

what's the difference between GPU TDP and Board TDP ?


so the R600's(GDDR3) power consumption is 160 or 225 ??


and G80's Board TDP is 225W too , no ?

^^;

[memberSince97]

That 16K+ 3dm06 is impressive , but that cpu getting over 2fps on both scores also seems super-duper clocked ? No ?

[SugarCoat]

Well, I ran down all the performance kings (as in, the fastest card when launched) from the last few years, in cronologically discending order

8800 GTX - $650
7950 G2 - $599
X1950 XTX -$450
X1900 XTX - $650
7800 GTX 512- $650 (Heh.... I don't think any of the 43 people who got it paid less than $700)
X1800 XT - $550
7800 GTX - $600
X850 XT PE - $550
6800 Ultra - $500
9800 XT - $500
9700 Pro - $400

So, to answer Geo's question.. ;) ... August 2002.


you forgot the X800XT PE @ $499 (i bought one at launch for $525 which seemed to be the new price. The XT just sorta popped up and took over the 475-499 price area.

[Jawed]

Jawed,

with the intel as of now and in a more english without to much tech lingo since I have no references to those,
what does the difference between Nvidias and ati´s cards really do in performance.
We can talk, dx9 and xp or dx10 and vista.
Open gl and d3d.

I understand that ati did something that seems more future oriented, longer shaders effiency and more dx10.

I am just a user of the cards, overklock a little and also read forums as these just for fun even if vector and tmu and such words brings little reference and meaning for me.

If you would put what is known and guess as of now, and translate and calculate what that means for games that are of now and are coming, I be interested to hear how you think the cards will do.
It looks likely that some DX9 games will perform markedly worse on R600, seemingly due to a lack of texturing rate. (Bear in mind I think 7900GTX performs markedly worse than X1950XTX, if you want a reference point.)

I suppose there'll be some exceptions. From all the benchmarking mess we've seen so far, it seems that the ROPs are pretty much equivalent in capability to G80's (e.g. at 4xMSAA with HDR) so it's the texturing that'll show the biggest difference.

I think D3D10 is a different ballgame. I think there's a decent likelihood that this will favour R600, prolly quite heavily - because I think ATI spent a lot of transistors on stuff that's new for D3D10 whereas NVidia seems to have spent a lot of transistors on texturing and ROPs.

If R600 is a D3D10 beast, how many games is that going to be relevant in? How long before you replace the card you buy this summer?

The best thing is to wait until the games you want to play are benchtested on the cards you're considering - hey there's less than a week to go now.

The expected bandwidth for R600 once looked like a probable indicator of performance, but now if it's any indication of performance it seems like it could only be for D3D10 games or better than 4xMSAA.

Maybe Call of Juarez's D3D10 patch (coming in the next week or so?) will make the crowds go "Woo!" but I'm pretty dubious we'll see any compelling D3D10 game content this year.

I personally found a big disconnect between the pre-launch G80 benchmarks and those that appeared in launch reviews, and those that have appeared since. I'm finding benchmarks less and less informative... Hell, I'm in a minority round here and think 8600GTS is a decent performer in DX9 games for its specification (not its price) so what do I know?

Jawed

[Anon Lamer]

Amazing.. the XT prices drop even before the card is out in retail yet..
€357 in holland.. (sapphire)

http://www.icomputers.nl//articledetail.aspx?A_ID=10878


Thats because the dollar is dropping against the Euro. Do not panic.

[Fox5]

It looks likely that some DX9 games will perform markedly worse on R600, seemingly due to a lack of texturing rate. (Bear in mind I think 7900GTX performs markedly worse than X1950XTX, if you want a reference point.)

I suppose there'll be some exceptions. From all the benchmarking mess we've seen so far, it seems that the ROPs are pretty much equivalent in capability to G80's (e.g. at 4xMSAA with HDR) so it's the texturing that'll show the biggest difference.

I think D3D10 is a different ballgame. I think there's a decent likelihood that this will favour R600, prolly quite heavily - because I think ATI spent a lot of transistors on stuff that's new for D3D10 whereas NVidia seems to have spent a lot of transistors on texturing and ROPs.

If R600 is a D3D10 beast, how many games is that going to be relevant in? How long before you replace the card you buy this summer?

The best thing is to wait until the games you want to play are benchtested on the cards you're considering - hey there's less than a week to go now.

The expected bandwidth for R600 once looked like a probable indicator of performance, but now if it's any indication of performance it seems like it could only be for D3D10 games or better than 4xMSAA.

Maybe Call of Juarez's D3D10 patch (coming in the next week or so?) will make the crowds go "Woo!" but I'm pretty dubious we'll see any compelling D3D10 game content this year.

I personally found a big disconnect between the pre-launch G80 benchmarks and those that appeared in launch reviews, and those that have appeared since. I'm finding benchmarks less and less informative... Hell, I'm in a minority round here and think 8600GTS is a decent performer in DX9 games for its specification (not its price) so what do I know?

Jawed


If the r600 is heavily geared for dx10, then perhaps ati tried to duplicate the success of the r300, which had exceptional dx9 performance. Even the cut down versions sold well, despite that the competing nvidia cards to things like the 9600pro were faster most of the time, they just couldn't compete in dx9 performance.
Of course, with the r300 ati also had a beastly top end card that won in everything, sometimes by a large margin, and much better AA performance. It's possible ATI could have much better AA performance (due to their larger memory bus) this time, but the desperately need a high end yet mass market priced card if they want that halo effect again. (perhaps the xt will be priced low enough and perform well enough to give them that)

[Rangers]

I think D3D10 is a different ballgame. I think there's a decent likelihood that this will favour R600, prolly quite heavily - because I think ATI spent a lot of transistors on stuff that's new for D3D10 whereas NVidia seems to have spent a lot of transistors on texturing and ROPs.

Does DX10 somehow reduce the need for texturing?

Although I guess, if the bottleneck shifts to other parts of the card in DX10, parts that ATI would do better in based on this presumption, it's somewhat irrelevant...

[Rangers]

Enter crackpot theory mode: (which can otherwise be completely disregarded because the 9700 pro was an all time high price point when it came out)

ATI was rather down and out (in terms of sales and debatably performance/features) when their performance champ came out, and had no choice but to price it low in order to get sales! Similar to the current situation with the X2900xtx!

9700 pro was probably just cheaper because of inflation. That was a while ago.

The rest of the cases typically Nvidia launched first, months before ATI, so they could charge a premium early. By the time ATI came in prices had already dropped.

[Colourless]

I think that you also need to look at the margins the cards and boards were bringing in to make a proper analysis. Nvidia is trying to ever increase it's margins, and at the same time the retail prices are going up. It's quite easy to get better margins by increasing your retail price (assuming of course that you are still getting the sales). This is an over simplification but it still needs to be considered. How much money is everyone making from the sales now compared to how much they were making 4 years ago as percentage of the board cost.

[Geo]

You mean like the fact that the $399 9700 Pro was driven by the petite 218mm^2 R300?

[INKster]

I think that you also need to look at the margins the cards and boards were bringing in to make a proper analysis. Nvidia is trying to ever increase it's margins, and at the same time the retail prices are going up. It's quite easy to get better margins by increasing your retail price (assuming of course that you are still getting the sales). This is an over simplification but it still needs to be considered. How much money is everyone making from the sales now compared to how much they were making 4 years ago as percentage of the board cost.

I think those assertions might be in need for re-consideration in light of actual DX10 competition coming in. Say, a few weeks from now...

[turtle]

Julidz, the difference is just what it says:

GPU wattage is wattage used by just the GPU (at default referance clocks)

Board wattage is wattage used by the GPU + rest of the board (RAM etc.) at default reference clocks.

The GPU on the XTX uses more power because presumably it is clocked higher.

Both use 225W because both are limited in clocks either/or because that is what is availble of the use of the 2x6-pin connector and/or difference in RAM used (2.0 is just the closest flat number.) Obviously 225W limits R600. IIRC overclocking is disabled when using 225W (2x6-pin) because it is THAT close to using 225W at default speeds.

All these overclocking scores are using 300W of available power, which in my mind explains the results, and the TDP is reality is going to be much higher when doing so.

G80's TDP I believe is 177W, but overclocking may be limited by available power (225W), which of course will make the power usage increase. I am not sure on that though (G80's case), but I surely believe R600 benefits from having that extra 75W available to it.

[Shtal]

R650 is the Radeon HD 2950XTX

ATI hasn't launched the R600 yet and it is already planning its next launch of a die shrinked version. It won't be launching the 1024 MB graphics card until Q3. It plans to launch the R650 chip and brand it as Radeon HD 2950XTX, 1024 MB GDDR4.

The chip is called R650 and it is a 65 nanometre core, how convenient. With this new chip ATI might have a fighting chance against the Geforce 8800 Ultra, but the real question is what will Nvidia have at that time.

The cards are scheduled for late Q3, so you should expect them to arrive in August or September, unless they get postponed. http://www.fudzilla.com/index.php?option=com_content&task=view&id=871&Itemid=1

Looks to me that they might scrap R600-XTX "HD 2900XTX 1024MB GDDR4" altogether.

So we will have Q2, 2007. R600 HD2900XT 512MB GDDR3
So we will have Q3, 2007. R650 HD2950XTX 1024MB GDDR4

I guess for R650 mostly be high clocks @ 1GHz GPU maybe ? :)

I wonder what Nvidia do? "I don't think Ultra will help, plus it is overpriced compare to ATI-R650"

[PSU-failure]

Does DX10 somehow reduce the need for texturing?

Although I guess, if the bottleneck shifts to other parts of the card in DX10, parts that ATI would do better in based on this presumption, it's somewhat irrelevant...
Think about what makes 3DMark06 such a "bad" GPU benchmark, except the CPU score weighing really too much.

What is texturing used for in most pre-DX10 games? Don't they use texture based bump mapping and lighting a lot?

In the light of 3DMark06 scores, it seems HDR is well handled by the R600 (almost the same SM3.0 score as the G80) and non-hdr tests suffer from a texturing weakness.

So, the question is: will upcoming games be full of dynamic lighting/geometry displacement or use mapped lights/shadows/bumps?

[Rangers]

Think about what makes 3DMark06 such a "bad" GPU benchmark, except the CPU score weighing really too much.

What is texturing used for in most pre-DX10 games? Don't they use texture based bump mapping and lighting a lot?

In the light of 3DMark06 scores, it seems HDR is well handled by the R600 (almost the same SM3.0 score as the G80) and non-hdr tests suffer from a texturing weakness.

So, the question is: will upcoming games be full of dynamic lighting/geometry displacement or use mapped lights/shadows/bumps?

Not sure what you're getting at. I dont know, I assume the former.

I think that could be where R600 makes ground. If it's poor at texturing but good at HDR texturing..well I think HDR is the future right?

[nAo]

in your tipical next gen game the vast majority of textures are LDR.

[seahawk]

I wonder what Nvidia do? "I don't think Ultra will help, plus it is overpriced compare to ATI-R650"

Do you beleive G80 will be the competition for R650 ? They shipped in Nov 2006.

[AlexV]

Think about what makes 3DMark06 such a "bad" GPU benchmark, except the CPU score weighing really too much.

What is texturing used for in most pre-DX10 games? Don't they use texture based bump mapping and lighting a lot?

In the light of 3DMark06 scores, it seems HDR is well handled by the R600 (almost the same SM3.0 score as the G80) and non-hdr tests suffer from a texturing weakness.

So, the question is: will upcoming games be full of dynamic lighting/geometry displacement or use mapped lights/shadows/bumps?


Come again on the bolded part?Unless you`re stuck somewhere in DX8 era, no, not quite, AFAIR. The thing is this:if you suck at math in DX9(it`s an example, I`m not saying that`s the case with R600 because, frankly, I dunno and I hate stating some crap personal assertion as fact), you`re still going to suck at it in DX10 mostly. There`s no huge paradigm shift, you`re not doing stuff essentially different. You have the ability to do them far more elegantly, that`s certain. The trump card with DX10, IMHO, is the GS...that one is new, and we don`t know how it`ll perform.

The noise about the GS was that first-gen DX10 hardware wouldn`t be fast enough to make it a viable large-scale general use option for games. That makes me think it`s a point where an advantage could be gained:you have two ass-slow parts, but one is only half-ass slow...the difference from 0 to 20 fps is far more perceivable than the difference from 60-80(this is oversimplified, but I think it serves to get the point across). If ATi is actually better than nV at this(as some MS noise suggested), and if they can actually get devs to make use of the GS in such a manner that creates the opportunity of illustrating this performance delta(which I kindof doubt, because they`ve sucked the big one on this front ever-since first introduction of DX9, 3Dc, Fetch4 and so on, whilst nV managed to push everything they had in one way or another), it may actually matter. But notice there`s a lot of ifs in there.

And again, consider AMDs and implicitly ATi's position as a company:are they really in such a situation that they can afford to release world-breaking performance at mid-range prices?Are they a charity?Ya think?There must be something else at play here.

[vertex_shader]

http://www.imagebanana.com/img/b11ska11/xh29.jpg

I'm bored with 3dmark scores, its say nothing about real ingame performance.

[Silent_Buddha]

And there's always the case that Dev's won't do much with Geometry Shaders even if R600 (and this is a gross exaggeration I'm sure) managed to outperform G80 by an order of magnitude (WRT GS).

After all IF the current high end market leader cannot run GS worth a damn, yet their competition can run it well, it's still not going to gain major traction. I'd imagine this is a major reason NV doesn't seem very concerned with GS, even if ATI goes gangbusters on it. If both companies can't do it at least reasonably well, why put a lot of resources into developing a game that will make major use of it?

Personally, the main selling point of a performant DX10 part at this point in time is for people like me. Someone who only upgrades a video card once every 1-2 years. (This year might be an anomaly for me as I might upgrade 2 systems graphics cards this year.)

So, considering that I'm getting either G80 or R600 with the expectation of using it at the VERY least until next summer (more likely next winter) then I, for one, am VERY interested to see anything that might give an indication of how each might possibly perform with regards to DX10 and its use of Geometry Shader.

Regards,
SB

[Unknown Soldier]

When Nvidia has been this quiet, I think you can always expect them to drop something beeg.

So the G80 has ruled the roost for 7 months and the Ultra might just own the HD2900XT for a short while until the XTX is released then ATI is expected to own for +- 5-6 months, but then I expect Nvidia to release it's next GPU which should put the hurting back(if what we've seen with both the NV40-G80).

Questions is, what does ATI plan to do for the following 6 months after that?

US

[Unknown Soldier]

I'm bored with 3dmark scores, its say nothing about real ingame performance.

And it says nothing about DX10 scores either. ;)

US

[Arnold Beckenbauer]

If you read horizontally, everything comes out just fine. These diagrams aren't consistent in the way they orientate the ALUs (if you go back to R5xx versions of these diagrams you'll see what I mean):

http://www.digit-life.com/articles2/video/r520-part1.html

e.g. X1600 should show the ALUs organised vertically if it follows the X1800 convention - which we know is 1:1 ALU:TEX. Or maybe X1800 is the odd one out, lol - except I've got an X1900 diagram here that agrees with the X1800 orientation.

:lol:

Jawed
http://directupload.com/files/j2rwj2tyjxmcamngnimz.jpeg
Reading horizontally I see two shader clusters with 12 5D ALUs and one TU per shader cluster (RV630). The RV610 has then one shader cluster with one TU and 8 5D ALUs.
My problem is: Why ATi says, RV630 has 3 SIMDs and RV610 2 SIMDs?
http://www.beyond3d.com/content/articles/4/7

[Pressure]

When Nvidia has been this quiet, I think you can always expect them to drop something beeg.

So the G80 has ruled the roost for 7 months and the Ultra might just own the HD2900XT for a short while until the XTX is released then ATI is expected to own for +- 5-6 months, but then I expect Nvidia to release it's next GPU which should put the hurting back(if what we've seen with both the NV40-G80).

Questions is, what does ATI plan to do for the following 6 months after that?

US

What if AMD does not care for the Ultra High-end market anymore? I mean, NVIDIA releasing a $800+ graphic card just to rain on AMDs parade seems like a major case of the silly-willies.

If the Radeon HD 2900XT releases at $399 and performs nicely I see it as a good trend.

Admittedly it is late, and very late indeed, if it were to compete against the G80 in the first place. I still think it has its place though and hopefully the mid-range and low-end parts is very competitive against anything NVIDIA currently has on the market.

[aeryon]

i dont mean to rant or anything....well actually i do

3DMARK SCORES ARE MEANINGLESS STOP PRAISING THEM.


One of the main reasons i liked these forums in the past was the lack of beating 3Dmark scores to death everytime a random number was leaked in conjunction to a new peice of hardware. Is this a new trend? Maybe its just me but it looks ignorant!

it's so true ! and talking about gaming performance, some food to mao5 :

KTE wrote:
I read Sampsa saying the HD2900XT 512MB GDDR3 is only a competitor to a 8800GTS, not able to compete with the 8800GTX or Ultra.

well know Kinc reply:
Yes in DX9 games in gamin resolutios with AA that is. There is a new driver coming today lets see wha that does.

Synthetic benchmarks doesnt awlays apply to real world gaming :sad:

source: nordichardware forum (http://www.nordichardware.com/forum/next-vt8428.html?postdays=0&postorder=asc&start=25)

[leoneazzurro]

http://directupload.com/files/j2rwj2tyjxmcamngnimz.jpeg
Reading horizontally I see two shader clusters with 12 5D ALUs and one TU per shader cluster (RV630). The RV610 has then one shader cluster with one TU and 8 5D ALUs.
My problem is: Why ATi says, RV630 has 3 SIMDs and RV610 2 SIMDs?
http://www.beyond3d.com/content/articles/4/7

They refer to the main block organization of both GPU, if you take a look at the Ultra threaded processor outputs, you see it is connected to 3 (in RV 630) or 2 (in R610) cluster of units, they are the SIMD units they refer.
Now, the disparity of TU in regard to SIMD can mean two things IMHO:

1) Each cluster can access every TU indipendently because they are fully decoupled or
2) organization of R6XX is far more complex, with the TU coupled "orthogonally" to SIMD clusters, i.e. every SIMD cluster is further divided into blocks (i.e. 4 for R600, 2 for RV630 and only one for R610) and each block is coupled with a TU (8 address, 4 filtering, 20 fetching units each).

I already wrote a lot of posts ago that a very reliable source told me that R600 is structured as 4-4-4-5 whereas G80 is 8-2-8. In this case it can be that in R600 every 80-SP cluster is divided in 4 blocks with (4x5)SP each each working on a quad, and every one of these blocks has a dedicated TU. In this way, block 1s in SIMD 1,2,3,4 access TU 1, block 2s in SIMD 1,2,3,4 access TU 2, and so on. If you take a look at the block diagram there could be some hints about this possibility.
The benefits of doing so I cannot understand, however. :oops:

Same souce told that R600 performs better with long shaders, but on simple shaders this organization, by having one "level" more with respect to G80, incurs in some penalty.

[Unknown Soldier]

What if AMD does not care for the Ultra High-end market anymore? I mean, NVIDIA releasing a $800+ graphic card just to rain on AMDs parade seems like a major case of the silly-willies.

If the Radeon HD 2900XT releases at $399 and performs nicely I see it as a good trend.

Admittedly it is late, and very late indeed, if it were to compete against the G80 in the first place. I still think it has its place though and hopefully the mid-range and low-end parts is very competitive against anything NVIDIA currently has on the market.

At the end of the day, people like us do look at the high-end, so don't kid yourself that if the XT only beats the GTS and can't beat the GTX that people won't go for the GTX instead. Of course some of us also prefer better IQ, so it'll have to have better IQ than the GTX, which we all know is pretty good atm.

I think though that the most important thing will be the DX10 performance. Of course we'll still have to wait for any worth while software to come out and by then(most probably) newer and faster hardware will be available.

US

[neliz]

AMD's power of three?

http://ati.amd.com/online/getall3/

[Arty]

From Tweaktown (http://www.tweaktown.com/)


http://i19.tinypic.com/4pbp4hw.jpg (http://i19.tinypic.com/6bbzas6.jpg)


The voucher will cost AIB partners roughly $6 - 7 USD each and will be bundled with all XT cards but it is unsure at this stage if the voucher will be bundled with lower end graphics cards.We have got some information form a reliable source that ATI will be launching the Radeon HD 2900XTX in the end anyhow but nobody is entirely clear on the exact details.

http://i12.tinypic.com/5yhbww9.jpg


It will most likely not be out until late July or early August, but it should be on display at Computex by at least a couple of ATI’s partners from what our Taiwan spies are telling us. This is the consumer version of the XTX and will look pretty much identical to the HD 2900XT except with a different sticker and more and faster GDDR-4 RAM under the cooler, i.e. this is not the 12.4in full length cards that will be available from OEM’s although both PCB length versions are floating around. The model name hasn’t been set as yet and it might be called XT GDDR4 or XTX, but we’ll stick with XTX for now. ATI partners have the option of ordering either the GDDR-3 XT or GDDR-4 version right now.

It will feature 1GB of GDDR memory and this means a higher memory clock, up from 1650MHz DDR for the HD 2900XT to 2020MHz DDR. The core will also be boosted, but only a laughable 10MHz from 740 to 750MHz. It seems like there will be as much point in buying the Radeon HD 2900XTX as it is to buy a GeForce 8800 Ultra. Maybe the era of the super high-end cards is finally over and we will all get decent performance from an affordable graphics card, but then again, miracles doesn't happen that easily.

ATI’s 1GB GDDR-4 card will likely be priced around the same as Nvidia’s GeForce 8800GTX 768MB at around $550 USD and you will probably be able to buy it in late July or sometime in August.

[Pressure]

At the end of the day, people like us do look at the high-end, so don't kid yourself that if the XT only beats the GTS and can't beat the GTX that people won't go for the GTX instead. Of course some of us also prefer better IQ, so it'll have to have better IQ than the GTX, which we all know is pretty good atm.

I think though that the most important thing will be the DX10 performance. Of course we'll still have to wait for any worth while software to come out and by then(most probably) newer and faster hardware will be available.

US

Not to mention actual reviews of the hardware we are discussing.

I do not really care whether or not the Radeon HD2900XT beats the Geforce 8800GTX, what I care about if it enables me to do more things in real-time when editing video and applying effects. In the end of the day I hardly play games anymore and there is hardly any games released for Mac OS X regardless. Hopefully Apple will use the Radeon HD 2900XTX 1GB OEM version in its Mac Pro line-up.

The Radeon 2k family should bring more to the table than just 3D-performance afterall ;)

But sure, performance figures is always fun to look at.

[bigtabs]

That's an odd statement to make in a forum such as this.

It seems like there will be as much point in buying the Radeon HD 2900XTX as it is to buy a GeForce 8800 Ultra.

The 1GB VRAM would be more than a small point for me. With that much memory I could use every texture mod I wanted in Oblivion and crank AA up to great levels without worry. Can't do that with 512mb.

I'd also expect UT3 to want more than half a gig of VRAM at max settings. So I think their comparison is a little wrong there.

[PSU-failure]

And again, consider AMDs and implicitly ATi's position as a company:are they really in such a situation that they can afford to release world-breaking performance at mid-range prices?Are they a charity?Ya think?There must be something else at play here.
The answer is quite easy to find: 6 months delay since nVidia competing part launch.

Did you ask the same question when Intel released Conroe at a lower price point than AMD's X2 competing parts?

[Dalton Sleeper]

I was hoping to get some X-Fire results in real games, i don't think one XT will be enough with only 512MB. But it's only going to be X-Fire if it performs good, alot better than one card. Otherwise it's going to be XTX...

[AlexV]

The answer is quite easy to find: 6 months delay since nVidia competing part launch.

Did you ask the same question when Intel released Conroe at a lower price point than AMD's X2 competing parts?

No. Why did the R520 launch at such a comparatively lower price?Don`t answer that if you`re in love with ATi and think that they`re in it for the good of the planet. They want to make a profit. If they had similar performance to the GTX across the board, they would`ve priced it similarly, and maybe even kick nV in the nuts about it. I don`t think the R600 is cheap enough to make to warrant such a lowish price. Even if you factor comparatively lower RAM size.

Intel released Conroe at the price-points that were characteristic for their market-a super high-end part is at 1000, for example, and from there down-wards. They could afford to sell it at that price and maintain considerable margins. AFAIR, Conroe had goodish yields. They also have much more flexibility:from a wafer they can get a lot of parts(let's say E6600/6700/X6800, depending on how they handle clock-rates). I don`t think you can make X2600s from failed X2900-I may be wrong here though.

If some of you ppl can`t see why it`s fairly illogical, from an economic standpoint, to release your presumably top-end card at this price-point, if it is so much better/at least equal to your competitor, that`s that, I doubt anything I or anyone say will change the apparently universally accepted idea that ATi=God`s arm on earth.

[3dilettante]

Conroe is also a process node ahead of the top X2s.

It's roughly 140mm2 compared to a 90 nm X2 that is just under 200 mm2.
Coupled with Intel's much larger fab capacity, a lower price point could be hit while simultaneously butchering AMD's margins.

I do not see a die size advantage for R600 over G80, unless the NVIO significantly impacts manufacturing costs.
Since both Nvidia and ATI are fabless and use the same foundry, there is no indication of any large differences in economies of scale between the two.

Perhaps the lower price is due to AMD's newfound allergy to inventory. It sounds like they don't want too many R600s to experience old age in their warehouses.

[Jawed]

http://directupload.com/files/j2rwj2tyjxmcamngnimz.jpeg
[...]
My problem is: Why ATi says, RV630 has 3 SIMDs and RV610 2 SIMDs?
Aha! Hmm, I hadn't spotted that, that definitely contradicts what I've been asserting (I've been referring to the first versions of those diagrams that leaked, without the green box - sigh - so I've missed this point).

If, as suggested (and you rightly link Xenos), these lower GPUs have finer-granularity, then that definitely puts a different complexion on things. They have more sequencer complexity than I was expecting, which means they have better dynamic branching than I was expecting, too.

Ouch :!:

That might point to the future of R6xx, where sequencer complexity increases, to bring batch sizes down. I'm expecting R600 to have batches of 64. It would appear that RV630 has a batch size of 32 and RV610 is 16.

3 SIMDs and 2 TUs in RV630 is really very curious just because it's a hybrid of R5xx and Xenos instruction (ALU + TEX) scheduling.

But, well, this is a definite change of course for me. Ooh that's quite a nice surprise :shock:

Jawed

[Razor1]

remember that rumor monger site..... how real can these be?

http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1&limit=1&limitstart=1

[trinibwoy]

Well if they're not real I give them 10 points for imagination. If real those points go to ATI for most impressive performance in the WTF category this year. I mean, there is absolutely no friggin way R600 should be losing to the GTS in ANYTHING.

[Razor1]

if they are fake, man lots of sites are going to be loosing their reputation:lol:

[Unknown Soldier]

I don`t think you can make X2600s from failed X2900-I may be wrong here though.

No, but you can make XL's. ;)

[Jawed]

I already wrote a lot of posts ago that a very reliable source told me that R600 is structured as 4-4-4-5 whereas G80 is 8-2-8. In this case it can be that in R600 every 80-SP cluster is divided in 4 blocks with (4x5)SP each each working on a quad, and every one of these blocks has a dedicated TU. In this way, block 1s in SIMD 1,2,3,4 access TU 1, block 2s in SIMD 1,2,3,4 access TU 2, and so on. If you take a look at the block diagram there could be some hints about this possibility.
Good thinking. This is very intriguing.

8-2-8 appears to be read as:

clusters
number of ALU arrays per cluster
number of pixels per arraySo 4-4-4-5 would need to be interpreted as:

clusters
number of ALU arrays per cluster
number of pixels per array
number of co-issues per pixelBy this nomenclature, RV630 would be 3-2-4-5 and RV610 would be 2-1-4-5.


The benefits of doing so I cannot understand, however. :oops:
Smaller batches, better dynamic branching.

Same souce told that R600 performs better with long shaders, but on simple shaders this organization, by having one "level" more with respect to G80, incurs in some penalty.
The "level" might be referring to the fact that R600 is 5 scalars co-issued, rather than G80 which (sort of) is just scalar. Dunno.

These ideas could indicate far more scheduler complexity in R600 than in R580. We need a leaked slide that says R600 is 16 SIMDs to confirm this, I think. Otherwise it's subtler and not as exciting :lol:

Jawed

[Arty]

Well if they're not real I give them 10 points for imagination. If real those points go to ATI for most impressive performance in the WTF category this year. I mean, there is absolutely no friggin way R600 should be losing to the GTS in ANYTHING.
Its a give away when the author/reviewer himself asks you to resort to salt .. ;)

"take these results with a tiny bit of salt"

[Sound_Card]

if they are fake, man lots of sites are going to be loosing their reputation:lol:

If their real, then alot of well known respected editors are going to be losing their credibility.

Why must you put faith in a review that looks half ass to say the least?

[Razor1]

If their real, then alot of well known respected editors are going to be losing their credibility.

Why must you put faith in a review that looks half ass to say the least?


hmm i questioned the results :wink:

[vertex_shader]

remember that rumor monger site..... how real can these be?

http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1&limit=1&limitstart=1

Without pictures, without driver info, without any real information about benchmarking method, i say like 1% :wink:

[trinibwoy]

Its a give away when the author/reviewer himself asks you to resort to salt .. ;)

That was in reference to a specific set of results.

It seems possible that AA/AF settings have some problems with CoH and Radeon HD2900XT, so take these results with a tiny bit of salt.

And Sound Card what do you want people to do? Ignore everything that you don't like? The rumour mill is churning and you can't conveniently filter out the stuff you don't want to see/hear. Just take it for five more days. You've stood strong this long, a little while longer won't kill you :) It's not like people are taking this stuff as gospel - it's just the usual hullabaloo that surrounds every launch.

[CarstenS]

We need a leaked slide that says R600 is 16 SIMDs to confirm this, I think.
I don't think you'll find any.

Reaon: Take the slide from p193 - RV630 is supposed to have 120 SPUs in 3 SIMDs. Now what would that give for 16 SIMDs? Definitely more than 320, right?

[INKster]

Well if they're not real I give them 10 points for imagination. If real those points go to ATI for most impressive performance in the WTF category this year. I mean, there is absolutely no friggin way R600 should be losing to the GTS in ANYTHING.

From the looks of it, the HD 2900 XT is loosing to a X1950 XTX (yeah, right... :D).

[Arty]

That was in reference to a specific set of results.
How convenient .. :lol:

[neliz]

How convenient .. :lol:

Anyway, their benchmark numbers are .. "whack" to state it in a very 80's way, but also conveniently compatible with DT's "OMG AMD IS d00meD!" review.

[IbaneZ]

Bah, fake! N00bs!! Everybody knows that you have to be a rocket scientist to benchmark a video card, and I'm sure that the guys at dailytech and it-review don't qualify. :wink:

Seriously though, the early previews/fakenews/whatever, and hints from people that the gaming performance ain't that good is a little spooky.

Less than five days left now, and you just know a few sites will post their stuff early. I'm thinking vr-zone. :smile:

[trinibwoy]

How convenient .. :lol:

Just wanted to make sure you weren't taking the statement out of context that's all :)

[vertex_shader]

Fruitzilla:

R600XT overclocks with two 2x3 pin

The Radeon HD 2900 XT will overclock to 840 MHz on the air and if you havent seen that story you can see it here. The 840 MHz core on the Radeon HD 2900 XT card is possible with the reference cooler.

Even if you don't have the 2x4 pin necessary for overclocking you will be able to overclock. The plain two 2x3 pin power connectors will do the job but the Catalyst control center won't recognise that the card has the Overdrive capability.


Link (http://www.fudzilla.com/index.php?option=com_content&task=view&id=906&Itemid=1)

[Arnold Beckenbauer]

I don't think you'll find any.

Reason: Take the slide from p193 - RV630 is supposed to have 120 SPUs in 3 SIMDs. Now what would that give for 16 SIMDs? Definitely more than 320, right?

But eight?

[Jawed]

I don't think you'll find any.

Reaon: Take the slide from p193 - RV630 is supposed to have 120 SPUs in 3 SIMDs. Now what would that give for 16 SIMDs? Definitely more than 320, right?
All three of these GPUs have varying ALU:TEX ratios, so I don't see how RV630 can lead to such a specific conclusion.

---

I think the smaller GPUs might need their smaller batch sizes for general batch throughput reasons.

Forgetting the advantage that smaller batches brings to dynamic branching, when the GPU is trying to timeslice amongst primitive, vertex and pixel batches, smaller batches make for more finely-grained scheduling.

I don't know how important this is...

Jawed

[cadaveca]

DX10.1 important, methinks.:wink:

[Pressure]

remember that rumor monger site..... how real can these be?

http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1&limit=1&limitstart=1

Just take a look at the components used:

- ASUS P5W DH Deluxe
- EVGA 122-CK-NF68-A1
- ASUS M2N32-SLI Premium
- DFI ICFX3200T2R/G
- AMD X2 6000+
- Intel Core 2 Duo E6700
- OCZ Platinum XTC PC2-8500 (2x2x1GB)
- AMD Radeon HD2900XT
- OCZ and EVGA 8800GTX
- ASUS 8800GTS 640MB
- OCZ GameXstream PSU (new revision, rewired for 8-pin PCI-E connector)
- Corsair HX620
- OCZ Vindicator coolers
- WD RaptorX's 150GB
- Pioneer DVR-110BK's

- Dell 3007WFP 30" LCD monitor (up to 2560x1600)
- Benq FP241W (up to 1920x1200)

I have my doubts that they even used the same platform.

[leoneazzurro]

A SIMD in this ATI nomenclature would be each main ALU cluster.
The basic "tile" of this architecture is the group of 20 SP, arranged in a 4x5 fashion.
Let's call it a "quad block". Each level of "Quad block" is tied to a TU.
So R600 is a 4 SIMD, with each SIMD being 4 Quad blocks, 4 TU GPU.
RV630 is a 3 SIMD, with each SIMD being 2 Quad Blocks, 2 TU GPU.
R610 is a 2 SIMD, wich each SIMD being a Quad Block, 1 TU GPU.

In any case, RV630 is much less texture limited in relation to R600 (in terms of ALU:TEX), because one TU serves 4 SIMD in R600 but only 3 in RV630.

EDIT: if this is really how R6XX GPUs are organized...

[Geeforcer]

IT-reviews could have benchmarked one of those 65nm R600s they have been insiting on. It had trouble keeping up with other cards, being imaginary and all.

[trinibwoy]

I have my doubts that they even used the same platform.

Yeah I'm not even sure what they're trying to say there....

[Nebula]

Its a give away when the author/reviewer himself asks you to resort to salt .. ;)

OK, I don't know the validity of those benchmarks but to be fair the reviewer said to take the 'CoH benchmarks with a tiny grain of salt' ! :smile:

It seems possible that AA/AF settings have some problems with CoH and Radeon HD2900XT, so take these results with a tiny bit of salt. We'll add sugar to "neutralize" a bit later.

[CarstenS]

All three of these GPUs have varying ALU:TEX ratios, so I don't see how RV630 can lead to such a specific conclusion.

I'm confused: Why are you talking about ALU-TEX-Ratios all of a sudden? I was under the impression, we were discussing the organization of the ALUs in SIMD-Arrays.

[allnighter]

Just take a look at the components used:



I have my doubts that they even used the same platform.

Hey the guy's promising more goodness tomorrow. We'll feast on some Crossfire vs. SLi 'results'. Than we'll see some 60's wacko stuff, all in pretty colors and shapes.

[Jawed]

A SIMD in this ATI nomenclature would be each main ALU cluster.
The basic "tile" of this architecture is the group of 20 SP, arranged in a 4x5 fashion.
Let's call it a "quad block". Each level of "Quad block" is tied to a TU.
So R600 is a 4 SIMD, with each SIMD being 4 Quad blocks, 4 TU GPU.
RV630 is a 3 SIMD, with each SIMD being 2 Quad Blocks, 2 TU GPU.
R610 is a 2 SIMD, wich each SIMD being a Quad Block, 1 TU GPU.
I reckon you're thinking about this much more clearly than me :smile:

I see what you are saying, that all of these GPUs have multiple batches in flight inside each SIMD cluster.

It's a very interesting hybrid. I'm now feeling more optimistic about batch size in R600. Your analysis indicates a likelihood of 16, if each quad block is independent of its brothers inside the SIMD cluster. This presumes that an instruction lasts 4 clocks.

In any case, RV630 is much less texture limited in relation to R600 (in terms of ALU:TEX), because one TU serves 4 SIMD in R600 but only 3 in RV630.

EDIT: if this is really how R6XX GPUs are organized...
Thanks, I think this new perspective has merit!

Jawed

[Jawed]

I'm confused: Why are you talking about ALU-TEX-Ratios all of a sudden? I was under the impression, we were discussing the organization of the ALUs in SIMD-Arrays.
I think you can forget what I said. leoneazzurro's argument is that 4 SIMDs are enough for R600 and so I'm barking up the wrong tree anyway! I didn't understand the 4-4-4-5 configuration properly.

Jawed

[{Sniping}Waste]

Kinc has now confirmed that theres no cold bug and has had the GPU around -90C and boots up fine.

[Robin B]

So its only in 3Dmark its doing fine,great guess its many thats palying it.:roll:

http://www.xtremesystems.org/forums/showpost.php?p=2180185&postcount=34

[Arty]

So its only in 3Dmark its doing fine,great guess its many thats palying it.:roll:

http://www.xtremesystems.org/forums/showpost.php?p=2180185&postcount=34
A hard oc GTS can even come close to a GTX .. Thats why I am not a big fan of cryptic hints especially so close to the reviews. :|

[_xxx_]

But eight?

You mean like 8x40, if roughly compared to G80's 8x16?

Edit: oops, just saw leoneazzurro's post...

[vertex_shader]

So its only in 3Dmark its doing fine,great guess its many thats palying it.:roll:

http://www.xtremesystems.org/forums/showpost.php?p=2180185&postcount=34

Orton days counted in amd :wink:

[Geo]

So its only in 3Dmark its doing fine,great guess its many thats palying it.:roll:

http://www.xtremesystems.org/forums/showpost.php?p=2180185&postcount=34

Okay, maybe yesterday's hint wasn't enough. No more sarcasm-laced rolly eyes in this thread, or maybe we'll ask the console guys to bring their moderation skills in here. This is Beyond3D; learn how to express your thinking with complete, thoughtful sentences (and preferably paragraphs). . . or do everyone a favor and just read the posts of those members who can.

[vertex_shader]

Slowly i start thinking r600 born for dx10, and this is why card is late, amd hope some dx10 patch can catch up the card, when everything goes fine we see dx10 game performance numbers monday :smile:

[IbaneZ]

So its only in 3Dmark its doing fine,great guess its many thats palying it.:roll:

http://www.xtremesystems.org/forums/showpost.php?p=2180185&postcount=34

Deep freeze rocks! :wink:

Ok, too many bad news in one day. Well, at least the 2900XT will be cheap (and rock in synthetic benchies apparently). Always something.

[Geeforcer]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance - almost as if people are saying "Well, those specks are got to be good for SOMETHING, right?". I am personally trying to think of the last card that was slower then similarly-featured card in DX(n) but significantly faster in DX(n+1).

Of course, if anyone can point me to the specific features of R600 architecture and specific provisions in DX10, that, when combined, would deliver significant advantage over G80, I am all ears.

[leoneazzurro]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance - almost as if people are saying "Well, those specks are got to be good for SOMETHING, right?". I am personally trying to think of the last card that was slower then similarly-featured card in DX(n) but significantly faster in DX(n+1).

Of course, if anyone can point me to the specific features of R600 architecture and specific provisions in DX10, that, when combined, would deliver significant advantage over G80, I am all ears.

But we have seen cards that in DX(n+1) were much slower even winning in DX(n) :grin:
Anyway, in this case it will happen IMHO only if G80 is really bad at some DX10-only features (GS) AND they will be used massively in games.

[trinibwoy]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance - almost as if people are saying "Well, those specks are got to be good for SOMETHING, right?".

Well to be fair those specs are pretty impressive and the reported performance is severely lacking in comparison. I think somebody pointed to 3dmark as maybe an example of what we can expect with tuned drivers. Maybe there's more of that in store for other apps. Though that doesn't bode well for the architecture as a whole if a lot of per-application tweaking is necessary to achieve its full potential.

[Geo]

I am personally trying to think of the last card that was slower then similarly-featured card in DX(n) but significantly faster in DX(n+1)

R300, of course, in DX8 with no AA/AF. But sometimes --even usually?-- a cigar is just a cigar.

[Geeforcer]

R300, of course, in DX8 with no AA/AF. But sometimes --even usually?-- a cigar is just a cigar.

Of course at any setting actually worth talking about, in DX8 OR DX9 it was faster than NV30 and NV35 too, for good measure.

[compres]

Of course at any setting actually worth talking about, in DX8 OR DX9 it was faster than NV30 and NV35 too, for good measure.

Faster sometimes in dx8 and much faster in dx9. So there is your example. I think r600 is going to be similar, great in dx9 and best in dx10, the difference is nVidia will have a really good chip too this time.

[Geo]

Of course at any setting actually worth talking about, in DX8 OR DX9 it was faster than NV30 and NV35 too, for good measure.

And yet if one were to look at actual reviews of the day and see how many sites actually tested R300 vs NV30* one might be amazed at what some people who were supposed to be experts have been known to consider as "actually worth talking about". . . :lol:



*with some honorable exceptions, of course.

[digitalwanderer]

If nothing else this will add some much needed melodrama to the launch. :)

[PeterAce]

Well I really looking forward to seeing a greater range of intersting synthetic benchmarks (other than 3DMark) here at Beyond3D *.

I was super-disapointed about the lack of results/graphs for synthetic benchmarks in the G80 Performance Analysis (esp after Bogobench, Rightmark3D, Marko Dolenc's fillrate tester and others were mentioned but no results were supplied!).

* Unless the 'R600 Performance Article' is seperate from the 'R600 Architecture Overview' article and then I'll have to be a bit more patient...

[Silent_Buddha]

- ASUS P5W DH Deluxe
- EVGA 122-CK-NF68-A1
- ASUS M2N32-SLI Premium
- DFI ICFX3200T2R/G
- AMD X2 6000+
- Intel Core 2 Duo E6700
- OCZ Platinum XTC PC2-8500 (2x2x1GB)
- AMD Radeon HD2900XT
- OCZ and EVGA 8800GTX
- ASUS 8800GTS 640MB
- OCZ GameXstream PSU (new revision, rewired for 8-pin PCI-E connector)
- Corsair HX620
- OCZ Vindicator coolers
- WD RaptorX's 150GB
- Pioneer DVR-110BK's

- Dell 3007WFP 30" LCD monitor (up to 2560x1600)
- Benq FP241W (up to 1920x1200)

That's my favorite part of the bench systems used in that article. No indication of what component was used with what video card or on what test.

Shoddy to say the least, even if the results are correct, the article is certainly questionable.

Regards,
SB

[Julidz]

Julidz, the difference is just what it says:

GPU wattage is wattage used by just the GPU (at default referance clocks)

Board wattage is wattage used by the GPU + rest of the board (RAM etc.) at default reference clocks.

The GPU on the XTX uses more power because presumably it is clocked higher.

Both use 225W because both are limited in clocks either/or because that is what is availble of the use of the 2x6-pin connector and/or difference in RAM used (2.0 is just the closest flat number.) Obviously 225W limits R600. IIRC overclocking is disabled when using 225W (2x6-pin) because it is THAT close to using 225W at default speeds.

All these overclocking scores are using 300W of available power, which in my mind explains the results, and the TDP is reality is going to be much higher when doing so.

G80's TDP I believe is 177W, but overclocking may be limited by available power (225W), which of course will make the power usage increase. I am not sure on that though (G80's case), but I surely believe R600 benefits from having that extra 75W available to it.


hmmm thx

[Julidz]

this is true ???



Here is the one of the key reasons why the R600 marchitecture falls behind the G80 GTX version. While Geforce 8800 GTX or G80GTX has 32 Texture Memory Units (TMUs) Radeon HD 2900 XT has 16 only.

TMUs are responsible for getting the textures in your games and the more you have, the faster you can handle the texture work.

Still, the Radeon HD 2900 XT at 750 MHz core is about 30 percent faster in raw clock speed from Geforce 8800 GTX at 575 MHz and finishes the texture cycle some 30 percent faster than Nvidia.

In spite of that this is not really an ideal solution as Nvidia has twice as many texture memory units. The only hope is that ATI has more effective TMUs, but this is one big marchitectural disadvantage over the G80 marchitecture.

[Silent_Buddha]

Of course at any setting actually worth talking about, in DX8 OR DX9 it was faster than NV30 and NV35 too, for good measure.

I also seem to remember at the time NV30 launched that it did in fact win some benchmarks vs R300 in DX8 games. However it got totally slaughtered in anything DX9 related. Were there even any DX9 games when NV30 launched? Don't think so. In which case in anything without AA and AF, the NV30 benched similarly to the R300 in DX8 games of the time. Totally different case with AA and AF however.

This was even more pronounced in NV35 where it was on par and often beat R300 in DX8 benches that were run. Yet again was beaten soundly in DX9 games until NV started to do massive shader replacement.

I also seem to remember there was quite a bit of hullabaloo about R300 cheating in order to beat NV35, as well as lots of hand waving that the only reason R300 was faster in early release Half-Life 2 benches was that Valve was obviously purposefully not optimizing the game for NV35.

Does that have any bearing on what is happening now? Doubtful.

Although there is a possibility that one or both of the cards will be significantly faster in either DX9 or DX10 vs. its competition.

At this point it's obviously too early to tell as the only indication of future DX10 performance was in 1 slide by AMD. Until actual proper reviews and analysis can be made of the cards, you can't just immediately call failure.

Regards,
SB

[leoneazzurro]

this is true ???



Here is the one of the key reasons why the R600 marchitecture falls behind the G80 GTX version. While Geforce 8800 GTX or G80GTX has 32 Texture Memory Units (TMUs) Radeon HD 2900 XT has 16 only.

TMUs are responsible for getting the textures in your games and the more you have, the faster you can handle the texture work.

Still, the Radeon HD 2900 XT at 750 MHz core is about 30 percent faster in raw clock speed from Geforce 8800 GTX at 575 MHz and finishes the texture cycle some 30 percent faster than Nvidia.

In spite of that this is not really an ideal solution as Nvidia has twice as many texture memory units. The only hope is that ATI has more effective TMUs, but this is one big marchitectural disadvantage over the G80 marchitecture.



It is not really true. Texture units in both chips are radically different in comparison to R580 or G71, and at the moment it's really difficult to compare them.
From the details we could say that in theory (based only on the unit number) R600 has an advantage in fetching textures and G80 has an advantage in filtering.

[Mintmaster]

And what's annoying me is that you think ATI has never written a co-issuing compiler ever before. R300 is a 4-issue ALU. R600 is 5-issue. R300's four instructions are all different (either in component count or capability or both). 4 of R600's ALUs are identical. The sky isn't falling in.Yet AGAIN you're missing the point. Nobody ever said ATI can't do it. DemoCoder simply asserted that it's more important to be good at it now.

1-clock instructions are extremely costly...I still don't agree with this unless you're talking about latency. Changing ops each clock isn't hard as long as you don't need the result immediately. Nearly every pipelined processor in the world, regardless of how simple, does this. You don't save much space by increasing this to more clocks.
I'm glad the penny's dropped. This is just one way that your sequencer complexity increases.

If instead of combining a 64-pixel ALU with a single-clock instruction pipeline your sequential scalar GPU has 16-pixel ALUs and four-clock instructions, you've still got increased sequencer complexity compared against R600, because you've just multiplied the number of batches in flight 4x in order to retain the same batch size (this is the "four Xenoses glued together" scenario).

So, now muse on how much of G80 is batch sequencing logic, since it has 16 batches in flight and compare that against the 4 batches in R600.I think one of the reasons we're having difficulty communicating is differing terminology when you say "batches in flight". Before G80 came it was 512 for R5xx and 6 (albeit enormous ones) for G70. Now you're talking about batches in the immediate vicinity of the ALU arrays, ignoring all the other batches in the pipeline.

Simple cycling between batches for predictable ALU instructions isn't hard, and doesn't add measurably to the sequencer complexity. The tough task is managing the many threads in flight that are waiting for texture fetch results. I don't consider what you're talking about to be sequencer complexity. I think the complexity arises from the larger data pool that each stage of the ALUs need to select from.

G80 may have more batches in flight in this sense, but the only reason for it to have more total batches in flight is the higher texture throughput.

[SugarCoat]

remember that rumor monger site..... how real can these be?

http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1&limit=1&limitstart=1

if they arent then obviously some people have way too much time. Thats the best attempt i've seen at discrediting a product if its all BS. It would be rather strange that they would go through all that to a product that is already badly disadvantaged by a severely delayed launch. Their "findings" arent going to mean a damn to anyone still left holding onto the "R600 dream" this late in and so close to the actual launch. It would also mean they have an agenda, again ask yourself, why? In all likelihood they're real.

From the looks of it, the HD 2900 XT is loosing to a X1950 XTX (yeah, right... :D).

With a new architecture if its poorly designed, or especially immature drivers, this is entirely possible. If i remember right (not to bring up this dead horse) the Geforce FX cards were actually slower or not giving near enough of a lead you'd expect in quite a few instances compared to the previous Geforce 4 cards in DX8 benchmarks at launch.

[Mintmaster]

It is not really true. Texture units in both chips are radically different in comparison to R580 or G71, and at the moment it's really difficult to compare them.
From the details we could say that in theory (based only on the unit number) R600 has an advantage in fetching textures and G80 has an advantage in filtering.How can you say that? R600 can do 16 fetches per clock, and G80 can do 32. G80 can do 64 bilinearly filter ops (8-bit per channel) per clock, and we don't know what R600 can do but it's probably 32 if we're lucky. Take clock differences into account and there's still a big disparity.

[Fornowagain]

Just revisiting the review (http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1) posted earlier they've added this, which is strangely exactly how I imagined him to look. :D Still think its fake?

http://img267.imageshack.us/img267/274/r600test12cj3.jpg (http://imageshack.us)

[leoneazzurro]

How can you say that? R600 can do 16 fetches per clock, and G80 can do 32. G80 can do 64 bilinearly filter ops (8-bit per channel) per clock, and we don't know what R600 can do but it's probably 32 if we're lucky. Take clock differences into account and there's still a big disparity.

Nope, from the supposedly real AMD slides there are 20 Texture fetch units in each of the 4 Texture Unit blocks of R600. That's 80 Unit total, and at higher frequency.

[Silent_Buddha]

How can you say that? R600 can do 16 fetches per clock, and G80 can do 32. G80 can do 64 bilinearly filter ops (8-bit per channel) per clock, and we don't know what R600 can do but it's probably 32 if we're lucky. Take clock differences into account and there's still a big disparity.

I'd imagine from this slide...

http://forum.beyond3d.com/showpost.php?p=980371&postcount=4164

It seems to imply that R600 can do 80 Texture Fetches per clock.

Regards,
SB

[AlexV]

if they arent then obviously some people have way too much time. Thats the best attempt i've seen at discrediting a product if its all BS. It would be rather strange that they would go through all that to a product that is already badly disadvantaged by a severely delayed launch. Their "findings" arent going to mean a damn to anyone still left holding onto the "R600 dream" this late in and so close to the actual launch. It would also mean they have an agenda, again ask yourself, why? In all likelihood they're real.



With a new architecture if its poorly designed, or especially immature drivers, this is entirely possible. If i remember right (not to bring up this dead horse) the Geforce FX cards were actually slower or not giving near enough of a lead you'd expect in quite a few instances compared to the previous Geforce 4 cards in DX8 benchmarks at launch.

Something to keep in mind:the difference in terms of work being done between DX8 and DX9 was rather huge(you went from a rather primitive programability to a much more general one, you went from integer to float etc.)-it`s not the same with DX9 to DX10(save for the GS which is new and as I`ve stated before could be a trump card). Again, IMHO, if you`re poor at DX9 math, you`re still going to be poor at it in DX10. If you suck at doing DX9 games, you`re not going to see huge improvements in DX10 UNLESS it involves some heavy Geometry shading and you`re adept at that.

Fact is, we don`t really know how good/bad R600 is(yet).

[Sound_Card]

Just revisiting the review (http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1) posted earlier they've added this, which is strangely exactly how I imagined him to look. :D Still think its fake?

http://img267.imageshack.us/img267/274/r600test12cj3.jpg (http://imageshack.us)


ROFL. Look very very very very close to the finer detail. I work with photshop almost everyday at school, and that is one horrible job. :razz:

I must say, that was a very valid try.

[digitalwanderer]

What makes you think that's a chop? I don't see anything tell-tale jumping out at me.

[Skrying]

Are the stickers on the fans in the same orientation?

[SugarCoat]

Something to keep in mind:the difference in terms of work being done between DX8 and DX9 was rather huge(you went from a rather primitive programability to a much more general one, you went from integer to float etc.)-it`s not the same with DX9 to DX10(save for the GS which is new and as I`ve stated before could be a trump card). Again, IMHO, if you`re poor at DX9 math, you`re still going to be poor at it in DX10. If you suck at doing DX9 games, you`re not going to see huge improvements in DX10 UNLESS it involves some heavy Geometry shading and you`re adept at that.

Fact is, we don`t really know how good/bad R600 is(yet).

regardless of your valid point, i think most people expect performance to be increased noticably across the board in pretty much every single thing that the previous gen did. I didnt mean to make it sound like i was saying DX8->DX9 was the same as DX9->10 but more so that when a new card is introduced, it does not always clobber its predecessor.

What makes you think that's a chop? I don't see anything tell-tale jumping out at me.

looks like a part number is blurred on the side of the HSF on every single card where its visible in the same way and the one against his abdomen has a very blurry back left corner where the 6pin and 8pin are suppose to be, could of just be taken by a cheapish camera though.

[leoneazzurro]

I'd imagine from this slide...

http://forum.beyond3d.com/showpost.php?p=980371&postcount=4164

It seems to imply that R600 can do 80 Texture Fetches per clock.

Regards,
SB

Yes, thanks. There are also some hints at the filtering capability, even if we don't know how much work per clock really each of the 16 TF units of R600 can do. Anyway what I heard is that R600 can not do trilinear per cycle.

[Bob]

It seems to imply that R600 can do 80 Texture Fetches per clock.
If you're going to start counting prefiltered texels, then G80 can do 256 of those per clock.

[Frank]

ROFL. Look very very very very close to the finer detail. I work with photshop almost everyday at school, and that is one horrible job. :razz:

I must say, that was a very valid try.
It looks real to me.

[leoneazzurro]

If you're going to start counting prefiltered texels, then G80 can do 256 of those per clock.

Sure? I knew 64. Have you some link to this info? I would appreciate very much :)

[nAo]

Sure? I knew 64. Have you some link to this info? I would appreciate very much :)
If it can provide 64 bilinear filtered 'samples' per clock then it should be able to read 256 texels per clock (from L1 cache) as well for obvious reasons.

[Fox5]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance - almost as if people are saying "Well, those specks are got to be good for SOMETHING, right?". I am personally trying to think of the last card that was slower then similarly-featured card in DX(n) but significantly faster in DX(n+1).

Of course, if anyone can point me to the specific features of R600 architecture and specific provisions in DX10, that, when combined, would deliver significant advantage over G80, I am all ears.

I'm sure some of ATI's r300 derivatives were slower than the 8500 yet could beat it in dx9...assuming 8500 could do dx9.
And the r300 derivatives were definitely slower than the geforce fx cards in dx8/8.1 but beat them badly in dx9.

[Mark]

Just revisiting the review (http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1) posted earlier they've added this, which is strangely exactly how I imagined him to look. :D Still think its fake?

http://img267.imageshack.us/img267/274/r600test12cj3.jpg (http://imageshack.us)
That dude looks like a character from Planet of the Apes...

[leoneazzurro]

If it can provide 64 bilinear filtered 'samples' per clock then it should be able to read 256 texels per clock (from L1 cache) as well for obvious reasons.

And to fetch all 256 to the ALUs? (because we are not speaking about filtering but about texture fetching to the ALUs)?

[NocturnDragon]

That dude looks like a character from Planet of the Apes...

He looks like a happy kid on Christmas day!

http://xs315.xs.to/xs315/07193/xmas.jpg

[Pete]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance - almost as if people are saying "Well, those specks are got to be good for SOMETHING, right?". That's exactly what I'm thinking. Surely ATI hasn't crammed so much into R600 to not have more to show than a GTS-beater, and even then conditionally? My skepticism waxes and wanes with each blurted NDA-buster.

I am personally trying to think of the last card that was slower then similarly-featured card in DX(n) but significantly faster in DX(n+1).One could argue that R300 (DX9), I think R520 (DX9 SM3 branching), and now apparently R600 (really long/complex shaders? HDR all the way?) are similar in that regard: trading off some current performance for greater anticipated future-API/IQ performance. If this is the case, it would seem that ATI is aiming a little beyond NV on the can-do/hope-to-do curve, and has fallen behind in two succeeding generations (R300 being on time relative to NV30). Or I'm oversimplifying, not seeing the forest, etc.

Of course, if anyone can point me to the specific features of R600 architecture and specific provisions in DX10, that, when combined, would deliver significant advantage over G80, I am all ears.This is the 64,000 pixel question, and I'm thinking we'll have to wait for NDA lift and possibly 3DM07 or simultaneous 360-PC releases like Shadowrun to do more than guess.

[Razor1]

That dude looks like a character from Planet of the Apes...


Good eye lol

[digitalwanderer]

He looks like a happy kid on Christmas day!

http://xs315.xs.to/xs315/07193/xmas.jpg
http://www.elitebastards.com/forum/images/smiles/biglol.gif

[Kaotik]

Regardless if it's a chop or not, the one on his lap / against his stomach does look too small compared to the others, even the one on his right knee looks bigger in my eyes, and it should be further away from the cam

[Silent_Buddha]

If you're going to start counting prefiltered texels, then G80 can do 256 of those per clock.

Ah ok, I think I see what you mean. If going by the block diagram and Jawed's interpretation of them. Then in one clock cycle R600 can do

16 Texture Fetches + 64 (fp32) Texel Fetches + EITHER 16 (fp32) Vertex fetches OR 16 (fp32) Texel Fetches.

And you're saying that G80 can do

32 Texture Fetches + 256 Texel Fetches in 1 cycle?

Regards,
SB

[Razor1]

Regardless if it's a chop or not, the one on his lap / against his stomach does look too small compared to the others, even the one on his right knee looks bigger in my eyes, and it should be further away from the cam


hard to tell from the angle, just did some filtering the artifacts SC is talking about are also there where the wall hits the ground, might be just bad compression. There is a wierd blue area with the one the guy is holding in the left hand though.

[jamis]

Regardless if it's a chop or not, the one on his lap / against his stomach does look too small compared to the others, even the one on his right knee looks bigger in my eyes, and it should be further away from the cam
That's because the one on his lap is not HD2900.

[Fornowagain]

That's because the one on his lap is not HD2900.
Thats right its a 2600.

[jamis]

looks like a part number is blurred on the side of the HSF on every single card where its visible in the same way and the one against his abdomen has a very blurry back left corner where the 6pin and 8pin are suppose to be, could of just be taken by a cheapish camera though.
It's not blurred, there is no power plugs. :wink:

[neliz]

I bet they tested the 2900 on the X2 6000+ system versus the E6700.

http://www.hexus.net/content/item.php?item=7913&page=6

add anywhere from 5 to 25% performance deficit to the r600 numbers..

[leoneazzurro]

Ah ok, I think I see what you mean. If going by the block diagram and Jawed's interpretation of them. Then in one clock cycle R600 can do

16 Texture Fetches + 64 (fp32) Texel Fetches + EITHER 16 (fp32) Vertex fetches OR 16 (fp32) Texel Fetches.

And you're saying that G80 can do

32 Texture Fetches + 256 Texel Fetches in 1 cycle?

Regards,
SB

http://forum.beyond3d.com/showpost.php?p=980606&postcount=4222

This is what Rys answered to me, now I don't know what's the right answer :???:

[SugarCoat]

I bet they tested the 2900 on the X2 6000+ system versus the E6700.

http://www.hexus.net/content/item.php?item=7913&page=6

add anywhere from 5 to 25% performance deficit to the r600 numbers..

with AA/AF and increasing resolutions? No way. Even with a conroe proc a GPU bottleneck will present itself more and more, especially as the resolution increases. The leads were pretty constant in ALL benchmarks, not just the lower res ones. A 6000+ isnt exactly a slouch either.

It's not blurred, there is no power plugs. :wink:

Hard to tell for me, i really dont like the flames and stock heatsink colour. If you didnt know what you were looking at you'd assume they were all the same.

[nAo]

And to fetch all 256 to the ALUs? (because we are not speaking about filtering but about texture fetching to the ALUs)?
Who cares about that? do you particularly like point filtered textures?

[Geeforcer]

ROFL. Look very very very very close to the finer detail. I work with photshop almost everyday at school, and that is one horrible job. :razz:

I must say, that was a very valid try.

Bodies of water, Africa, combat boots, heat sinks.

[hwfan12]

I bet they tested the 2900 on the X2 6000+ system versus the E6700.

http://www.hexus.net/content/item.php?item=7913&page=6

add anywhere from 5 to 25% performance deficit to the r600 numbers..

Hmm, very interesting... but there are other articles on the net about the subject.
Check this one, for example:
http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1 . It seems that R600 still has some room for improvement...

[Topman]

Raptor from techzone.pt has tested a DX10 demo (CoJ?) in HD 2900XT... < 15 fps in 1024 x 768... 8.37 drivers...

Time to forget high-end cards from ATi...


I think HD 2600XT will be a good card for your segment...

:)

[leoneazzurro]

Who cares about that? do you particularly like point filtered textures?

What I'm saying is that we don't know actually how many texture samples R600 can read from the cache, we know that R600 has 80 FP32 texture samplers units and G80 has , AFAIK, 64 and not 256 FP 32 units.
Then, I could be also wrong, and G80 in this case has simply around 4x more global texturing power per clock with respect to R600.

[jamis]

Hard to tell for me, i really dont like the flames and stock heatsink colour. If you didnt know what you were looking at you'd assume they were all the same.
Yeah, well as someone said it's 2600. It's smaller, single slot and even the "flames" are bit different than on the others.

[Pressure]

Hmm, very interesting... but there are other articles on the net about the subject.
Check this one, for example:
http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1 . It seems that R600 still has some room for improvement...

Which is what we are discussing here.

Currently we think they tested the Radeon HD2900XT in a whole AMD environment. AMD processor, AMD motherboard and AMD graphic card. A trinity if you like.

And the NVIDIA setup with an, obviously, superior Core 2 Duo 6700.

[nAo]

What I'm saying is that we don't know actually how many texture samples R600 can read from the cache, we know that R600 has 80 texture samplers units and G80 has , AFAIK, 64 and not 256 units.
No one is saying that G80 has 256 samplers (whatever this word means), I'm saying that if G80 can generate 64 bilerped samples per clock cycle it means it has to be able to read 256 4 bytes words from its texture cache(s).
That R600 leaked slides is not quoting 32 bits textures filtering rate so I suppose is the same as FP64 textures -> 16 per clock cycle -> 64 texels per clock retrieved from its texture cache.
Looking at the diagram you can see as some of this so called samplers are moving data to filtering units while other samplers are not, so I wouldn't be surprised if the other 16 samplers are used to fetch vertices and to support plain point filtering.

[Fornowagain]

I bet they tested the 2900 on the X2 6000+ system versus the E6700.

He does say.

[QUOTE][
There are two things we can conclude from our first round of testing on this pretty obsolete, but still the highest-end chipset on the market for Socket AM2./QUOTE]

So the X2 6000+?

[SugarCoat]

And the NVIDIA setup with an, obviously, superior Core 2 Duo 6700.

There is nothing obvious about it. High res high setting = CPU bottleneck negated. As i said as well, the 6000+ isnt a slouch. 15-20% or more performance from just the differing processors would only be possible by playing at rather low resolutions and shutting all ingame quality settings off. Doesnt anyone pay attention to CPU benchmarks and why they always reduce settings to crayon quality?

[neliz]

There is nothing obvious about it. High res high setting = CPU bottleneck negated. As i said as well, the 6000+ isnt a slouch. 15-20% or more performance from just the differing processors would only be possible by playing at rather low resolutions and shutting all ingame quality settings off. Doesnt anyone pay attention to CPU benchmarks and why they always reduce settings to crayon quality?

That's true. but when sites do review both low and high settings (hothardware) you can see that the gap increases when resolution and AA/AF options are applied.

Quake at 640x480 and 16x12 16af/4aa
http://www.hothardware.com/articles/AMD_Athlon_64_X2_6000/?page=9

FEAR at low-res low qual and then at a bottlenecked res 16x12 16af/4aa
http://www.hothardware.com/articles/AMD_Athlon_64_X2_6000/?page=10

oh well.. why am I actually arguing about something which I know is wrong?

[leoneazzurro]

No one is saying that G80 has 256 samplers (whatever this word means), I'm saying that if G80 can generate 64 bilerped samples per clock cycle it means it has to be able to read 256 4 bytes words from its texture cache(s).
That R600 leaked slides is not quoting 32 bits textures filtering rate so I suppose is the same as FP64 textures -> 16 per clock cycle -> 64 texels per clock retrieved from its texture cache.
Looking at the diagram you can see as some of this so called samplers are moving data to filtering units while other samplers are not, so I wouldn't be surprised if the other 16 samplers are used to fetch vertices and to support plain point filtering.

The 64 bilerps per clock are INT8, AFAIR, not FP32.
http://www.beyond3d.com/content/reviews/7/16

[nAo]

The 64 bilerps per clock are INT8, AFAIR, not FP32.
http://www.beyond3d.com/content/reviews/7/16
Yep, as the vast majority of textures you need to sample are INT8.
While filtering FP32 textures R600 runs still at half G80 rate per clock, according that slide.

[SugarCoat]

not al reviews do that. even at 1600x1200 16xaf 4xaa the E67 will still show better performance on something like a 1900...

the amount of reviewers that do CPU reviews at real world high res high quality settings is almost zero, the reason is because all the bars in their pretty graphs are within single digit frames of eachother so there is no point to it.

One of the more recent examples i can remember of how useless CPU speed is at high res gaming on a new game was in Oblivion and Driverheavens' 8800 launch review. They did the test at 1920x1200 with 4AA/16AF at 2.66GHz and then at 3.6GHz, ~40% improvement in clock speed, the result was a whopping 0-2% without HDR and a MAX of 5% improvement with HDR, this is from a 40% increase in clock speed. Therefore coming to the conclusion that a top end AMD processor is going to cause an automatic 15-25% performance loss in all new games (or at least the ones in question in the review) with simuliar settings or even more taxing ones, is quite simply ridiculous.

The most i could grant you is the 1280x1024 benchmarks might get a noticeable (5-15%) improvement on a 6700, but as i said, the other tests are fine and the trend is pretty much always there, so there is nothing horribly flawed with the results as far as what CPUs were used goes.

[leoneazzurro]

Yep, as the vast majority of textures you need to sample are INT8.
While filtering FP32 textures R600 runs still at half G80 rate per clock, according that slide.

Yes, but I already said G80 has huge advantage in filtering. But when unfiltered data are requested R600 has more units than G80. And it has also 20% more clock, so this also shorten the distance in filtering.
Anyway, I would really know why ATI was so conservative with TU.

PS: I saw a slide in which they claimed 10x filtering improvement on R580... seems really strange to me.

[Julidz]

hmm , so it isn't comparable


just one more question

R'600s architeture is Vec5 (vec4 + 1 scalar) ok ?



so , it can do a vec4 + vec1 instruction per clock ?

[digitalwanderer]

Did they mention what drivers they used in the review?

[Kaotik]

Did they mention what drivers they used in the review?

if you mean the it-review thing, no, they didn't.
they also "forgot" to mention nv drivers, chipset drivers, driver settings besides aa/af "level", operating system etc etc.

[digitalwanderer]

I can forgive 'em that, but without knowing the drivers used I can't tell spit about performance. :???:

[leoneazzurro]

hmm , so it isn't comparable


just one more question

R'600s architeture is Vec5 (vec4 + 1 scalar) ok ?



so , it can do a vec4 + vec1 instruction per clock ?

It's not really simply "vec5", but yes, it can do vec 4 + scalar

[nAo]

Yes, but I already said G80 has huge advantage in filtering. But when unfiltered data are requested R600 has more units than G80.
maybe they decided GPGPU is more important than games for them

PS: I saw a slide in which they claimed 10x filtering improvement on R580... seems really strange to me.
r580 could only filter FP16 and Fp32 textures using shaders..wouldn't be suprised if they're comparing thei new hw implementation to the old software implementation.

[INKster]

Raptor from techzone.pt has tested a DX10 demo (CoJ?) in HD 2900XT... < 15 fps in 1024 x 768... 8.37 drivers...

Time to forget high-end cards from ATi...


I think HD 2600XT will be a good card for your segment...

:)

He's not getting optimal results yet due to driver issues and problems regarding the somewhat "hacked" 8pin auxiliary power plug. ;)
BTW, it's techzonept.com, not techzone.pt. :D

[SugarCoat]

I can forgive 'em that, but without knowing the drivers used I can't tell spit about performance. :???:

Launch drivers are just as useless in my opinion. I dont call a battle won/lost until its at least 6-8 months after a launch, especially with a new architecture. The R520 was pretty weathered (badly delayed and released already for 3-4 months if i remember right) when it got its app specific OpenGL improvements. Personally if it does indeed perform as bad as it does in this many titles at launch im blaming driver immaturity.

[Kaotik]

He's not getting optimal results yet due to driver issues and problems regarding the somewhat "hacked" 8pin auxiliary power plug. ;)
BTW, it's techzonept.com, not techzone.pt. :D

6pin vs 8pin powerplug won't change your FPS at all, the card runs at default and even OC's with 2x6pin just fine, just overdrive won't work with 2x6pin

[Bob]

Yes, but I already said G80 has huge advantage in filtering. But when unfiltered data are requested R600 has more units than G80. And it has also 20% more clock, so this also shorten the distance in filtering.
Well, if you feel like using property A on GPU X with property B on GPU Y gives you a reasonable basis for comparison, more power to you!

Hint: Check your units.

[leoneazzurro]

maybe they decided GPGPU is more important than games for them

I don't think so, it's still a small niche market


r580 could only filter FP16 and Fp32 textures using shaders..wouldn't be suprised if they're comparing thei new hw implementation to the old software implementation.

What if in R600 they could use both HW and SW shader filtering?
Anyway, this makes me wonder... if R580 is SO texture limited, then the whole argument of most of current games being texture limited falls... because R580 performs not so bad even comparing to a 8800 GTS. And R600 has WAY more power than R580...

[Julidz]

It's not really simply "vec5", but yes, it can do vec 4 + scalar

that's because i've read in the inquirer that it's not vec4 + scalar , but superscalar vec5...what the hell ?

[leoneazzurro]

Well, if you feel like comparing property A on GPU X with property B on GPU Y gives you a reasonable basis for comparison, more power to you!

Hint: Check your units.

Sorry, I don't understand what you are saying.
Are you telling me G80 has 256 of something similar to R600 FP32 samplers?
Are you suggesting that 64 of R600 units can perform only fetch to filtering units so they are not available to fetching unfiltered samples?
What I know by reading is that G80 has 64 texture fetch and 64 texture filtering units. I can be wrong.

PS: I'm not an expert, I'm only a person wanting to learn :)

PS2: I understand now what your're saying, but maybe it was only me that could not explain well :)
I tried to say that having less filtering power (less filtering units) per clock, but higher clocks, R600 filtering power is less far to G80's than the mere unit comparison can say.
And it was not tied to what I was saying about fetch units.
Conclusion: if R600 has half of filtering power of G80, has also 20% more clock, so theoretical number peak power in this case is 60% than G80
R600 has more fetch units and 20% more clock, so in theory it should perform better in this regards, except if there are dependencies

[digitalwanderer]

Launch drivers are just as useless in my opinion. I dont call a battle won/lost until its at least 6-8 months after a launch, especially with a new architecture. The R520 was pretty weathered (badly delayed and released already for 3-4 months if i remember right) when it got its app specific OpenGL improvements. Personally if it does indeed perform as bad as it does in this many titles at launch im blaming driver immaturity.
I agree/disagree with you. I think you have to wait 3-4 months to until it gets fully optimized, but I think launch drivers are real damned important because that's what the most people are going to look at.

Bad launch drivers = bad launch, IMHO.

[Geeforcer]

I am with Digi. You can only get away with bad launch drivers if your hardware is otherwise absolutely superior. People who have been waiting for upgrade are not going to wait ANOTHER 3-4 months to see how the drivers shake out. 6-8 month is crazy - at this point the replacement cards will come in.

[INKster]

6pin vs 8pin powerplug won't change your FPS at all, the card runs at default and even OC's with 2x6pin just fine, just overdrive won't work with 2x6pin

Precisely.

[Geo]

Bad launch drivers = bad launch, IMHO.

<kaff> Radeon 8500 <kaff>

Nvidia thinks they got turned over a slow spit for G80 Vista drivers, but that'd been nothing compared to the howling they'd have faced if R600 had launched around the same time with working Vista drivers.

[digitalwanderer]

<kaff> Radeon 8500 <kaff>
Ah, but you forget I'm one of the minority who think that was the last of the awful ATi cards and not the start of the good ones. ;)

I never cared for the 8500. :razz:

[Geo]

Neither here nor there --just a gruesome example of how icky drivers on launch can break your heart (and damage your part's reputation).

[Silent_Buddha]

the amount of reviewers that do CPU reviews at real world high res high quality settings is almost zero, the reason is because all the bars in their pretty graphs are within single digit frames of eachother so there is no point to it.

One of the more recent examples i can remember of how useless CPU speed is at high res gaming on a new game was in Oblivion and Driverheavens' 8800 launch review. They did the test at 1920x1200 with 4AA/16AF at 2.66GHz and then at 3.6GHz, ~40% improvement in clock speed, the result was a whopping 0-2% without HDR and a MAX of 5% improvement with HDR, this is from a 40% increase in clock speed. Therefore coming to the conclusion that a top end AMD processor is going to cause an automatic 15-25% performance loss in all new games (or at least the ones in question in the review) with simuliar settings or even more taxing ones, is quite simply ridiculous.

The most i could grant you is the 1280x1024 benchmarks might get a noticeable (5-15%) improvement on a 6700, but as i said, the other tests are fine and the trend is pretty much always there, so there is nothing horribly flawed with the results as far as what CPUs were used goes.

Although there are some instances where games are also heavily CPU limited even at extremely high resolutions and graphics quality. Especially if they do any sorts of physics processing (cloth physics, particle physics, collision physics, etc)

I'm not sure about the games that were tested, but games such as EQ2 and Vanguard show huge differences in framerate even at 1920x1200 or 2560x1600 res.

Especially in EQ2, you'll get much more performance at high resolutions by upgrading your CPU than you will by upgrading your Graphics card.

Granted, neither of those two were benched. However, my point is that any game that uses lots of physics calculations will be both CPU and GPU bound even at very high resolutions. And Core 2 Duo I would expect to have much better performance with regards to physics calculations than an AMD X2.

Benching with different CPU's on different graphics card is just shoddy, lazy, and in extreme cases biased when you are trying to only compare the graphics cards.

Regards,
SB

[Geeforcer]

<kaff> Radeon 8500 <kaff>

Nvidia thinks they got turned over a slow spit for G80 Vista drivers, but that'd been nothing compared to the howling they'd have faced if R600 had launched around the same time with working Vista drivers.

I will say this: R600 drivers better delivery-room clean after all the "ggpwnd" statements they've made.

[Luminescent]

Any word on whether R600s texture filtering and addressing arrays are globally available to all units for fetch and filter or are they limited to certain ALU groups ala G80 and R580?

[Jawed]

Yet AGAIN you're missing the point. Nobody ever said ATI can't do it. DemoCoder simply asserted that it's more important to be good at it now.
I disagree, because if you treat R600 as a vec4+scalar architecture and feed it the same code as for R300 (vector or pixel shader), the throughput will be no worse.

Actually the DX8 modifiers seem like they could be a sore point in R600. What's the betting that, at the very least, source modifiers have to be issued as a distinct instruction? Pretty certain I'd say. So R600 will be slowed down compared with R300. Even the best compiler in the world can't make R600 fully overcome this deficit since R600's not wide enough for all combinations. But R300 has a pipeline hazard related to the DX8 modifier which means that R600 will claw back its loss in situations where R300 had to issue a NOP on the main ALU's prior clock.

The difference with R600 is that it's capable of running at higher instruction throughput and should average a higher percentage of its theoretical peak FLOPs. So the compiler writers have got something to get their teeth into.

It's important for maximum performance that the compiler is good. The difference is that this pipeline has a higher baseline to work from, even with a compiler that can do no more than issue a vec4/vec3 and (optionally) issue an alpha channel instruction, or issue a special function. Compared with R300, R600 has less corner cases. Well, at least on the surface ahead of NDA, anyway.

The way I see it, the dumbest compiler will get more out of the R600 pipeline than the same dumb compiler on R300. There are some corner cases centred on the DX8 modifier mini-ALU - see the CTM documentation if you feel like enumerating them. The example code snippet I gave last night is one of them, actually...

But, in conclusion, I think the stark simplicity of R600's ALU pipeline, when presented with vec3/vec4 + scalar code, makes it extremely hard to argue that "it will run worse than R300 unless the compiler is significantly better than R300's." At the same time I agree, R600 will benefit from a tricksy compiler, and those tricksy bits are new, uncharted, territory. When done right it'll show R300 a clean pair of heels.

I still don't agree with this unless you're talking about latency. Changing ops each clock isn't hard as long as you don't need the result immediately.
Well, I think the latency/pipeline-turnaround issue is a big deal.

Nearly every pipelined processor in the world, regardless of how simple, does this. You don't save much space by increasing this to more clocks.
It's interesting to note that CPUs with simultaneous multithreading tend to go with just 2 threads (batches, effectively). It really isn't easy to just keep ratcheting-up the number of hardware threads your pipeline will support. For each extra hardware thread you want to support, you have to correspondingly speed-up your "search" across available threads to identify what's issuable.

I think one of the reasons we're having difficulty communicating is differing terminology when you say "batches in flight".
I was careful to specify this at the beginning (sorry, that's a few hundred posts ago now), to mean merely the batch "at the end of the pipe" (or at any single position marked off in the pipeline, effectively) - because the actual number of batches in flight is subject to the pipeline length. Something we don't always know. But we know that both Xenos and R5xx use an 8-clock pipeline, with 2 batches each running for 4 clocks. We just don't know what R600 is and I tried to keep away from that issue.

Before G80 came it was 512 for R5xx and 6 (albeit enormous ones) for G70. Now you're talking about batches in the immediate vicinity of the ALU arrays, ignoring all the other batches in the pipeline.
Sorry about the confusion, it gets tedious to qualify terminology every time. It might be better to use the CPU terminology "hardware thread" I guess.

Simple cycling between batches for predictable ALU instructions isn't hard, and doesn't add measurably to the sequencer complexity. The tough task is managing the many threads in flight that are waiting for texture fetch results. I don't consider what you're talking about to be sequencer complexity. I think the complexity arises from the larger data pool that each stage of the ALUs need to select from.
In order to fill your pipeline's hardware threads, you have to have hardware that's fast enough to "survey" the status of all available batches.

A batch is in one of a number of states:

executing as both ALU and TU hardware threads (will wait for both to finish)
executing as an ALU hardware thread (waiting for clause completion)
executing as a TU hardware thread (waiting for texturing result to appear in destination register, will then enter state 4)
waiting to be issued to ALU
waiting to be issued to TU
waiting for instruction cache page-in
[other stuff I can't think of right now...]TU hardware threads have indeterminate latency, but in theory ALU hardware threads are fully predictable. Erm, except for when there's dynamic branching in the clause that's issued, etc.

G80 may have more batches in flight in this sense, but the only reason for it to have more total batches in flight is the higher texture throughput.
Well, with the suggestions of Arnold Beckenbauer and leoneazzurro

http://forum.beyond3d.com/showpost.php?p=983673&postcount=4812

http://forum.beyond3d.com/showpost.php?p=983679&postcount=4815

http://forum.beyond3d.com/showpost.php?p=983799&postcount=4847

it looks like R600 has rather more hardware threading than I surmised last night, so that means there's prolly rather more sequencer logic in there, in order to cut the batch size.

This also means that fine-grained ALU redundancy will cost more in terms of overhead, e.g. comparing 1-in-4 with the 1-in-16.

So, ahem, R600 right now looks significantly more costly there...

Jawed

[Jawed]

Any word on whether R600s texture filtering and addressing arrays are globally available to all units for fetch and filter or are they limited to certain ALU groups ala G80 and R580?
I think they'll be restricted, simply because of the difficulty of writing from the TU to an arbitrary register file location.

This question only seems to apply to R600. I think RV630 and RV610 are both small enough that there's only a single shader unit. The reason I say this is that both of them have just one RBE (ROP), 4 pipes. They're just like RV530 and RV510 in this respect - just that they have beefier SIMD and TU configurations inside that single shader unit.

---

You could interpret these diagrams as scalings based upon Xenos architecture. That would require the kind of routing you describe. I don't know how to affirm or deny this...

http://pcweb.mycom.co.jp/articles/2005/09/09/cedec1/images/012l.jpg

Jawed

[Rangers]

So from the It-review, which dovetails with a lot of the other rumors we've heard, X2900 is about 90% of a 8800GTS.

This product is a utter disaster for ATI. Why would anybody even buy it over a G80 product, even a 8800GTS, when it's such a power hog?

I can see sales being next to nil for this. As I said, the it-review shows it inferior to a 8800GTS. Lets say driver polishing or a stronger review CPU can get it to 100% or 105% of 8800GTS, even so, why would you purchase it when its so hot and power hungry? Therefore the market for this card is maybe 10% of people who just really prefer to buy the ATI brand. That's about it. The other 90% of people, the vast majority of sales will go to Nvidia.

I cant believe how bad ATI has gotten..

[WaltC]

...
I also seem to remember there was quite a bit of hullabaloo about R300 cheating in order to beat NV35, as well as lots of hand waving that the only reason R300 was faster in early release Half-Life 2 benches was that Valve was obviously purposefully not optimizing the game for NV35....



I don't recall anything remotely close to that during that period...;) What I recall most vividly about the period was that as nVidia was scrambling to design a competitive nV40, which took the majority of its design cues directly from R300, was nVidia being roundly and properly called on the carpet for literally advertising nV3x as an 8-pixel-per-clock gpu when it was later discovered to have been a 4-pixel-per-clock gpu from the start--which of course neatly explained why R300 ran away with most everything at the time, and why the performance discrepancies between the gpus mystified many of us for so long. The second best-deserved criticism of nVidia at the time that I recall was even while nVidia was bragging publicly about its "128-bit FP pipeline" (fp32) and stating flatly that ATi's R300 "96-bit pipeline" (fp24) was, quote, "not enough," unquote--nVidia was in fact configuring its drivers to run fp16 in the benchmarks where R300 was running fp24--but attempting to maintain the public illusion that nVidia was running fp32 the whole time.

This was found out, too, later on, and in fact it was also revealed that the first couple of official nV3x drivers from nVidia never even permitted fp32 operation of the gpu--but would in fact run at fp16 while reporting to the end user/customer that he was running at fp32 precision--again, even while nVidia PR was boasting about nV3x's wonderful fp32 capabilities. Of course it was obvious that such a bold and shameless sham would be found out, which it was. That also cleared up any minor performance discrepancies that emerged, as when after public pressure nVidia finally released drivers that put nV3x into fp32 mode, running at fp24 the R300 walked all over nV3x running in fp32. nV3x was also horrible at running SM2.0 code of any description compared to R300, which also explained neatly nVidia's vigorous protests about how "ATI and Microsoft were taking 3d gaming in the wrong direction."....;) It was the "wrong direction" for nVidia at the time, but it seemed to be exactly the right direction for everybody else...;) Then there were the benchmarks like Eidos' Tomb Raider benches that starkly showed how poorly nV3x was as an SM2.x gpu compared to R300, and nVidia was so incensed that the company pressured Eidos to actually can the benchmark and Eidos knuckled under to the pressure. And last but not certainly least was the hullabaloo begun by Extreme 3d and then our very own B3D, which demonstrated to the world how nVidia had deliberately compromised and cheated the 3dMark benchmark in order to produce benchmark scores that showed much better results than nV3x was actually capable of delivering. I cannot imagine that there is anybody alive who fails to clearly remember that...;)

To cap, I simply do not remember a single thing in the life of nV3x wherein ATi was accused of cheating--at least, I recall no such accusations by people who attempted even a modicum of objectivity about the situation. In terms of a comparison between nV3x and R3xx, the differences in the gpus were so great and so stark that the concept of ATi "having to cheat" to best nV3x in most every category, if not all of them, was simply required by nobody...;) Once we moved beyond the misleading falsehoods deliberately sanctioned and advanced by nVidia for nV3x, and we moved into the areas of fact in terms of what nV3x actually was as compared to R300, the notion of a cheating ATi simply wasn't required to illustrate the differences in the gpu performances.

I want to hasten to add that with nV40 nVidia turned the corner and showed that old dogs indeed can learn new tricks, and joined the 3d party that ATi started with R300, and I think we are all immeasurably better off because of it--even though it is clear to me at least that nVidia had absolutely no choice in the matter at all if it wished to remain a viable 3d gpu company for the long haul. The R3xx-nV3x saga, as sorry as it was, stands as a testament, I think, to the enduring value of competition and how it can improve everyone's lot over time. The companies that fall behind are the companies that fail to learn the new tricks that other companies teach them. It could happen to ATi as easily as it happened to nVidia back in '02, should ATi ever become so comfortable in its position that it feels it can rest on its laurels. I can imagine what a slap in the face R300 must've been in '02 to a cocksure nVidia convinced that after swallowing 3dfx whole it now reigned supreme in perpetuity in the 3d gpu marketplace--especially as nVidia failed to see--as most all of us did--just what a potent new bag of tricks ATi was bringing to the table in terms of the 9700 Pro. Indeed, I was no different, and it took some convincing for me to appreciate R300 at the time for what it was--but as the realization dawned after buying an R300--it was very easy to refrain from ever looking back.

I cannot say what R600 will in fact bring to the table in the current horse race. But my gut feeling is that it is going to be something very, very good, and probably pretty special in a number of ways. That's what I expect, anyway--and it's good to know that we have not very long at all to wait before discovering whether or not my gut feeling is on track. Competition is a wonderful thing!

[Jawed]

http://www.cupidity.f9.co.uk/b3d97.gif


What are those 5 ops? Is this the direct ancestor of R600's ALU organisation or is it something rather less exciting?

This question's been bugging me for a while, so hopefully someone understands the capability of the vertex ALU pipeline in ATI SM2/3 hardware.

Jawed

[cadaveca]

float4+float?

[Bob]

What I recall most vividly about the period was that as nVidia was scrambling to design a competitive nV40, which took the majority of its design cues directly from R300,
Yeah, NV40 was entirely architected from the group up, developed, layed out, emulated and tested in just 9 months, instead of the more usual 2-3 years. Or maybe your perception of reality has little to do with things like "facts" (which, as I hear, have a well known NVIDIA bias).

To cap, I simply do not remember a single thing in the life of nV3x wherein ATi was accused of cheating--at least, I recall no such accusations by people who attempted even a modicum of objectivity about the situation
Your memory is faulty (http://209.85.165.104/search?q=cache:atPtB0opyfkJ:www.digit-life.com/articles2/radeon/herc-r9800-r7500.html+radeon+9700+cheating&hl=en&ct=clnk&cd=10&gl=us&client=opera)

[SugarCoat]

Although there are some instances where games are also heavily CPU limited even at extremely high resolutions and graphics quality. Especially if they do any sorts of physics processing (cloth physics, particle physics, collision physics, etc)

I'm not sure about the games that were tested, but games such as EQ2 and Vanguard show huge differences in framerate even at 1920x1200 or 2560x1600 res.

Especially in EQ2, you'll get much more performance at high resolutions by upgrading your CPU than you will by upgrading your Graphics card.

Granted, neither of those two were benched. However, my point is that any game that uses lots of physics calculations will be both CPU and GPU bound even at very high resolutions. And Core 2 Duo I would expect to have much better performance with regards to physics calculations than an AMD X2.

Benching with different CPU's on different graphics card is just shoddy, lazy, and in extreme cases biased when you are trying to only compare the graphics cards.

Regards,
SB

Anyone who's played either or both MMOs knows that those engines perfom shoddy due to massive memory requirements and dont scale well with hardware in a normal sense compared to your FEAR or HL2. Vanguard in particular i found the average framerate badly, and i mean badly, limited by the amount of memory bandwidth available. So no offense to you, but i think you cherry picked two MMOs because they simply dont react like what most benchmarks show. Its also worth mentioning both games hardly react to SLI/CrossFire no matter the resolution and settings even though both do infact show plenty signs of GPU limitation. I disagree its mainly physics in either case as well, but more so, again, memory bottlenecking due to the design of the game itself, afterall its an MMO so theres a heck of a lot of info being passed around on the fly all the time. Oblivion has plenty of physics and hardly reacts at all to CPU speeds at very high settings as i already pointed out. Oblivion also doesnt need to render a massive area without loading where you can run in either direction for 10-15minutes without one hitch, and it doesnt need to track other player characters and their actions which puts significant strain on other components like the system memory and the HDD.

Im not saying the processor isnt important, it is, and im sure in time as software is tailored more and more for multicore, as we see things offloaded onto other cores such as physics, we'll see significant gains, but right now pretty much all gaming engines are mainly single threaded. The result is all you have to depend on to increase speed is raw MHz and in many instances this doesnt help because the bottleneck is infact elseware for those of us gaming at high settings high resolutions.

What they did can be considered lazy but i'd argue the bias part. The advantage they may of given the 8800 cards you and a few others are complaining about is MINIMAL at best. Literally 2-5% if they did average fps. At the highest res they did there i'd be pretty damn shocked if it wasnt under 2% for pretty much all of the tests due to the different processors.

[digitalwanderer]

Your memory is faulty (http://209.85.165.104/search?q=cache:atPtB0opyfkJ:www.digit-life.com/articles2/radeon/herc-r9800-r7500.html+radeon+9700+cheating&hl=en&ct=clnk&cd=10&gl=us&client=opera)
Futuremark's audit revealed cheats in NVIDIA Detonator FX 44.03 and 43.51 WHQL drivers.

Earlier GeForceFX drivers include only some of the cheats listed below:

1. The loading screen of the 3DMark03 test is detected by the driver. This is used by the driver to disregard the back buffer clear command that 3DMark03 gives...
2. A vertex shader used in game test 2 (P_Pointsprite.vsh) is detected by the driver. In this case the driver uses instructions contained in the driver to determine when to obey the back buffer clear command and when not to...
3. A vertex shader used in game test 4 (M_HDRsky.vsh) is detected. In this case the driveradds two static clipping planes to reduce the workload... This cheat was introduced in the 43.51 drivers as far as we know.
4. In game test 4, the water pixel shader (M_Water.psh) is detected. The driver uses this detection to artificially achieve a large performance boost - more than doubling the early frame rate on some systems...
5. In game test 4 there is detection of a pixel shader (m_HDRSky.psh). Again it appears the shader is being totally discarded and replaced with an alternative more efficient shader in a similar fashion to the water pixel shader above...
6. A vertex shader (G_MetalCubeLit.vsh) is detected in game test 1. Preventing this detection proved to reduce the frame rate with these drivers, but we have not yet determined the cause.
7. A vertex shader in game test 3 (G_PaintBaked.vsh) is detected, and preventing this detection drops the scores with these drivers. This cheat causes the back buffer clearing to be disregarded...
8. The vertex and pixel shaders used in the 3DMark03 feature tests are also detected by the driver. When we prevented this detection, the performance dropped by more than a factor of two in the 2.0 pixel shader test.

ATI was also accused of some cheats but they were not so strong.

Is that the bit you're talking about Bob? Because if so I think you're doing a bit of revisionism, you can't compare how ATi "cheated" to how nVidia "cheated" during that time. :lol:

Don't ya remember all the "magic drivers are coming!" talk, and the whole "REAL WORLD testing" that nVidia tried to blow smoke up our ass with back in those days? :|

[Silent_Buddha]

Anyone who's played either or both MMOs knows that those engines perfom shoddy due to massive memory requirements and dont scale well with hardware in a normal sense compared to your FEAR or HL2. Vanguard in particular i found the average framerate badly, and i mean badly, limited by the amount of memory bandwidth available. So no offense to you, but i think you cherry picked two MMOs because they simply dont react like what most benchmarks show. Its also worth mentioning both games hardly react to SLI/CrossFire no matter the resolution and settings even though both do infact show plenty signs of GPU limitation. I disagree its mainly physics in either case as well, but more so, again, memory bottlenecking due to the design of the game itself, afterall its an MMO so theres a heck of a lot of info being passed around on the fly all the time. Oblivion has plenty of physics and hardly reacts at all to CPU speeds at very high settings as i already pointed out. Oblivion also doesnt need to render a massive area without loading where you can run in either direction for 10-15minutes without one hitch, and it doesnt need to track other player characters and their actions which puts significant strain on other components like the system memory and the HDD.

Im not saying the processor isnt important, it is, and im sure in time as software is tailored more and more for multicore, as we see things offloaded onto other cores such as physics, we'll see significant gains, but right now pretty much all gaming engines are mainly single threaded. The result is all you have to depend on to increase speed is raw MHz and in many instances this doesnt help because the bottleneck is infact elseware for those of us gaming at high settings high resolutions.

What they did can be considered lazy but i'd argue the bias part. The advantage they may of given the 8800 cards you and a few others are complaining about is MINIMAL at best. Literally 2-5% if they did average fps. At the highest res they did there i'd be pretty damn shocked if it wasnt under 2% for pretty much all of the tests due to the different processors.

I would agree with you except that in my case. Going with faster memory netted next to zero improvement in EQ2, other than the initial stuttering when loading all textures for a zone or when new characters entered your field of view with armor textures that weren't already loaded into memory.

However, overclocking the CPU or better yet just replacing the CPU with a faster version see's rather large gains in performance, especially when you enabled balanced or higher settings.

Maybe I'll have to play Oblivion again, but number of instances that physics calculations is done in that game was absolutely minimal compared to EQ2. Especially if you have cloth simulation on with many people in capes, clothing, flying carpets, etc.

EQ2 is quite graphically nice to look at, especially when you get into the higher quality settings. However, it is also extremely CPU bound at all settings. Going from a X800XTPE to a X1800XT netted a whopping 1-2 fps more on average. Upgrading the CPU from a X2 3800 to a X2 5000+ netted closer to 5-15 fps on average depending on scene complexity at 1920x1200 res with 4x AA/16x AF.

My friend that just upgraded his 9700 pro to a 8800 GTX see's minimal (less than 4fps) gain in EQ2.

I will agree however that Vanguard is CPU, GPU and memory bandwidth limited. You will get increases in FPS with all those at ALL resolutions. Brad McQuaid has already gone to say that you cannot just put in a faster GPU and expect to get better performance from the game. It is limited by virtually all components in a system.

It's why you see two people with 8800 GTXs in game both running the same graphics settings and resolutions. Yet one complains he's only getting 15-20 FPS while the other one is saying he's getting 50-60 FPS.

To get away from those two games. Rise of Nations - Rise of Legends also benefits much more from higher CPU speed than it does from a faster GPU.
So it isn't just limited to those 2 games.

Yes, I used a couple of extreme cases. But it's to illustrate that just because you are at a resolution that is more likely to be GPU bound than CPU doesn't mean the CPU can't or won't affect framerates.

And to come to any conclusions at all where each GPU might be paired with a different CPU isn't serving any purpose.

Regards,
SB

[edit - last post on this as I think this is starting to go WAAAY off topic.]

[Topman]

Strange...

HD 2900XT 1GB > GF 8800 Ultra in Specviewperf 9?
Denny from xtremesystems has the cards... tests in same system...

http://www.iamxtreme.net/video/r600/8800gtx_12.PNG
http://www.iamxtreme.net/video/r600/x2900xt1024_02.png

Drivers?
HD 2900XT 'FireGL' optimized?
GF 8800 Ultra uses 'game' drivers, not Quadro drivers(FX5600)?

???

[INKster]

Strange...

HD 2900XT 1GB > GF 8800 Ultra in Specviewperf 9?
Denny from xtremesystems has the cards... tests in same system...

http://www.iamxtreme.net/video/r600/8800gtx_12.PNG
http://www.iamxtreme.net/video/r600/x2900xt1024_02.png

Drivers?
HD 2900XT 'FireGL' optimized?
GF 8800 Ultra uses 'game' drivers, not Quadro drivers(FX5600)?

???

The Quadro FX 5600 has double the memory (1.5GB).
The Quadro FX 5500 is actually an old G71 with 1GB of GDDR3 RAM.
The closest thing would be the Quadro FX 4600 (don't let the GTS-like PCB fool you, that is a full 128 sp, 384bit, 768MB GDDR3 G80 GPU).

[Sc4freak]

Just revisiting the review (http://it-review.net/index.php?option=com_content&task=view&id=1314&Itemid=1) posted earlier they've added this, which is strangely exactly how I imagined him to look. :D Still think its fake?

http://img267.imageshack.us/img267/274/r600test12cj3.jpg (http://imageshack.us)

It's photoshopped. One of his fingers on his left hand is eating into the cooler of the card on his leg.

[Rangers]

It's photoshopped. One of his fingers on his left hand is eating into the cooler of the card on his leg.

It's not photoshopped and the review is real, is my bet.

[digitalwanderer]

It's photoshopped. One of his fingers on his left hand is eating into the cooler of the card on his leg.
That's just the locking bracket pushing into his finger a bit.

[Silent_Buddha]

To cap, I simply do not remember a single thing in the life of nV3x wherein ATi was accused of cheating--at least, I recall no such accusations by people who attempted even a modicum of objectivity about the situation.

You mean like this article that was specifically written to determine whether ATI was cheating or that Valve was unfairly optimizing for ATI hardware as Nvidia's offering couldn't possibly be that bad?

http://www.anandtech.com/video/showdoc.aspx?i=1863&p=1

Or something like this, where Valve has to defend itself by stating that they spend 5x the amount on time on Nvidia's hardware as on ATI hardware? Something they had to do with the outcry that Nvidia's flagship couldn't possibly be that bad since it was quite competitive in many other games and thus Valve was either sabotaging performance on Nvidia's card or not optimizing for it.

http://www.gamespot.com/pc/action/halflife2/news.html?sid=6074909

Or maybe something like this where a gaming review site had to try to calm the masses that were all out to get Valve because they were unfairly optimizing for ATI?

http://www.computerpoweruser.com/editorial/article.asp?article=articles/archive/c0312/28c12/28c12.asp&guid=

I wasn't saying that there was a lot of hullabaloo about how ATI was cheating or that X company was unfairly optimizing for ATI from rational thinking people.

Just that there was a rather large and vociferous outcry by many so called "enthusiasts" at the time stating that ATI was either cheating or certain companies (mainly ones making DX9 games) were unfairly optimizing for ATI.

After all, if the FX5900 was competitive and sometimes faster in BF1942 or UT2003, it was obviously cheating that the 9700 pro and 9800 pro were so much faster in the Half-Life 2 benches that Valve presented. Or at least, that's what a lot of people would like to have thought back then.

It wasn't until months even years later, that the majority of people finally agreed that the 9700/9800 pro was just plain better at DX9.

Regards,
SB

PS - Found this thread amusing. Brings back memories of the times when the 5900 Ultra was consistently beating the 9800 pro in benchmarks and games.

http://discuss.extremetech.com/forums/327992001/ShowPost.aspx

Granted, this was before an actual DX9 game was released and thus it was only benches on DX8 games that people are talking about.

[tEd]

Will B3D have a review up on the day NDA lifts?

[Silent_Buddha]

Strange...

HD 2900XT 1GB > GF 8800 Ultra in Specviewperf 9?
Denny from xtremesystems has the cards... tests in same system...

http://www.iamxtreme.net/video/r600/8800gtx_12.PNG
http://www.iamxtreme.net/video/r600/x2900xt1024_02.png

Drivers?
HD 2900XT 'FireGL' optimized?
GF 8800 Ultra uses 'game' drivers, not Quadro drivers(FX5600)?

???

Er, stock HD 2900XT 1gig GDD4 vs stock 8800 GTX 768meg.

3dsmax-04
ATI 26.29
NV 23.68

Catia-02
ATI 19.80
NV 7.655

Ensight-03
ATI 32.65
NV 25.63

Light-08
ATI 24.28
NV 12.87

Maya-02
ATI 44.20
NV 46.82

Proe-04
ATI 13.79
NV 17.60

SW-01
ATI 30.29
NV 13.55

tcvis-01
ATI 8.280
NV 5.381

ugnx-01
ATI 46.89
NV 5.608

So does this mean that the R600 is a better workstation/3d rendering card? Just how relevant are these numbers to the workstation market?

Er wait, or is a lower number better in this bench?

Regards,
SB

[Geeforcer]

Just how relevant are these numbers to the workstation market?



3dsmax-04
NV 23.68
ATI 26.29
Quadro 4600 38.57

Catia-02
ATI 19.80
NV 7.655
Quadro 4600 47.67

Ensight-03
ATI 32.65
NV 25.63
Quadro 4600 41.95

Light-08
ATI 24.28
NV 12.87
Quadro 4600 38.14

Maya-02
ATI 44.20
NV 46.82
Quadro 4600 174.25

Proe-04
ATI 13.79
NV 17.60
Quadro 4600 44.28

SW-01
ATI 30.29
NV 13.55
Quadro 4600 85.87

tcvis-01
ATI 8.280
NV 5.381
Quadro 4600 23.09

ugnx-01
ATI 46.89
NV 5.608
Quadro 4600 25.28

[Russell]

It looks real to me.

I'm not sure. The one on the left leg seems way less blurrier than I would expect from a picture taken at this angle, especially compared ot how out of focus the leg its sitting on is. The one he's holding closest to him doesn't seem to blend with the background very well....what's with the kind of aura of colors there? Shouldn't we see pants behind it? Why the black fan on that one? That particular card also seems shorter. I could just be looking too hard for signs of tampering that don't exist simply because Sound_Card suggested it's fake. One thing's for certain is that I don't have a good eye for this sort of thing. Just throwing out what I see.

Even if it is real, why would he need 6 of them? It would be hard enough for a tester to get a pair for Crossfire testing. Why would a vendor, who would risk their ass getting by lending the guy a single card, chance lending him 6 of the damned things?

EDIT: Ok, didn't see the last two pages of this thread until I refreshed. So the card in front of him is a 2600. But why would he have a single 2600 and 5 2900's then?

And yeah, it really does look like his finger is passing through the frontmost card on his leg. Maybe he's a ghost? Would that explain the bluish/pinkish aura around his groin.

[trinibwoy]

Well the "2600" is casting a shadow so it's one hell of a photochop if fake.

[flopper]

I would agree with you except that in my case. Going with faster memory netted next to zero improvement in EQ2, other than the initial stuttering when loading all textures for a zone or when new characters entered your field of view with armor textures that weren't already loaded into memory.

However, overclocking the CPU or better yet just replacing the CPU with a faster version see's rather large gains in performance, especially when you enabled balanced or higher settings.

Maybe I'll have to play Oblivion again, but number of instances that physics calculations is done in that game was absolutely minimal compared to EQ2. Especially if you have cloth simulation on with many people in capes, clothing, flying carpets, etc.

EQ2 is quite graphically nice to look at, especially when you get into the higher quality settings. However, it is also extremely CPU bound at all settings. Going from a X800XTPE to a X1800XT netted a whopping 1-2 fps more on average. Upgrading the CPU from a X2 3800 to a X2 5000+ netted closer to 5-15 fps on average depending on scene complexity at 1920x1200 res with 4x AA/16x AF.

My friend that just upgraded his 9700 pro to a 8800 GTX see's minimal (less than 4fps) gain in EQ2.

I will agree however that Vanguard is CPU, GPU and memory bandwidth limited. You will get increases in FPS with all those at ALL resolutions. Brad McQuaid has already gone to say that you cannot just put in a faster GPU and expect to get better performance from the game. It is limited by virtually all components in a system.

It's why you see two people with 8800 GTXs in game both running the same graphics settings and resolutions. Yet one complains he's only getting 15-20 FPS while the other one is saying he's getting 50-60 FPS.

To get away from those two games. Rise of Nations - Rise of Legends also benefits much more from higher CPU speed than it does from a faster GPU.
So it isn't just limited to those 2 games.

Yes, I used a couple of extreme cases. But it's to illustrate that just because you are at a resolution that is more likely to be GPU bound than CPU doesn't mean the CPU can't or won't affect framerates.

And to come to any conclusions at all where each GPU might be paired with a different CPU isn't serving any purpose.

Regards,
SB

[edit - last post on this as I think this is starting to go WAAAY off topic.]

MMOG have games that stress the machine to the max.
Vanguard has 16x texture that for example WoW has.
Have a bottleneck in your system and playing vanguard will show it fast.

Each component in a machine needs to be top notch due to the requirements of coming MMOG titles. Conan will be dx10.

A lot of people look at cpu or gpu but ram and harddrive and overall systemspeed is neglected.

[flopper]

So from the It-review, which dovetails with a lot of the other rumors we've heard, X2900 is about 90% of a 8800GTS.

This product is a utter disaster for ATI. Why would anybody even buy it over a G80 product, even a 8800GTS, when it's such a power hog?

I can see sales being next to nil for this. As I said, the it-review shows it inferior to a 8800GTS. Lets say driver polishing or a stronger review CPU can get it to 100% or 105% of 8800GTS, even so, why would you purchase it when its so hot and power hungry? Therefore the market for this card is maybe 10% of people who just really prefer to buy the ATI brand. That's about it. The other 90% of people, the vast majority of sales will go to Nvidia.

I cant believe how bad ATI has gotten..

I am seldom suprised in hardware but more often in peoples judgement before they actually have a card in the machine.

Pure logic says,
in 3Dmark the card performs as the GTX from Nvidia, indicating good performance with the cards.
What seems then to be lacking are drivers that benefit from the performance.
Now, its to early to say how well the card will do since each driver version show improvement.
The card will be judged and set as to a GTS and the 65nm XTX will be here augusti/september ish?

Price/performance there will be a hard call,
windows XP or win Vista?
Want to play old dx9 games or the new DX10?
The gaming platform are changing, either we like it or not.
So, for me, what platform do I want to use for games and what games do I want to play?
For me, I look forward a new technology, Vista, crysis, alan wake, (he needs a wake up call, really) and other titles, mods and such and then there is Unreal engine 3.

The card to get for the future is ati.
If people think a little they will see that coming and besides, people buy card based on 3Dmark numbers more than anything else.
If someone has a mx440 and a socket A or a pentium 3 cpu and 512ram, I bet they upgrade the gpu but not much else.

I have a friend who do computers a lot, he play games, I told him 2gb ram will make the machine faster, well he got more, he agreed, he got a dualcore, he again went with a faster machine.

I am waiting to upgrade my X1950xtx which is a card that runs even with Nvidias GTS being a old card of an previous generation to a twice as fast card as 2900xt is.
I still just use a 1600x1200 resolution so it be nice to run stuff a tad faster again.

[AlexV]

I am seldom suprised in hardware but more often in peoples judgement before they actually have a card in the machine.

Pure logic says,
in 3Dmark the card performs as the GTX from Nvidia, indicating good performance with the cards.
What seems then to be lacking are drivers that benefit from the performance.
Now, its to early to say how well the card will do since each driver version show improvement.
The card will be judged and set as to a GTS and the 65nm XTX will be here augusti/september ish?

Price/performance there will be a hard call,
windows XP or win Vista?
Want to play old dx9 games or the new DX10?
The gaming platform are changing, either we like it or not.
So, for me, what platform do I want to use for games and what games do I want to play?
For me, I look forward a new technology, Vista, crysis, alan wake, (he needs a wake up call, really) and other titles, mods and such and then there is Unreal engine 3.

The card to get for the future is ati.
If people think a little they will see that coming and besides, people buy card based on 3Dmark numbers more than anything else.
If someone has a mx440 and a socket A or a pentium 3 cpu and 512ram, I bet they upgrade the gpu but not much else.

I have a friend who do computers a lot, he play games, I told him 2gb ram will make the machine faster, well he got more, he agreed, he got a dualcore, he again went with a faster machine.

I am waiting to upgrade my X1950xtx which is a card that runs even with Nvidias GTS being a old card of an previous generation to a twice as fast card as 2900xt is.
I still just use a 1600x1200 resolution so it be nice to run stuff a tad faster again.

Yeah, really, no argument there...coz history has really taught as that 3DMark is the most indicative thing in the universe when it comes to performance across the board. I like fanbase selective memory:everybody's favorite rag-on target, the NV30(5) retard child fared really well in 3DMark...guess it simply failed to have the drivers providing that performance. Honest.

The card isn`t to be judged as ATi want it, as they really can`t impose judgement to the whole wide world-no matter how they market it, it`ll be seen as their new high-endish thing, za new beast, if you like. ATi is a top-dog and it`s perceived as such, so you can`t really do a-"we`re more into mainstream now, don`t really care about high-end" thing in an instant. It won`t quite work. Considering that AMD aren`t retards(they're not), they`ll have to turn their 5.5 bln expense into an investment. ATM, the best way is to capitalize on ATi's current strengths(top-competitor, not middle-centric one) and slowly align them to their goals, because ATi still has quite some impetus from it`s single-virgin days, even considering latest slip-ups(R520).

What everybody seems to think is that AMD did a hugely stupid thing like going to ATi and telling them:gut the basterd so it`s no more than a GTS competitor, sell an expensive to make chip/card with a low-enough price to ensure your margins are crap, start a silly-price war with your no.1 competitor who has had just about 7 months of milking and improving fabrication process, so they`re likely to afford to cut on their current prices a bit-because we want to gain ?market-share? ?From whom?Intel?Intel has a huge market share because most of the lowish ended Celeron dudes will get some integrated crap and that`s that. And it`s Intel 9xx doodoo or something. But that ties to the fact that Intel has a HUGEish market-share...their CPUs drive their GPU market, AMD doesn`t have nearly enough of the low-end CPU market(where integrated matters the most) to warrant spending 5.5bln on getting some GPU making abilities.

If they want(and they probably do want, at least for the time being, because corporate mergers aren't like Legos) to capitalize on ATi's strength in the GPU market, they need a high-end part. Because high-end parts drive low-end aquisitions here. Because the bling reviews are on the high-end parts, not on some 1 fps 8200 or whatever. Even if that`s the major dough-maker. Because when Bob sees a Yahoo review or whatever crap mag/site he reads of the mega doodoo X a billion XTX he'll go with his 150$ to the store and reason the following:well, that X a billion XTX is cool...that a billion-600 one can`t be that much worse, can it?And it has a helluva lot of RAM too...gimme one.

So, with the risk of sounding like a broken record:everything that has been shown about the R600(save for the tech-specs, which are what they should be) doesn`t jibe with what would be logical/efficient economically. We're talking about a business here, so their prime goals are of an economic nature(no, they`re not into it for the good of the planet, manking, shits and giggles etc., they`re in it for the money), thus I think that either ATi made some poor choices for the R600 somewhere along the line OR what we know ATM is skewed by something else.

[nicolasb]

Okay, another question: is it likely (or even possible) that games coming out in the next year or so will put comparatively more emphasis on shaders and less on texturing, as compared with games that have already been released? And, regardless of whether they do or not, is it possible that R600 was designed on the assumption that they would?

I ask because I'm not convinced by the argument "R600 is designed for DX10 rather than DX9". But "R600 is designed to colour pixels using shaders rather than textures" might be a bit more plausible. If games do actually go that way then we might see R600 pulling ahead of G80 in time. (Or not, if they don't).

[Nebula]

Well the "2600" is casting a shadow so it's one hell of a photochop if fake.

The card against his shirt is making a pressure mark on his shirt and the bracket for that card is casting a shadow on his shirt. Doesn't seem fake though.

The photo of the man with the cards was ~75KB and therefore the IQ.
http://i176.photobucket.com/albums/w193/NebulasPhotoPocket/r600test12cj3.jpg

[CarstenS]

http://www.cupidity.f9.co.uk/b3d97.gif


What are those 5 ops? Is this the direct ancestor of R600's ALU organisation or is it something rather less exciting?

This question's been bugging me for a while, so hopefully someone understands the capability of the vertex ALU pipeline in ATI SM2/3 hardware.

Jawed
Do not confuse "ops" with "instructions". A trivial pixel is already a 4-op when it's using RGBA-channels. With Vertex-ALUs, it has been a long time since they got 4+1 wide.

[Pressure]

So from the It-review, which dovetails with a lot of the other rumors we've heard, X2900 is about 90% of a 8800GTS.

This product is a utter disaster for ATI. Why would anybody even buy it over a G80 product, even a 8800GTS, when it's such a power hog?

I can see sales being next to nil for this. As I said, the it-review shows it inferior to a 8800GTS. Lets say driver polishing or a stronger review CPU can get it to 100% or 105% of 8800GTS, even so, why would you purchase it when its so hot and power hungry? Therefore the market for this card is maybe 10% of people who just really prefer to buy the ATI brand. That's about it. The other 90% of people, the vast majority of sales will go to Nvidia.

I cant believe how bad ATI has gotten..

Wow, slow down there. First off, the entire market does not consist of High-end solutions. In fact, that is were the minority of money are made.

If AMD can make their Mid-range and low-end better than their NVIDIA counterparts, they should be golden.

Besides, the biggest growing market are notebooks and not desktop models.

[leoneazzurro]

Okay, another question: is it likely (or even possible) that games coming out in the next year or so will put comparatively more emphasis on shaders and less on texturing, as compared with games that have already been released? And, regardless of whether they do or not, is it possible that R600 was designed on the assumption that they would?

I ask because I'm not convinced by the argument "R600 is designed for DX10 rather than DX9". But "R600 is designed to colour pixels using shaders rather than textures" might be a bit more plausible. If games do actually go that way then we might see R600 pulling ahead of G80 in time. (Or not, if they don't).

Question is that we don't know if really R600 is SO texture limited. Its theoretical capability should be way off R580, and R580 is not faring so bad against the GTS in present games. Also theoretical math throughput is better, and efficiency of an unified architecture should be also better. Yes, there are a lot of "should be" but I'm beginning to think there's another reason (bottleneck? Scheduling? lack of registers?) for the worse than expected performance of R600.

[Unknown Soldier]

The notion that R600 is "born for DX10" strikes me personally as a reaction a lower than expected performance

Remember when Orton announced(almost a year ago if not longer) that the R600 would be the fastest DX9 card ever, well since then the G80 was released and I think ATI had to bite deep and now will settle for Fastest DX10 card. ;)

US

[nicolasb]

Remember when Orton announced(almost a year ago if not longer) that the R600 would be the fastest DX9 card everActually he announced that it would be ATI's fastest DX9 card ever (which it is). He didn't say it would be faster than G80. (Although he probably believed it would be; I think even Nvidia was a little surprised by G80).

[neliz]

Remember when Orton announced(almost a year ago if not longer) that the R600 would be the fastest DX9 card ever, well since then the G80 was released and I think ATI had to bite deep and now will settle for Fastest DX10 card. ;)

US

I checked that quote.. a hundred pages back or so and Orton said it would be their fastest DX9 card.

[Kaotik]

According to Sampsa, on his system with Enermax 850W PCIe 2.0 compatible PSU, the system has peaked @ 298W - that's for the whole system, not just HD2900, so I guess the 225W+ figures end up being quite far from the truth in real world applications :cool:

edit:
For those who can read finnish; http://plaza.fi/muropaketti/bbs/t470333,850
Dunno if it's in english somewhere (extremesystems?)

[CarstenS]

According to Sampsa, on his system with Enermax 850W PCIe 2.0 compatible PSU, the system has peaked @ 298W - that's for the whole system, not just HD2900, so I guess the 225W+ figures end up being quite far from the truth in real world applications :cool:

That would depend on the load he's driving his peaks with (i hope it's not Atitool...he'd better check his clockrates) and of course with the rest of the system's components.

For example running one giant MADD-Kernel on the GPU with the Rest of the System basically sitting idle would rather strengthen the 225 Watt assumptions.

FWIW with my testing rig, i currently top out at about 299 Watts for the whole system with a GF8800 GTX.

[Kaotik]

That would depend on the load he's driving his peaks with (i hope it's not Atitool...he'd better check his clockrates) and of course with the rest of the system's components.

For example running one giant MADD-Kernel on the GPU with the Rest of the System basically sitting idle would rather strengthen the 225 Watt assumptions.

FWIW with my testing rig, i currently top out at about 299 Watts for the whole system with a GF8800 GTX.

Well knowing Sampsa, I'm fairly sure those peaks are from 3DMark06 which under his experience has given the highest peaks in his normal testing routines of cards

The rest of the system components, hard to say for sure, but based on his last reviews, I'd guess there's C2E X6800, i975X based mobo, 2x 1GB mem and SATA2 HDD in there

[Jawed]

Do not confuse "ops" with "instructions". A trivial pixel is already a 4-op when it's using RGBA-channels. With Vertex-ALUs, it has been a long time since they got 4+1 wide.
So, what is one op? Why is one op the norm?

Why would you sometimes get 2?

Jawed

[Unknown Soldier]

Sorry, I was wrong, it was Richard Huddy that said it and the context implies that it would be the fastest DX9 card. Remember that the X1950XTX was faster than the G70 and I believe he was implying(or at least thought) that the R600 would be faster than the G80.

“The R600 will be [absolutely] the fastest DirectX 9 chip that we had ever built,” said Richard Huddy, the head of ATI Technologies’ software developers relations department, at a press conference in London, UK.

http://www.xbitlabs.com/news/video/display/20060525104243.html

You read it as the fastest DX9 card that ATI built, I read it as the fastest DX9 card ever.

I still do think that the G80 shocked ATI and might even have caused the delay of the R600.

US

[CarstenS]

So, what is one op?
ADD, MUL, MAD - you name it.

Why is one op the norm?
Because that is what you'll get normally. If there's no macros involved for example.


Why would you sometimes get 2?

Because sometimes it could be possible to schedule a scalar op in addition to the normal vec4-op.

[CarstenS]

Well knowing Sampsa, I'm fairly sure those peaks are from 3DMark06 which under his experience has given the highest peaks in his normal testing routines of cards

The rest of the system components, hard to say for sure, but based on his last reviews, I'd guess there's C2E X6800, i975X based mobo, 2x 1GB mem and SATA2 HDD in there

Thanks, but 298 Watts seems pretty low to me.

[neliz]

I still do think that the G80 shocked ATI and might even have caused the delay of the R600.

US

It shocked ATI because it was twice as fast as their previous gen GPU? I don't think so.. that's quite the norm already. Plus we've seen that the GPU's in the first batches were produced in the beginning of February, a much more logical reason would be a terrible yield of the die. I've heard there won't be much cards in europe for launch (somewhere between 1000 and 2000) but the cards should be coming in by the thousands near the end of the month.

[vertex_shader]

I not listen often to fruitzilla, but here (http://www.fudzilla.com/index.php?option=com_content&task=view&id=922&Itemid=1) is a very good and true write.

I can clearly say they misled not only the investors, they misled the users, partners too.

[NocturnDragon]

3dsmax-04
NV 23.68
ATI 26.29
Quadro 4600 38.57

SNIP

ugnx-01
ATI 46.89
NV 5.608
Quadro 4600 25.28

What's that supposed to mean? The quadro has optimized drivers of those apps...
While neither the 8800 ultra nor (that we know of at least, and I don't think they just unified the consumer drivers with the pro drivers) the R600 are optimizied for those taks.
It would be a fair comparison between the quadro and the new FireGL.

[Mariner]

Page 200! :shock:

Thank heavens that the card itself will be released soon and we'll be able to shut down this monster thread.

[neliz]

I not listen often to fruitzilla, but here (http://www.fudzilla.com/index.php?option=com_content&task=view&id=922&Itemid=1) is a very good and true write.

I can clearly say they misled not only the investors, they misled the users, partners too.

Are they launching 10 cards Monday, or not?

If someone says that the 610 and 630 had serious bugs, they need to get their brains checked, their launch was planed for May/June for almost half a year.

vr-zone's roadmap (oct. 2006):
http://www.vr-zone.com/?i=4188

vr-zone's roadmap (aug. 2005)
http://www.vr-zone.com/?i=2612&s=1
So HEY!.. they missed one quarter with R600...

Softpedia (oct 6. 2006)
http://news.softpedia.com/news/AMD-ATI-Roadmap-Leaked-37324.shtml

[Unknown Soldier]

Sorry Neliz, I'll disagree with you. Huddy was already talking R600 in May '06 and the fact that they had an advantage with Xenos already done and the fact that the R600 is supposed to be like the Xenos except pumped up with drugs, everyone was expecting ATI to release the R600 by November, December at the latest. People even believed that the R600 would be released before the G80. They didn't and everyone expected January at the latest, then it was announced towards the end of January that the release would be March and then later announced that the March deadline would not happen.

I still think the performance of the G80 had an impact, and it didn't help that it had unified shaders.

Remember that the R600 was supposed to be released around the time of Vista .. or so people believed.

US

[Razor1]

What's that supposed to mean? The quadro has optimized drivers of those apps...
While neither the 8800 ultra nor (that we know of at least, and I don't think they just unified the consumer drivers with the pro drivers) the R600 are optimizied for those taks.
It would be a fair comparison between the quadro and the new FireGL.


Work station cards have certain features activated over the regular consumer cards (and some ogl extenstions), which do improve performance in certain 3d apps, like Max and Maya or other digital art programs.

[neliz]

Sorry Neliz, I'll disagree with you.

Remember that the R600 was supposed to be released around the time of Vista .. or so people believed.

US

True, I was expecting that too, but if you look at all the "hard" evidence in hindsight (leaked road maps) they did expect it to be released at the turn of 06/07, with March at it's latest.
But that's just our enthausiasm playing there.

http://www.vr-zone.com/?i=2612&s=1

Some interesting info from an investor meeting held with ATI's CEO Dave Orton on August 17th 2005.

• R600 targeted for (DirectX 10, targeting Vista and WGF 2.0) in Q4/F06 or Q1/F07

[nicolasb]

Are they launching 10 cards Monday, or not?If you believe The Inquirer, not.

http://www.theinquirer.net/default.aspx?article=39526

It's quite a long article, but a couple of choice quotes:

When the X2k launch day comes, next Monday, a single major SKU will be tipping up - the Radeon HD 2900 XT. Those journos nominally bound by EhnDeeEhs have been playing with the card for a few weeks now, and the results don't look good.

The card goes tit for tat with the GeForce 8800 GTS, winning some and losing some in a vaguely equal proportion, whilst coming in at the same price point.

Meanwhile, the XTX languishes in no-man's land, available only to system builders stupid enough to pay for a 1GB DDR4 card with barely better performance.

The delay of the Radeon series was, at the last check, designed to enable a top to bottom launch of the series. Forgeddaboutit. We're hearing from multiple sauces that the 2400 and 2600 are both delayed until the beginning of July, as the metal goes back for another respin. The silicon just ain't there, yet.

[vertex_shader]

I still do think that the G80 shocked ATI and might even have caused the delay of the R600.

US

You mean ATi eat the misleding g80 interviews what is come from the other side? :wink:
It's a shame when ATi eat this, and looks like ATi not have any good factory "sources".

[Unknown Soldier]

Ye :)

And i'd hate to be called a fanboi. ;)

US

[Unknown Soldier]

You mean ATi eat the misleding g80 interviews what is come from the other side? :wink:
It's a shame when ATi eat this, and looks like ATi not have any good factory "sources".

And maybe they did have good 'sources' hence the delays from end '06 to beginning '07. ;)

That said the delay at end March for a Family release in May '07 was also lame.

US

[neliz]

According to this april 2006 article, they never were planning on beating G80 in 2006.

http://www.theinquirer.net/default.aspx?article=30745

[trinibwoy]

If R600 sucks ATI would be the first to know so I doubt they were surprised by any of this. Hopefully it's just software issues at work here cause I can't fathom how that calibre of hardware could be having trouble with a significantly hobbled G80 part.

[Rangers]

The disaster just keeps getting worse..and just as AMD was taking Nvidia market share with bundles.

They've got to get some more texturing capability in their cards or they're dead, it's that simple. Who knows if they care right now, though.

If R580 was any indication, I'd expect all their mid-range parts to suck also, as they slavishly maintained some artificial crippling texture/shader ratio even at the lower ends last gen. I mean, they put out a 4 TMU mainstream part. It took them months to recover from that last time and get decent mid-range parts out.

[Dalton Sleeper]

At first I was waiting for 2*2900XT, but now it almost seem that 8800 ULTRA is a better choice. But I'll wait a few days hoping to see better results.