The NEXT LAST R600 Rumours & Speculation Thread [Archive] - Page 13

jimmyjames123

26-Apr-2007, 19:56

you dont see the glaring problem in your view? The cards launching almost 6 months after nVidia and if it offers nothing really earth shattering thats really about as bad as you can get short of performing worse in many titles as well.

The XT is also launching at a reactive price DUE to its late introduction so i think you're being far too lenient. IF the card had launched within a reasonable time frame to the competition it would easily have been a $499 part rather then a $399. They've basically been forced into the $399 pricepoint, that you think looks so good, due to how late the launch is.

We're also talking about the flagship card as well until an XTX shows its face in quantity, the same goes for any 8800Ultra, the GTX is still the flagship card. This launch, at least from this distant perspective that leaked information is offering, is turning out to be as exciting as the X800XT was against the 6800Ultra (not a huge performance gap or large advantage in IQ), except this time instead of launching withing a few weeks of eachother ATi is coming in 6 months later which makes it look terrible.

I don't claim that it is a rosy picture for AMD. No doubt, it would have helped their margins tremendously to launch earlier if they could have done it at a reasonable level of performance/reliability/availability. However, just because they can't quite match the 8800 GTX after six months doesn't mean everything is a failure. If the 512MB XT can be equal to or outperform a 640MB GTS, probably have an edge in DX10 games, and sell for a similar price point (noting that it does have 128MB RAM less in comparison, which would reduce RAM cost in comparison), then that is better than nothing.

Jawed

26-Apr-2007, 19:57

Not asking you, specifically, Jawed, but quoting for context, and assuming an architecture somewhat based on Xenos as above...

How on God's green earth can they take a 232M transistor chip, add 16 more shader ALUs (vec 4+1 or vec5) to the existing 48, and end up with 700M+ trannies????

For cripes' sake! Even adding a GS and more int functionality...surely D3D10 compliance does not require 400M transistors. :???:

I'm really missing something here.
(I realize the 232M figure does not include the ROPs, etc. but still...)

ERK
Considering that ALUs don't consume much of the die anyway, the picture's even more puzzling. So, ahem, D3D10 "done right" :lol: may well cost 400M transistors :lol: :lol:

There was a time when R600 was thought to be 500/550M transistors (at least by me...) so the whole damn thing is just a mystery.

I do expect lots of RAM - but I've been saying that for years practically and, well, RAM doesn't take that much space...

Jawed

Geeforcer

26-Apr-2007, 19:59

I do expect lots of RAM - but I've been saying that for years practically and, well, RAM doesn't take that much space...

Jawed

Space, no - but it can drive the transistor count up pretty hard.

neliz

26-Apr-2007, 20:00

neliz, who did you pre-order from?

Salland, the dutch e-tailer showing up yesterday.

ERK

26-Apr-2007, 20:01

Thanks! I'm glad I'm not the only one slightly baffled by some of this.

ERK

ChronoReverse

26-Apr-2007, 20:02

Hrm. If these benchmarks are true, AMD would be wise to release a REALLY uber midrange card. They'll lose the top-end but will get a good reputation in the section that really counts.

jimmyjames123

26-Apr-2007, 20:05

A success for consumers how? Consumers have had 'as good' a card since last year...

From the benchmarks shown so far, isn't the 512MB XT as fast or faster than the 640MB 8800 GTS, and don't most people believe that the XT will be stronger in DX10 games? From a positioning standpoint, I don't think that the stock 640MB GTS would be in a very good position against the XT, since it has in some cases lower performance and it also has higher RAM costs with extra 128MB RAM on board. Of course, it's nice to have six months with no competition in that segment, but clearly something needs to happen now for the GTS, be it lower prices/reduced margins, and/or performance improvement (pushing clock speeds higher, etc).

On a side note, releasing an 8800 Ultra won't do anything to give the 640MB 8800 GTS more favorable footing against the XT. The only way to do that would be through performance and/or price improvements for the 640MB 8800 GTS (which is not hard to imagine, given that prices are already starting to come down to below $400, and given that higher clocked 640MB GTS in some cases comes very close to the performance of a stock GTX).

Sound_Card

26-Apr-2007, 20:12

Jawed

26-Apr-2007, 20:12

So, next week, are we expecting to hear about the next version of 3DMk? Will there be any further reason to hold it back?

Jawed

jimmyjames123

26-Apr-2007, 20:16

Sound_Card

26-Apr-2007, 20:17

Who are you directing this unintelligible statement at? Are you drunk man? :D

I could use a beer actully.:shock:

Lux_

26-Apr-2007, 20:19

Will there be any further reason to hold it back? R650:wink: :lol:

jimmyjames123

26-Apr-2007, 20:20

You know what, now I'm really puzzled as to why AMD held back the launch date of the 512MB XT model. Clearly, there must have been something holding them back other than "family launch". What happened, Hector go on a long vacation or something? :D

nAo

26-Apr-2007, 20:24

How on God's green earth can they take a 232M transistor chip, add 16 more shader ALUs (vec 4+1 or vec5) to the existing 48, and end up with 700M+ trannies????
That chip is far from being DX10 complaint (no integer support for example..) and also it does not carry the weight of compression techniques being used on modern GPUs that does not have an embedded frame buffer, so you can't just compare them on that basis.
Moreover we still know nothing about R600 architecture..and I can't wait to find out more about it!

CarstenS

26-Apr-2007, 20:33

How on God's green earth can they take a 232M transistor chip, add 16 more shader ALUs (vec 4+1 or vec5) to the existing 48, and end up with 700M+ trannies????

For cripes' sake! Even adding a GS and more int functionality...surely D3D10 compliance does not require 400M transistors. :???:

I'm really missing something here.
(I realize the 232M figure does not include the ROPs, etc. but still...)

You're going to need (almost) the whole memory controller, all of the ROPs including their requirements for DX10 (FP-Texture filtering, FP-Alpha-Blending, Compression etc.), INT-functions, massive amount of Registers (if you don't want to cripple your chip - DX10 requires 4.096 of them at any one time, though not necessarily all of them in hardware), and the whole lot of things i just plainly forgot.

Just take a look at the TC of the smaller G8x-Chips: You'll get 2 Shader-Quads with DX10 for the transistor-price of about 6 Quads DX9-Style. :)

Geeforcer

26-Apr-2007, 20:38

pjbliverpool

26-Apr-2007, 20:41

One does have to wonder what it means for Xenos's actual performance if these benchmarks are real. Afterall it should be a no brainer that the XT is A LOT faster than Xenos.

Kaotik

26-Apr-2007, 20:41

Sound Card,

While I think it's never a good idea to take one source as gospel, Dailythech has shown to be fairly reliable in the past with their pre-release information. Since you are skeptical (which is not a bad thing by any measure), what do you suggest happed here?

1) They just made up the numbers. To me that seems unlikely - they are not the kind of site that would trade a short-term hits spike for a long term credibility gap.
2) They did the test to the best of their ability but something was wrong: Card, drivers, system in general, etc. This would be just reporting and if the numbers are off, oh well - they told us what they saw. Still, it would be in their best interests to make sure that doesn't happen because their credibility will be damaged.
3) The numbers are accurate.
4) Alternative explanation.

Well one thing which in my eyes is lowering their reliability is the claim that XTX would be released at the same time as XT, while someone who actually attented the tech day at Tunis said that there won't be XTX, also their clocks for the XT are wrong.
And I still wonder how changing CPU from QX6700 to QX6800 suddenly doubled their Oblivion scores :D

Geeforcer

26-Apr-2007, 20:46

Changing the setting and the area where you benchmark in Oblivion can have huge effect on your scores.

trinibwoy

26-Apr-2007, 20:51

It's doom and gloom before the second can make a move in their defence? What I see is a monkey zoo having the last laugh before their was ever a first. Let's dare not qustion the evidence and instead danm it law? :roll:

Agreed but what else did you expect given the current info? All we have is a history of delays and broken promises and now underwhelming performance. What other reaction should a rational person have? Yes, DT's numbers are far below expectations given the technical specs but it's all we have. If you have some good reason for people to be optimistic I'd love to hear it.

Silent_Buddha

26-Apr-2007, 20:53

Probably from Salland Automatiseringen. They took the Sapphire HD2900XT of their site though after some phonecalls from AMD and Sapphire. But they said that all people who pre-ordered the HD2900XT will not have to worry. All pre-orders will remain in their computers and they had some more good news, but they couldn't tell people about it yet (NDA).

Maybe the fact that it might actually ship on May 2nd, if VR-Zone is actually correct in the launch date being pushed forward? Forcing to the reseller to push out the "official" ship date to well beyond the actual launch date could be just to obscure when the retailer will actually be allowed to ship said card since it wasn't supposed to be even available for pre-order.

A rather light slap on the wrist IMO for breaking NDA on ship/release date.

I'm still waiting for high IQ comparisons, and "real" reviews before declaring R600 a total disappointment.

People also seem to think that one bad product immediately puts a company out of business. I guess few remember when ATI was a top Video Card maker pre-1997 with their 2D cards. Failed misearably with virtually all their 3D cards until R300, and yet still managed to survive.

And apparently Matrox is still finding a way to survive by focusing on their niche market.

It's possible that ATI will cease to be a serious contender for the high end crown year to year as they have been since the R300, yet that will hardly doom the company to oblivion. After all, AMD mainly purchased ATI for their MB chipsets and integrated graphics. Worst case is that ATI will be relegated to the mid-range, low-end, integrated chipset, MB chipsets and stream computing and completely abandon the high end.

However, I currently don't see any indications that AMD/ATI will abandon the high end completely.

That said, if Nvidia will get off their arses and make a solid Vista 64 driver, I may still be getting an 8800 GTX if R600's features don't sufficiently impress me.

Regards,
SB

Jawed

26-Apr-2007, 20:53

4) Alternative explanation.
http://www.anandtech.com/storage/showdoc.aspx?i=2969
http://www.anandtech.com/storage/showdoc.aspx?i=2974

You be the judge.

Jawed

Geeforcer

26-Apr-2007, 20:55

Jawed, I swear I am not trying to be difficult, but what exactly am I looking for?

Sound_Card

26-Apr-2007, 20:55

Sound Card,

While I think it's never a good idea to take one source as gospel, Dailythech has shown to be fairly reliable in the past with their pre-release information. Since you are skeptical (which is not a bad thing by any measure), what do you suggest happed here?

1) They just made up the numbers. To me that seems unlikely - they are not the kind of site that would trade a short-term hits spike for a long term credibility gap.
2) They did the test to the best of their ability but something was wrong: Card, drivers, system in general, etc. This would be just reporting and if the numbers are off, oh well - they told us what they saw. Still, it would be in their best interests to make sure that doesn't happen because their credibility will be damaged.
3) The numbers are accurate.
4) Alternative explanation.

I think it's just best to wait and see. I'm not going to curse DT to hell, but I don't have a problem qustioning their data either. Komabatant is a person I trust, and that alone is enough to play skeptical. I see major potential flaws in their reviews and the numbers are not adding up.

With that said, I take option 4 for 4 hundred. :cool:

Geeforcer

26-Apr-2007, 20:57

With that said, I take option 4 for 4 hundred. :cool:

:wink: Well don't keep us in suspense - what do you think happened?

BlizzardOne

26-Apr-2007, 20:59

Maybe they had AA set to 24x or something in the control panel..:lol:

Jawed

26-Apr-2007, 21:00

Jawed, I swear I am not trying to be difficult, but what exactly am I looking for?
Incompetence.

Jawed

Sound_Card

26-Apr-2007, 21:07

:wink: Well don't keep us in suspense - what do you think happened?

I rest my case on the evidence that I have given.

On 1280x1024 they tested the HD 2900XT getting 47.9fps in Oblivion. However, their latest benchmarks put the HD 2900XT getting 73.4fps in a resoluton of 1920x1200. Same CPU and same game settings.

"I checked Tomshardware VGA chart for oblivion score. It reports that
Oblivion outdoor, 1920*1200*32, no AA, 8xAF, Max quality, HDRR manages 28 FPS

From Anh description I understand that the test was an outdoor.
While there are different outdoor areas where the performance will be different still I am not sure if the 98.4 FPS for the 8800 GTX is correct.

Looking at the numbers for 2900XTX, they seem to be very close to the 1950XTX which is difficult to believe.

For example:

Company of heroes 1280*1024:
1950 XTX 99 (tweaktown) E6600
2900 XTX 97 (dailytech) QX6800

Company of heroes 1600*1200:
1950 XTX 70 (tweaktown) E6600
2900 XTX 73 (dailytech) QX6800

http://www.tweaktown.com/articles/1084/6/page_6_be...

FEAR 1280*1024:
1950 XTX 80 (tomshardware) No softshadow 4AA 8AF , X6800
2900 XTX 84 (dailytech) with softshadow 4AA 16AF , QX6800

FEAR 1600*1200:
1950 XTX 57 (tomshardware) No softshadow 4AA 8AF , X6800
2900 XTX 58 (dailytech) with softshadow 4AA 16AF , QX6800

http://www.tomshardware.com/2007/02/12/the_amd_squ...

In Fear the softshadows can make a big difference but in company of heroes it seems the 2900 XTX has no advantage over 1950 XTX."'

Don't know what daily tech is thinking.

I have heard your side of the case, and rest assured, I'm not taking the bait. Slap me silly on May 2nd or the 14th, which ever the case.:razz:

It's a little bit of option 1 and 4. 4 because I highly trust kombatant. Gibbo's statment also alligns with what Kombatant staes.

trinibwoy

26-Apr-2007, 21:09

Maybe they had AA set to 24x or something in the control panel..:lol:

Won't that be a coup :lol:

Geeforcer

26-Apr-2007, 21:10

I rest my case on the evidence that I have given.

I have heard your side of the case, and rest assured, I'm not taking the bait. Slap me silly on May 2nd or the 14th, which ever the case.:razz:

Since you must have missed my previous post, allow me to copy/paste:

Let's see: The benchmarks you provided for comparison were CoH from TweakTown and FEAR from Tom's.

DTs CoH settings: High shader quality, High model quality, Anti-aliasing enabled (in game), Ultra texture quality, high quality shadows, high quality reflections, Post processing On, High building detail, High physics, high tree quality, High terrain detail, Ultra effects fidelity, Ultra effects density, Object scarring enabled and the model detail slider all the way to the right.

TweakTown CoH settings: ???

See a problem here?

ERK

26-Apr-2007, 21:22

You're going to need (almost) the whole memory controller, all of the ROPs including their requirements for DX10 (FP-Texture filtering, FP-Alpha-Blending, Compression etc.), INT-functions, massive amount of Registers (if you don't want to cripple your chip - DX10 requires 4.096 of them at any one time, though not necessarily all of them in hardware), and the whole lot of things i just plainly forgot.

Just take a look at the TC of the smaller G8x-Chips: You'll get 2 Shader-Quads with DX10 for the transistor-price of about 6 Quads DX9-Style. :)

Yes, your're right I suppose. And thanks nAo, for reminding me about the compression. I had forgotten about that--lots of benefits from that EDRAM!

About the G8x-chips, they do have lots more transistors per quad than the G7x-chips, but I didn't know if that was more related to D3D10 or to the scalar architecture. Very apples to oranges.
Thanks again, ERK

Sound_Card

26-Apr-2007, 21:31

http://www.anandtech.com/storage/showdoc.aspx?i=2969
http://www.anandtech.com/storage/showdoc.aspx?i=2974

You be the judge.

Jawed

This kind of stuff really is not all that un-common either.

BlizzardOne

26-Apr-2007, 21:50

not sure if this lil bit has been posted yet:

http://dailytech.com/Article.aspx?newsid=7044&commentid=131638#comments

&

http://images.dailytech.com/files/gpudip_results.html

no idea what it all means though...

**edit - quick comparison to one of my X1950 XTX's show the R600 to be yeilding results roughly 2.5 to 3x higher. will try and get some graphs or something whipped together :)

Arun

26-Apr-2007, 22:04

Well, well, well... I like 14 and 15 myself.
http://www.notforidiots.com/gpudip%20results.html

Geeforcer

26-Apr-2007, 22:08

Texture Image Units: 16
...

Jawed

26-Apr-2007, 22:14

Texture Image Units: 16
...
Ooh! Or should that be ARGH!

Jawed

vertex_shader

26-Apr-2007, 22:26

AMD CEO Hector Ruiz said in the CC last week "we don't do soft launches".

Not Hector Ruiz said this, Henri Richard said :wink:

Arnold Beckenbauer

26-Apr-2007, 22:42

Well, well, well... I like 14 and 15 myself.
http://www.notforidiots.com/gpudip%20results.html

My results are better:

[18:51:08] Folding@Home GPU Core - Beta
[18:51:08] Version 0.10 (Mon Oct 30 12:32:17 PST 2006)
[18:51:08]
[18:51:08] Compiler : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077 for 80x86
[18:51:08] Build host: CYGWIN_NT-5.1 vishal-gpu 1.5.19(0.150/4/2) 2006-01-20 13:28 i686 Cygwin
[18:51:08] Preparing to commence simulation
[18:51:08] - Looking at optimizations...
[18:51:08] - Files status OK
[18:51:08] - Expanded 87191 -> 443713 (decompressed 508.8 percent)
[18:51:08]
[18:51:08] Project: 2742 (Run 3, Clone 69, Gen 14)
[18:51:08]
[18:51:08] Assembly optimizations on if available.
[18:51:08] Entering M.D.
[18:51:14] Will resume from checkpoint file
[18:51:15] Working on 1258 p2742_GPUlambda
[18:51:15] Starting GUI Server
[18:51:21] Resuming from checkpoint
[18:51:21] Verified work/wudata_06.log
[18:51:21] Verified work/wudata_06.edr
[18:51:21] Verified work/wudata_06.trr
[18:51:21] Verified work/wudata_06.xtc
[18:51:21] Completed 80
[19:01:38] Completed 81
[19:12:55] Completed 82
[19:23:20] Completed 83
:lol:
But: 68530.1587? Too slow, other 8800GTX users get lower results.:?:

Xmas

26-Apr-2007, 22:45

Ooh! Or should that be ARGH!
That's the number of simultaneous textures allowed in the fragment shader. And it's obviously less than the hardware can do.

jamis

26-Apr-2007, 22:47

That's the number of simultaneous textures allowed in the fragment shader. And it's obviously less than the hardware can do.
So what's this about then?:?:

"Texture Image Units : 32"

http://www.notforidiots.com/gpudip%20results.html

overclocked_enthusiasm

26-Apr-2007, 23:00

Well you can't avoid the architecture to architecture comparisons. You seem to be implying that they can just set a low price on their fastest card and say "Hey look, we're fastest at $xxx". I think the price point of the XT is a lot more telling than any numbers from DT. If they thought it could compete favorably at a higher price point this move would boil down to initiating a price war. Exactly why would they do that?

This made me chuckle because of the recent Barcelona news:

"AMD also disclosed updated performance projections for its upcoming native Quad-Core AMD Opteron(TM) processors, code-named 'Barcelona.' The new Barcelona projections are based on the latest SPECcpu2006 benchmarks and show that AMD expects to have up to a 50 percent advantage in floating point performance and 20 percent in integer performance over the competition's highest-performing quad-core processor at the same frequency."

"Hey look, we're fastest at $xxx".
"at the same frequency"

Maybe that really IS what they are going to do! I guess it really doesn't matter that Nvidia has higher performing parts at a higher price point or Intel has higher clocks on their CPUs as long as you compare price point to price point and frequency to frequency.

ERK

26-Apr-2007, 23:15

After half a second's glance at the gpudip results (http://www.skenegroup.net/fornvidiots/gpudip/results/), the R600 looks very slow in that test.

IbaneZ

26-Apr-2007, 23:21

Bah! This thread is pathetic.

I miss Hellbinder, doomtrooper and WaltC. :wink: You're awake?

No launch without the big boys. :smile:

trinibwoy

26-Apr-2007, 23:22

After half a second's glance at the gpudip results (http://www.skenegroup.net/fornvidiots/gpudip/results/), the R600 looks very slow in that test.

Yeah Arun, what was that file you linked earlier? It's like 4x slower than other 8800GTX results?

tEd

26-Apr-2007, 23:28

I don't think you can compare them , as even the 1950xtx results are better than r600 comparing to the DT results:

http://www.skenegroup.net/fornvidiots/gpudip/results/ATI/R5xx/x1950xtx_xp_001.html

1950xtx gets 36880,4974

r600 gets 50450.6369 (DT results)

Geeforcer

26-Apr-2007, 23:35

Guys, those G80 gpudip results... how sure were we that they are not SLI?

BlizzardOne

26-Apr-2007, 23:36

BlizzardOne

26-Apr-2007, 23:37

Guys, those G80 gpudip results... how sure were we that they are not SLI?

We can't be, but if it treats SLI the same was as Crossfire, then it has bugger all effect - I already checked :)

trinibwoy

26-Apr-2007, 23:37

Don't think so - several people have reported in GTX numbers in the 17,000 range. Something is really screwy with those DT results. Maybe ATI threw in a sleep after every few lines of code in the leaked driver :)

tEd

26-Apr-2007, 23:38

My X1950 xtx gets nothing like that.. http://img123.imageshack.us/img123/8695/xtxmq1.jpg

that's why we shouldn't compare results from different sources :wink:

trinibwoy

26-Apr-2007, 23:39

My X1950 xtx gets nothing like that.. http://img123.imageshack.us/img123/8695/xtxmq1.jpg

Great, so basically these gpudip tests tell us nothing. :sad:

Arnold Beckenbauer

26-Apr-2007, 23:40

My X1950 xtx gets nothing like that.. http://img123.imageshack.us/img123/8695/xtxmq1.jpg

Vista?

Geeforcer

26-Apr-2007, 23:40

LOL, maybe we should start a gpudip thread in "boards and software" forum and compare results among ourselves.

BlizzardOne

26-Apr-2007, 23:41

Vista?

XP Pro, Cat 7.3's

*edit - just re-ran with CF enabled, but disabled my regular AA/AF settings, and total latency dropped to 81756.1576, while better that what it was, is a far cry from tEd's numbers.

Maybe it just guesses or randomises the results :p

Rangers

26-Apr-2007, 23:45

Texture Image Units: 16
...

What a real surprise/.

Rangers

26-Apr-2007, 23:45

Texture Image Units: 16
...

What a real surprise/.

8800GTX texture image units:32
http://www.notforidiots.com/gpudip%20results.html

16 for our hero

http://images.dailytech.com/files/gpudip_results.html

There's your problem right there.

That can explain why in many cases it's barely faster than X1950. Anytime that is the limiter..all you are gaining is 100mhz clock on 16 TMU's over R580. So you will gain 10 or 15% speed and that is all.

My first guess for substandard R600 results was to look at the shaders..what went wrong that the shaders were deficient? That was most likely wrong.

It's mindboggling that ATI did not even up the TMU's to 24 in a next gen part..while Nvidia has 32 to make sure they are not bottlenecked..

Arnold Beckenbauer

26-Apr-2007, 23:50

What a real surprise/.

What about this:
GL_RENDERER : ATI Radeon X1950 GT x86
GL_VENDOR : ATI Technologies Inc.
GL_VERSION : 2.0.6388 WinXP Release
Driver Name : atioglxx.dll
Driver Company : ATI Technologies Inc.
Driver Version : 6.14.10.6388
Driver Description : ATI OpenGL driver

ARB_fragment_program
ALU Instructions : 512 [ ABS ADD COS DP3 DP4 DPH DST EX2 FLR FRC LG2 LIT LRP MAD MAX MIN MOV MUL POW RCP RSQ SCS SIN SUB XPD ]
TEX Instructions : 512
TEX Indirections : 512
Program Instructions : 1536
Temporaries : 64
Parameters : 64
Local Parameters : 64
Env. Parameters : 64
Texture Coords : 8
Texture Image Units : 16
Texture Size : 4096 x 4096
Texture Rectangle Size : 4096 x 4096
Draw Buffers : 4
Renderbuffer Size : 4096 x 4096
Now surprised?

ERK

26-Apr-2007, 23:51

I think there are different versions of the program: 0.7, 0.8, released demo, etc. I don't see the version number listed with the results, except I think the DT results are v0.8.

EDIT: FYI: it was also written by a guy that does not own an ATI x1k graphics card(#31 (http://www.nvnews.net/vbulletin/showthread.php?t=87907&page=3)).

Rangers

26-Apr-2007, 23:53

What about this:

Now surprised?

Why? I knew R580 had 16.

So did X800 before it, years ago.

Arnold Beckenbauer

26-Apr-2007, 23:55

Why? I knew R580 had 16.

So did X800 before it, years ago.

And my X1950GT has 12. :wink:
GL_RENDERER : ATI Radeon X1950 GT x86
...
Texture Image Units : 16
Texture Size : 4096 x 4096
Texture Rectangle Size : 4096 x 4096
Draw Buffers : 4
Renderbuffer Size : 4096 x 4096

Rangers

26-Apr-2007, 23:57

And my X1950GT has 12. :wink:

So the program is innacurate? Fair enough.

Still odd it lists out 8800 at 32 though.

Arnold Beckenbauer

26-Apr-2007, 23:58

So the program is innacurate? Fair enough.

Still odd it lists out 8800 at 32 though.

No:
That's the number of simultaneous textures allowed in the fragment shader. And it's obviously less than the hardware can do.

jamis

26-Apr-2007, 23:59

And my X1950GT has 12. :wink:
And so has my Pro (well,GT actually but that's not the point..:lol: )

GL_RENDERER : Radeon X1950 Pro x86
GL_VENDOR : ATI Technologies Inc.
GL_VERSION : 2.0.6458 WinXP Release
---
Texture Coords : 8
Texture Image Units : 16
Texture Size : 4096 x 4096
Texture Rectangle Size : 4096 x 4096
Draw Buffers : 4
Renderbuffer Size : 4096 x 4096

Ailuros

26-Apr-2007, 23:59

Just a small note: I personally haven't been able to date to combine soft shadows with MSAA in FEAR. And before anyone is quick enough to state "I do" first watch the darned timedemo if there are actually softshadows present.

As for the DT results in general, it's too early to jump to any conclusions. That doesn't change though my personal opinion either that R600 will have only a modest performance advantage over it's direct competitor on average.

Rangers

27-Apr-2007, 00:01

One does have to wonder what it means for Xenos's actual performance if these benchmarks are real. Afterall it should be a no brainer that the XT is A LOT faster than Xenos.

Xenos has plenty of games to prove itself. It's a good chip. And it's games have looked just as good as it's competitor (that came out a year later), which is really all that counts in that sector.

It probably has 16 TMU's too.

INKster

27-Apr-2007, 00:06

Xenos has plenty of games to prove itself. It's a good chip. And it's games have looked just as good as it's competitor (that came out a year later), which is really all that counts in that sector.

It probably has 16 TMU's too.

Xenos' competitor didn't came out a year later, the game console carrying it did. ;)
My bet is that the RSX essential design was finished before the Xenos, since it is a much closer relative to the desktop GPU's available at that time.

R300King!

27-Apr-2007, 00:09

Well, here's what I think will happen. AMD will release the XT with newer stable drivers and it will beat the GTS. It might beat the GTX if you OC it to 845+ on the core. This is of course with the GTX factory OC'd to 650MHz. I heard the OC's GTX in the DT article is watercooled and costs $939.00. But no matter. I think the GTX Ultra will just be a 650MHz card with a larger cooler. It most likely won't OC much higher than that on an air cooler.

It seems there is a problem with the XTX, something probably about the GDDR4, etc. It could just be a driver bug on the XTX cuz it's hard to believe the XTX is the same speed as the XT with faster memory. Also, I've seen some benchmarks showing the X1950XTX beating the HD2900XT-X in a few games. This has got to be wrong.

I think DT made a mistake somewhere in benching. I don't think they just made up the numbers but they could have put the settings of the new XT/XTX cards at settings that take a big hit. Maybe there's some IQ settings that look great but just kill performance. Who knows?

Also, at the same time DT could have used some performance optimizations on the GTX(or maybe not) to boost its numbers. Many things could have happened. I think the XT's drivers are young and there are a few different versions floating around so who knows which one actually work best.

Many questions to answer. Yes DT is getting a sh*tload of hits on their website for these "reviews". I wouldn't put it past Fud and DT to throw number and rewiews back and forth for more traffic. And all DT has to do after the real reviews come out that contradict their "reviews" is to say, oops, our bad, we had some settings wrong, or the AA was at 24, we used the wrong drivers, etc, etc.

Gibbo over at Overclockers rated the XT at a 5/6 out of rating of 1-10 with the GTX being in the middle as a 5. Later Gibbo heard some other people that had XTs were getting some higher marks with newer drivers and so then Gibbo said he would probably rate it a 7 with these newer drivers compared to his numbers. (This is what I surmized, don't shoot me if not 100% right, but that's my assesment)

I guess we'll see May 2nd, right? ;)

Rangers

27-Apr-2007, 00:13

Xenos' competitor didn't came out a year later, the game console carrying it did. ;)
My bet is that the RSX essential design was finished before the Xenos, since it is a much closer relative to the desktop GPU's available at that time.

Well, of course I know that, but I dont feel like explaining every little thing in every post :smile:

They were more like same generation parts, yes. Xenos is still a kickbutt little chip. The OP just doesn't understand consoles at all by the very nature of his question :smile:

Xenos was very forward looking compared to RSX, and it sports 10MB of EDRAM in the same die space as the latter with none. It's a very excellent design. One could say ATI's last, really.

Jawed

27-Apr-2007, 00:16

Those GPUDIP results are not very useful, e.g. on G80 "18: Cel Shading" has 23 instructions/70 FLOPs and executes 6% slower than a 1 instruction/6 FLOP shader (MAD). In both cases, the shader has a single texture fetch.

Jawed

Nebula

27-Apr-2007, 00:42

Just a small note: I personally haven't been able to date to combine soft shadows with MSAA in FEAR. And before anyone is quick enough to state "I do" first watch the darned timedemo if there are actually softshadows present.

There is a trick to get it to run with AA and SS on, screenshot included.

http://forum.beyond3d.com/showpost.php?p=954174&postcount=207

hoom

27-Apr-2007, 00:53

Am I just misreading or (aside from questions of accuracy/comparibility) are people saying that those Gpudip results indicate poor performance for R600?

I read most of those results as being about twice as fast as the G80 O_o

Ailuros

27-Apr-2007, 00:56

There is a trick to get it to run with AA and SS on, screenshot included.

http://forum.beyond3d.com/showpost.php?p=954174&postcount=207

I might have neglected to mention that I meant the timedemo, but I thought it was obvious since the debate was about benchmark results.

FEAR was never meant to combine Soft shadows and MSAA; in game it does lead to occassional side-effects.

Nebula

27-Apr-2007, 01:05

I might have neglected to mention that I meant the timedemo, but I thought it was obvious since the debate was about benchmark results.

I know you meant the benchmark, thats why I linked the trick to you. Yes it works in the FEAR inbuilt benchmark and it is very obvious (just ran the inbuilt benchmark for memories sake and it's visible).

FEAR was never meant to combine Soft shadows and MSAA; in game it does lead to occassional side-effects

True.

flippin_waffles

27-Apr-2007, 01:30

Where is this score on GPUDIP that people are talking about? That number at the bottom of the page? Isn't that the column under processing time totalled up? I never added them, but I thought for sure that was total time taken. I'm at about 168 secs or so with my x800 pro modded.

Rangers

27-Apr-2007, 01:30

That's the number of simultaneous textures allowed in the fragment shader. And it's obviously less than the hardware can do.

Wouldn't it still be a bottleneck the same though? Just moved up the chain a bit?

BTW, if it's obviously less than the hardware can do, why is it more than the X1950GT can do?

I do hope ATI put at least 24 in there though.

Anon Lamer

27-Apr-2007, 01:36

Considering that ALUs don't consume much of the die anyway, the picture's even more puzzling. So, ahem, D3D10 "done right" :lol: may well cost 400M transistors :lol: :lol:

I suspect it may be GPGPU done right that cost an unknown amount of transistors.

WRT the XTX, I dont understand why they dont just respin the card with +100 Mhz over the XT (DTs overclock went to about 850 Mhz) and 1 gig of 2 ghz GDDR3 memory. The G80 ultra has 2.16 Ghz GDDR3 ram, thats not far from 2.2 Ghz GDDR4 ram. It may be that the XTX has to suffer the same GPU speed as the XT in order to conserve power consumption. As usual, we wont know until more technologically skilled review sites get their hands on samples.

Jawed

27-Apr-2007, 01:45

turtle

27-Apr-2007, 01:56

http://dailytech.com/DailyTech+Digest+Making+Sense+of+R600/article7079.htm
I won't pretend that the tests we did were exhaustive. Since DailyTech does not sign embargos, we sort of get what we can take when it comes to publishing early tests. Scott Wasson, Ryszard Sommefeldt and Anand Shimpi are true experts when it comes to GPU performance, and if you have any reservations about our preliminary examinations, certainly wait for their benchmarks.

Rys, where are you? Take a bow. :)

Ailuros

27-Apr-2007, 02:02

Thumbs up for the above statement.

3vi1

27-Apr-2007, 02:13

I suspect what DT tested was one of the first OEM spins of the XT - remember the threads from February, just before Chinese new year?

Essentially, they [DT] are pumping their site with sensationalism just to get those banner impressions. Those numbers don't add up that they published. Again, I think the XTX they said they tested was really just an OEM XT.

Anyhow, hello B3D!

PS: GPGPU, AVIVO and sound proccessing/physics... That is where those extra trannies went.. i bet!

Arty

27-Apr-2007, 03:12

Some more (http://www.chiphell.com/viewthread.php?tid=3336&extra=page%3D1) pics

http://img249.imageshack.us/img249/7600/010vl1.jpg

http://img249.imageshack.us/img249/1975/012dp3.jpg

http://img175.imageshack.us/img175/9567/011nh7.jpg

Cuthalu

27-Apr-2007, 03:18

Why is there two connectors for the fan, when it uses only one of them?

SugarCoat

27-Apr-2007, 03:18

Is that an 1024mb XTX? Correct me if im wrong but there isnt suppose to be memory on the back of the card if its 512mb.

Sobek

27-Apr-2007, 03:39

Why is there two connectors for the fan, when it uses only one of them?

Redundancy? Who knows...

This x800xt in my office system here has dual fan headers when it only uses one too.

INKster

27-Apr-2007, 03:51

Anarchist4000

27-Apr-2007, 04:04

Why does the copper inside that thing look rather small? It looks almost completely hollow. I could have sworn it was a solid block of copper before.

Also I wouldn't think they would need to expend very many transistors for GPGPU work. R580 seems to work very well at it and other than shared cache I can't think of anything else specific that would be required. And the sound processing should more or less be a pass-through setup so I wouldn't expect very many transistors to be spent there.

Geo

27-Apr-2007, 04:09

http://dailytech.com/DailyTech+Digest+Making+Sense+of+R600/article7079.htm

Rys, where are you? Take a bow. :)

:lol: He'll hate feeling a momentary instinct of affection for DT. :razz:

INKster

27-Apr-2007, 05:32

Did anyone notice how first the cooler and now the PCB look eerily similar to a 8800 GTX ?

Sobek

27-Apr-2007, 05:37

Did anyone notice how first the cooler and now the PCB look eerily similar to a 8800 GTX ?

Yeah I noticed that too...it's like the cooler's been modified to mimic the GTX's cooler. There's SFA copper under there and yet they insist on using a massive plastic casing extremely reminiscent of the shape and design that Nvidia's currently using. I suppose one benefit is the additional space allows for a tad more airflow, but I agree, the similarities are there.

SugarCoat

27-Apr-2007, 06:10

thats essentially the same cooler design they used on the X1950 reference cards...

Geeforcer

27-Apr-2007, 06:11

According to the latest blurb on DailyTech, Crossfire seems to be working (I remember CJ previously mentioning that there were some issues with it).

Sobek

27-Apr-2007, 06:22

thats essentially the same cooler design they used on the X1950 reference cards...

R600's cooler looks larger and more squared off to me...

http://www.3dnews.ru/documents/13027/X1950XTX-cooler-big.jpg

mao5

27-Apr-2007, 07:22

Negative:

- Kristopher Kubicki (DailyTech Staff)

BTW, welcome ! :smile:

I think they used early GDDR4 XTX sample card, remember this? 07/04 on the chip.

http://bbs.souyo.com/uploadimage/igdbol_2007427_9791.jpg
http://bbs.souyo.com/uploadimage/igdbol_2007427_7295.jpg

erick

27-Apr-2007, 07:33

EDIT: Nevermind. :)

Kaotik

27-Apr-2007, 07:34

Interesting, I thought R600 chip was turned by 45 degrees, but the footprint of the chip in the thermal paste would indicate it is not turned at all...

That's X1950 cooler, not R600 cooler :wink:

mao5

27-Apr-2007, 07:36

Kaotik

27-Apr-2007, 07:46

What's up with 2 fan power-plugs in on the PCB btw? :???:
In the pic with cooler attached, it only fills 1.

Sobek

27-Apr-2007, 08:02

What's up with 2 fan power-plugs in on the PCB btw? :???:
In the pic with cooler attached, it only fills 1.

-> Why is there two connectors for the fan, when it uses only one of them?

->

Redundancy? Who knows...

This x800xt in my office system here has dual fan headers when it only uses one too.

And it never went past that :razz:

HAL

27-Apr-2007, 08:03

http://www.dailytech.com/DailyTech+Digest+Making+Sense+of+R600/article7079.htm
:wink:

R300King!

27-Apr-2007, 08:28

Someone (Bum_JCRules) on Tomshardware (http://forumz.tomshardware.com/hardware/HD2900-XTX-Benchftopic-234334-days0-orderasc-75.html) stated he was under NDA and said these few quotes.

While I am required to follow the NDA, the stuff up on Daily Tech today is almost worthless. Yes Anandtech was present in Tunisia (signing Non-disclosure agreements like the Inquirer), why they are posting this stuff is beyond me because their numbers are off. They must be only using the XP drivers and OS because the numbers in CF vs the GTX are very much different. So until I can officially comment on the architecture and the performance.. hold all of this as useless until the rest of the world writes about it.

I really would love to comment on this stuff...

I understand that DT and Anand are seperate but that is so childish. Derek was there and his cards got to his place of business before he returned home from Tunisia. That long board they have ... Not what Derek should have gotten in his delivery. That is all I will say before I go too far.

So, who knows? These DT benches could be misleading.

_xxx_

27-Apr-2007, 08:31

I don't completely understand why people are saying that R600 in general is a complete failure. Maybe the XTX is a failure if it doesn't outperform the GTX. However, the 512MB XT model seems to perform as good if not better than the 640MB GTS, for a similar price point.

It is a failure, because releasing it in that price range (which I still can't believe) will cost ATI/AMD an arm and leg and pull their earnings further down. Let alone the investments it ate already throughout the development.

_xxx_

27-Apr-2007, 08:35

Why is it that nobody else has this demo?

Because they're not as imaginative, since it doesn't exist.

_xxx_

27-Apr-2007, 08:42

Space, no - but it can drive the transistor count up pretty hard.

There is no RAM on the GFX-chip.

That aside, I'd rather think it's the 512-bit bus, the ROP's, the video part and so on. The ALU's are surely a bit more complex as well.

epicstruggle

27-Apr-2007, 09:07

Someone (Bum_JCRules) on Tomshardware (http://forumz.tomshardware.com/hardware/HD2900-XTX-Benchftopic-234334-days0-orderasc-75.html) stated he was under NDA and said these few quotes.

So, who knows? These DT benches could be misleading.
Good find, this whole release has become more interesting as the days go by. Cant wait for the official release.

silent_guy

27-Apr-2007, 09:14

There is no RAM on the GFX-chip.
I'm pretty sure they were not talking about DRAM, but regular SRAM (L1/L2/fifos/...) :wink:

_xxx_

27-Apr-2007, 09:27

I'm pretty sure they were not talking about DRAM, but regular SRAM (L1/L2/fifos/...) :wink:

Well that's not how I read it, but I guess you're right :)

neliz

27-Apr-2007, 09:32

Everytime I think about DailyTech, I think about Sander Sassen and his HardwareAnalysis (http://www.hardwareanalysis.com/content/reviews/article/1813.6//) R520 review. When push came to shove it just wasn't that bad as everyone expected (http://www.hexus.net/content/item.php?item=3603&page=5) because of all the negative fud that was popping up all over the web

satein

27-Apr-2007, 09:44

Someone (Bum_JCRules) on Tomshardware (http://forumz.tomshardware.com/hardware/HD2900-XTX-Benchftopic-234334-days0-orderasc-75.html) stated he was under NDA and said these few quotes.

So, who knows? These DT benches could be misleading.

Kudo!! Nice find.

That would help explaining a bit on what it went during this few days. I still insist I will wait until I could read on B3D analysis on this Doggy-Delay-R600 :twisted:. It might be a twist at the end.

Regards,

dizietsma

27-Apr-2007, 09:49

I get the feeling that the 2900XT and XTX will be good cards and shine in the areas that ATi normally shine in however will that be enough ?

I think that releasing your card later than the opposition means you are on a hiding to nothing unless you can pull out a big performance jump over the current best of breed and that is not likely to happen this time.

Looking at history one unwanted, but positive, side effect of the FX 5800U for nvidia was that at with nv40 they were starting from a low base with not great expectations and so when nv40 came along in the same time frame and similar performance to the ATi offering it was seen as a big improvement from them even though really only the 6600GT was only a fair way above equivalent ATi card.

Since then nvidia seem to be always getting the first shot in with the impressive 40-60% gains over the last generation. To get these gains and the kudos you need to be first out the door with the next generation. Something ATi has failed twice with now. For G80 I was just expecting a 40% faster card, which we more than got, but we also got some interesting architecture to think about and also some better IQ as well ( gasp! ) so really they wowed the audience. With the new R6xx series cards they will be good but that wow factor will be missing again.

AMD really do need to get the next gen card out first again to regain the upper hand in this respect.

IbaneZ

27-Apr-2007, 09:53

Yeah, hopefully Mr Bum_JCRules knows something we (and DT) don't.

Anyway, I need a new PSU to power the 2900XT, my 480W won't be enough. Will this booster thingy do the job? http://www.overclockers.co.uk/showproduct.php?prodid=CA-008-SK&groupid=701&catid=123&subcat=484

It has 2 PCI express outputs. The price is right. :smile:

satein

27-Apr-2007, 10:10

Unknown Soldier

27-Apr-2007, 10:17

lol IbaneZ, I too currently have a 480W PSU and almost ordered a OCZ GameXStream 700W Power Supply before realising that it could be an incorrect PSU to buy for the R600 so cancelled my order.

Will that booster have the correct connectors for the R600 and will it have enough connectors for X-fire R600?

US

Reputator

27-Apr-2007, 10:39

Is that an 1024mb XTX? Correct me if im wrong but there isnt suppose to be memory on the back of the card if its 512mb.512-bit with only 8 channels...?

IbaneZ

27-Apr-2007, 10:39

Will that booster have the correct connectors for the R600 and will it have enough connectors for X-fire R600?

US

Apparently the 2900 has one 6 pin and one 8 pin, but the card should run just fine with two 6 pins. This is just based on rumors though, I have no idea. :smile:

Two R600's in X-fire will probably kill that little booster baby. :lol:

neliz

27-Apr-2007, 11:14

Apparently the 2900 has one 6 pin and one 8 pin, but the card should run just fine with two 6 pins. This is just based on rumors though, I have no idea. :smile:

Two R600's in X-fire will probably kill that little booster baby. :lol:

Fudo keeps saying that the 8-ping is only if you want to use overdrive..

pjbliverpool

27-Apr-2007, 11:28

They were more like same generation parts, yes. Xenos is still a kickbutt little chip. The OP just doesn't understand consoles at all by the very nature of his question :smile:

Sorry, but I believe I do, in fact I think you just misinterpreted what I was saying. I know this is off topic and once we have real benchmarks I may start a new thread about it if the point is still relevant but all I was saying is that if the XT, which is clearly much faster than Xenos no matter what is scoring similar to R580 then it has major implications for the old R520/R580 vs Xenos question which in turn has implications for RSX vs Xenos.

Of course its way to early to draw those kinds of conlusions so lets just leave it there until we know more :wink:

Twinkie

27-Apr-2007, 11:31

from THG:

why they are posting this stuff is beyond me because their numbers are off. They must be only using the XP drivers and OS because the numbers in CF vs the GTX are very much different.

The second sentence doesnt actually explain why the numbers are low on DTs review, but it brings other questions to the table.

Firstly, DT didnt even mention about benching crossfire/SLi, but notice how he mentions about the numbers for the crossfire/SLi performance?

Secondly, XP drivers and OS? what OS? so they benching in Vista where in fact the XT is alot closer to the GTX?

and just maybe.. maybe DT didnt sign the NDA and got hold of the card somehow.. *cough* AT * cough*.

leoneazzurro

27-Apr-2007, 11:41

DT surely hasn't got their card from ATI (because they should have signed the NDA in this case) but if they received their cards from Anandtech then Anand is in trouble: I don't think the NDA they signed allows the card to be given to benchmark to an external source.
My guess is that they got cards from another source (vendor, maybe).

Xmas

27-Apr-2007, 12:20

Wouldn't it still be a bottleneck the same though? Just moved up the chain a bit?

BTW, if it's obviously less than the hardware can do, why is it more than the X1950GT can do?

I do hope ATI put at least 24 in there though.
Maybe my wording wasn't clear: This is not the number of TMUs. It's the number of texture objects you can use in a fragment shader.

Actually, I was wrong in saying the hardware is certainly capable of more. D3D10 requires 128 simultaneous "resources" but only 16 samplers (which I thought to be 32). And since OpenGL texture objects combine both texture image and sampler state, R600 could indeed be limited to 16.

Kaotik

27-Apr-2007, 12:47

trinibwoy

27-Apr-2007, 12:56

The second sentence doesnt actually explain why the numbers are low on DTs review, but it brings other questions to the table.

Firstly, DT didnt even mention about benching crossfire/SLi, but notice how he mentions about the numbers for the crossfire/SLi performance?

Well he didn't mention SLI. Maybe he means that his crossfire numbers are more than twice DT's single XT numbers - which would lead one to believe that DT's numbers are bogus.

leoneazzurro

27-Apr-2007, 13:12

If they're from vendor, why do they have wrong clocks? Or do you mean they'd have some early samples from vendor rather than retail models?

Yes, it could be a vendor sample, I don't know how much early.

Sound_Card

27-Apr-2007, 13:19

Sound_Card

27-Apr-2007, 13:28

Unknown Soldier

27-Apr-2007, 13:38

So is that 2nd May release date for the R600 officially official? :)

US

_xxx_

27-Apr-2007, 13:41

No.

satein

27-Apr-2007, 13:59

No.

I cannot believe you because if you are under 'NDA' you cannot say 'No' :grin:

Hopefully, it will be soonnnnn....

neliz

27-Apr-2007, 14:03

I cannot believe you because if you are under 'NDA' you cannot say 'No' :grin:

Hopefully, it will be soonnnnn....

No retailer has set a date before May 14th. I expect retail availability to be after the 14th... more like the end of the month...

bangalter

27-Apr-2007, 14:13

No retailer has set a date before May 14th. I expect retail availability to be after the 14th... more like the end of the month...

But we might see official reviews by may 2nd, right?

I want to read Beyond3d's review, pretty much the only site I trust.

trinibwoy

27-Apr-2007, 14:36

Ok, so before this goes totally out of hand, let me say this, and this will be my final say on the matter, until the NDA is lifted: AMD made certain decisions concerning this card. I took a hard look out there, to see what's being leaked, and it seems there are still some stuff that are totally made up - tbh, it smells like a FUD campaign to me, if I take into consideration certain emails that are flying around lately. Certainly there are some stuff out there that are true, and you will know which is which when the NDA lifts soonish. The journalists that were in Tunis certainly know, and are probably laughing at some of those at this minute.

Wow that is seriously cryptic. Komb seems to be highlighting the fact that some things are true and some are false but he's not hinting at what type of info is accurate. We have rumours about performance, clocks, configuration, price and release dates. Anybody have a clue which of those he's referring to?

Also, why the emphasis on the architectural decisions that ATI made? Do these decisions impact IQ, performance, scaling? Is it an excuse for lackluster performance in some scenarios or is it something more revolutionary?

_xxx_

27-Apr-2007, 14:39

I cannot believe you because if you are under 'NDA' you cannot say 'No' :grin:

Hopefully, it will be soonnnnn....

I'm not under NDA, just saying that it's not confirmed anywhere in public, thus only a rumour so far :)

Arty

27-Apr-2007, 14:49

Wow that is seriously cryptic. Komb seems to be highlighting the fact that some things are true and some are false but he's not hinting at what type of info is accurate. We have rumours about performance, clocks, configuration, price and release dates. Anybody have a clue which of those he's referring to?

Also, why the emphasis on the architectural decisions that ATI made? Do these decisions impact IQ, performance, scaling? Is it an excuse for lackluster performance in some scenarios or is it something more revolutionary?
My guess? Mem Clocks, Shaders & Price.

Also note that he said AMD made decisions (delay->pricing) regarding the card doesnt specifically say 'architectural' decisions.

vertex_shader

27-Apr-2007, 14:52

Anybody have a clue which of those he's referring to?

Last part of his message say more than 1000 word.

IbaneZ

27-Apr-2007, 15:08

If kombatant is lying I'm gonna jump on the next plane to Greece and tickle him with my pen0r.

neliz

27-Apr-2007, 15:10

If kombatant is lying I'm gonna jump on the next plane to Greece and tickle him with my pen0r.

If that's the best you can do with it ;)

He'd better do €390 worth of pleasuring

nicolasb

27-Apr-2007, 15:12

No retailer has set a date before May 14th. I expect retail availability to be after the 14th... more like the end of the month...If you believe the current rumours, May 2nd is a paper launch, with actual retail availability in mid-May.

http://www.theinquirer.net/default.aspx?article=39223

Sound_Card

27-Apr-2007, 15:13

If kombatant is lying I'm gonna jump on the next plane to Greece and tickle him with my pen0r.

I know your trying to be funny, but man, that sounds.... well I don't want to get into that.

Sound_Card

27-Apr-2007, 15:15

From what I understand, NDA lifts May 2nd, retail channel opens May 14th.

IbaneZ

27-Apr-2007, 15:21

From what I understand, NDA lifts May 2nd, retail channel opens May 14th.

Yeah, lets hope so.

We're dying here! :yelling:

Sound_Card

27-Apr-2007, 15:21

http://www.theinquirer.net/default.aspx?article=39223

We have now confirmed with DAAMIT that launch and retail availability will be on the 14th, one day ahead of 8800 Ultra's arrival in retail, following a paper launch on the 2nd. This paper launch was also the cause for the ATI R600 rumour to appear.

Intresting....

Keep in mind that this must be one of the first times the inq has actully signed a NDA.

Kaotik

27-Apr-2007, 15:26

http://www.theinquirer.net/default.aspx?article=39223

Intresting....

Keep in mind that this must be one of the first times the inq has actully signed a NDA.

World would be better place without the Inquirer.
NDA ending has nothing to do with "launching a card", it usually is the same day, but having it ending 2 weeks early doesn't mean it would be "paperlaunched 2 weeks early"

neliz

27-Apr-2007, 15:28

I only follow the hard data that I have and that is that my card will be delivered somewhere between the second and fourth week of May.
http://home.wanadoo.nl/deachterberg/order.PNG

Sound_Card

27-Apr-2007, 15:29

World would be better place without the Inquirer.
NDA ending has nothing to do with "launching a card", it usually is the same day, but having it ending 2 weeks early doesn't mean it would be "paperlaunched 2 weeks early"

Well, were expected to see waves of benchmarks on that date. So I would call that a paper launch sort of speak.

Arty

27-Apr-2007, 15:35

Anyone think its worth speculating on why the NDA expiry was preponed two weeks? That is a pretty big move IMO.

1. Reaction/response from press in Tunisia
2. Leaks/FUD starting to get out of control
3. AIB Partners getting restless after months of delay
4. Launch of 8800 Ultra
5. Something else

Hopefully we will get atleast a few good (detailed) reviews on launch.

HAL

27-Apr-2007, 15:47

http://www.tweaktown.com/news/7397/radeon_hd_2900xt_launch_and_for_sale_may_14_not_th e_2nd/index.html
:wink:

satein

27-Apr-2007, 15:49

Well, were expected to see waves of benchmarks on that date. So I would call that a paper launch sort of speak.

Well, if the benchs are supposed to appear next week, it would be the weekend nightmare of all editior :wink: Returning from Tunisia and benching a new card and analysis a new architectue. I could only imagine on how the editor will want to slap DT for this thing as at first they are supposed to get 3 weeks long benching the thing.

Next week will be interesting if it is ture for early lift of NDA, but the following week would be the real heat up.

Silent_Buddha

27-Apr-2007, 15:55

Wooo, this makes me a truly happy camper if NDA actually does lift on the 2nd. The return of review weeks before actual availability. Meaning I get a chance to mull over reviews and whatnot before plunking down my cash and being one fo the first to own a card.

/happy dance.

It's nice to see both ATI and NV going back to doing this. Hopefully it's not a one shot thing.

And thank god NDA lifts next week, with so many conflicting reports, rumors and whatnot floating around, it'll be nice to get something solid.

So, will B3D be releasing their analysis on R600 when NDA lifts? :)

Regards,
SB

vertex_shader

27-Apr-2007, 16:09

http://www.tweaktown.com/news/7397/radeon_hd_2900xt_launch_and_for_sale_may_14_not_th e_2nd/index.html
:wink:

This saga never ends :wink:

From what we’ve been hearing in Taipei though, ATI are working busier than ever on new driver revisions which MIGHT increase performance but of course nobody knows what will happen for sure. One company claimed that ATI are waiting for “the best driver” which means if the driver is still not 100% by May 14th, we could see further delays.

When the driver 100% for may14th, than what? reviewers already making the reviews, sounds like another "great" information :smile:

mao5

27-Apr-2007, 17:05

one slide of $399 Radeon HD 2900XT confirms 16 texture units.

http://r800.blogspot.com/2007/04/radeon-hd-2900xt.html

XMAN26

27-Apr-2007, 17:16

Someone (Bum_JCRules) on Tomshardware (http://forumz.tomshardware.com/hardware/HD2900-XTX-Benchftopic-234334-days0-orderasc-75.html) stated he was under NDA and said these few quotes.

So, who knows? These DT benches could be misleading.

you should read what he said again. "the numbers in CF vs the GTX are very much different" With that he is talking about CrossFire mode vs a single GTX where as DT is 1 XTX vs 1 GTX.

Jawed

27-Apr-2007, 17:25

one slide of $399 Radeon HD 2900XT confirms 16 texture units.

http://r800.blogspot.com/2007/04/radeon-hd-2900xt.html
It also appears to confirm 16 ROPs.

So we get basic stuff based on a 742MHz clock (basing that on the rate of 742M triangles/s):

106 GB/s
bilinear rate 11872MT/s
pixel fillrate 11872MP/s
475 GFLOPsAnd, erm, that's about it for now folks...

700M transistors is seemingly a bit less than we were previously expecting (720-750M I think).

I'm not sure what the 47.5GP/s rate is meant to be? Fragment rate seems likely, 64*742.

Jawed

nAo

27-Apr-2007, 17:31

320 scalar madds, wow!
fragment rate is probably computed as the shaded fragment count per clock running a single clock shader

trinibwoy

27-Apr-2007, 17:40

I know ATI likes better pixels over more pixels but those unit counts lead me to believe they were aiming for higher clocks. Even if those TMU's are full speed on FP16 12GT/s still seems like a low target.

nao, we know they've adopted Nvidia's "stream processor" marketing but have we heard anything to the effect that they're scalar?

nAo

27-Apr-2007, 17:42

nao, we know they've adopted Nvidia's "stream processor" marketing but have we heard anything to the effect that they're scalar?
I have faith ;)

trinibwoy

27-Apr-2007, 17:44

Heh, well yeah there's always that :smile:

Anybody think that these 16 TMU's are a global sampler array? If they are they should get better utilization than G80's dedicated TMU's per cluster. (I'm sure there's some implication for cache structure and usage there but since I don't know anything about that stuff I wont try to comment).

Jawed

27-Apr-2007, 18:06

Anybody think that these 16 TMU's are a global sampler array? If they are they should get better utilization than G80's dedicated TMU's per cluster. (I'm sure there's some implication for cache structure and usage there but since I don't know anything about that stuff I wont try to comment).
I'd guess that the TMUs will retain their independence, as per R3xx...R5xx.

But your thoughts on cache are very apt:

Virtual memory fragment aware cache (http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=US2007073996&F=0)

http://www.cupidity.f9.co.uk/b3d94.gif

and it would appear that each shader unit (containing 4 TMUs) has a dedicated L2 cache. Then, the L2 cache for each shader unit can act as an L3 cache for other shader units.

I'm guessing that if an L2 miss occurs, L2 can query the 3 other L2s (functioning as L3) and if there's success, the cache line will be copied into L2. Otherwise a fetch from memory will be requested.

Remember that G80 has one L2 cache per memory channel. So, in effect, the L2 cache architecture is shared across all TMUs equally.

So, ahem, similar but different. R600 might be ATI's first GPU with an L2 and they went and put in L3 at the same time it seems.

Jawed

caboosemoose

27-Apr-2007, 18:06

Repost. Sorry.

jimmyjames123

27-Apr-2007, 18:11

From the cryptic messages from Kombatant, and looking at comments about Crossfire vs GTX, 24xAA, improved crossfire solution for R600, better image quality on the R600, no XTX model will be released, you will be suprised by the performance compared to 8800, etc, I can only come to the following conclusion:

AMD will launch the following to compete against 8800 GTS and 8800 GTX:

~$399 HD 2900 XT 512MB RAM
~$699 CROSSFIRE HD 2900 XT 1024MB RAM (basically two the above in crossfire, at $50 discount each)

The 2900 XT 512MB should in general outperform the 8800 GTS, and the crossfire solution with 1024MB RAM in total should outperform the 8800 GTX/Ultra (while providing better image quality through Crossfire).

This isn't the greatest way to achieve high margins, but it's a good way to gain the price-to-performance crown, especially if their recent implementation of Crossfire is more efficient than SLI.

SugarCoat

27-Apr-2007, 18:19

http://forum.beyond3d.com/showpost.php?p=976886&postcount=3083

thats an XTX right there with the full gig of ram, nobody seemed to care though. :lol:

jimmyjames123

27-Apr-2007, 18:21

I am not referring to a 1024MB single card higher clocked XTX, but rather two 512MB XT Crossfire cards positioned to fight the GTX/Ultra. Clearly, the Crossfire would outperform the GTX/Ultra, with better IQ too.

Sound_Card

27-Apr-2007, 18:24

http://forum.beyond3d.com/showpost.php?p=976886&postcount=3083

thats an XTX right there with the full gig of ram, nobody seemed to care though. :lol:

aye.

Take a good look at DT's XTX and notice where the power connectors are. Then take a good look at this XTX sugar provided.

SugarCoat

27-Apr-2007, 18:25

I am not referring to a 1028MB single card higher clocked XTX, but rather two 512MB XT Crossfire cards positioned to fight the GTX/Ultra. Clearly, the Crossfire would outperform the GTX/Ultra, with better IQ too.

yea but again though you and a few others are swinging the conversation over to how great it is via price wars, this isnt exactly what i and quite a few others care about, we want to know who has the better architecture. What exactly makes up the R600 and so on. I dont really care if they do a 2 for 1 or give cards away for free for that matter, it doesnt mean the cards themselves are better, which is what i and i believe quite a few others are interested in finding out. If waiting an additional 6 months only got AMD/ATi to do some strong pricing moves and the performance itself is lackluster, well then thats pretty boring.

Mize

27-Apr-2007, 18:30

Geeforcer

27-Apr-2007, 18:34

http://forum.beyond3d.com/showpost.php?p=976886&postcount=3083

thats an XTX right there with the full gig of ram, nobody seemed to care though. :lol:

That's not what the sorce for those pictures says: Radeon HD 2900 XT显卡的最终工作频率定在740/825MHz（核心/显存）。

turtle

27-Apr-2007, 18:37

If waiting an additional 6 months only got AMD/ATi to do some strong pricing moves and the performance itself is lackluster, well then thats pretty boring.

True, but you know...Occam's razor and all that.

Jawed

27-Apr-2007, 18:42

That's not what the sorce for those pictures says: Radeon HD 2900 XT显卡的最终工作频率定在740/825MHz（核心/显存）。
Yeah, a 512-bit bus with 32-bit devices always requires 16 chips.

Jawed

Kaotik

27-Apr-2007, 18:42

http://forum.beyond3d.com/showpost.php?p=976886&postcount=3083

thats an XTX right there with the full gig of ram, nobody seemed to care though. :lol:

That's XT, having 16 memchips is required for their 512bit membus anyway.

Unknown Soldier

27-Apr-2007, 18:42

That's not what the sorce for those pictures says: Radeon HD 2900 XT显卡的最终工作频率定在740/825MHz（核心/显存）。

Ye but that card looks bigger than 9.5" and you can clearly see the extra memory on the back(1024MB) so it's not an XT.

US

Geeforcer

27-Apr-2007, 18:45

Ye but that card looks bigger than 9.5" and you can clearly see the extra memory on the back(1024MB) so it's not an XT.

US

EDIT: LOL, both Jawed AND Kaotic beat me to it...:grin:

turtle

27-Apr-2007, 18:46

Yeah, a 512-bit bus with 32-bit devices always requires 16 chips.

Jawed

I thought it was 8x64?

Silent_Buddha

27-Apr-2007, 18:46

you should read what he said again. "the numbers in CF vs the GTX are very much different" With that he is talking about CrossFire mode vs a single GTX where as DT is 1 XTX vs 1 GTX.

Hmmmm, I wonder if this is him hinting at a 7950 GX2-ish setup for R600. A dual GPU 2xR600 single card (2 boards ala 7950 GX2?) setup to position against the 8800 GTX?

Personally I don't like the thought of this, but that quote sure does hint at it rather strongly IMO.

Regards,
SB

turtle

27-Apr-2007, 18:48

Hmmmm, I wonder if this is him hinting at a 7950 GX2-ish setup for R600. A dual GPU 2xR600 single card (2 boards ala 7950 GX2?) setup to position against the 8800 GTX?

Personally I don't like the thought of this, but that quote sure does hint at it rather strongly IMO.

Regards,
SB

My power supply just pissed it's self.

Geeforcer

27-Apr-2007, 18:50

I think powering and cooling a board with 2 R600s and 32 RAM chips would be quite a challenge. With design costs thrown in, it would probably be easier and cheaper to just go with 2 cards.

jimmyjames123

27-Apr-2007, 18:52

yea but again though you and a few others are swinging the conversation over to how great it is via price wars, this isnt exactly what i and quite a few others care about, we want to know who has the better architecture. What exactly makes up the R600 and so on. I dont really care if they do a 2 for 1 or give cards away for free for that matter, it doesnt mean the cards themselves are better, which is what i and i believe quite a few others are interested in finding out. If waiting an additional 6 months only got AMD/ATi to do some strong pricing moves and the performance itself is lackluster, well then thats pretty boring.

I never said anything about R600 being a "better" architecture than G80. How could one know that without having the cards on hand for extensive testing? I'm sure that both architectures have strong points and weak points in comparison to one another.

Being able to offer a high performance 1024MB total RAM crossfire solution (where NVIDIA right now can only offer 1280MB or 1536MB total RAM on GTS and GTX) is certainly a notable price-to-performance advantage for ATI in terms of cost of RAM, and that is tied to the fact that their card also uses a 512 bit bus. I said weeks ago that NVIDIA would need a 512 bit bus in order to better compete on a cost basis (in terms of on-board RAM cost).

This is a rumors and speculation thread on R600. I would think it would be relevant to talk about what R600 offerings they would position against NVIDIA. It's alot easier to speculate about that vs speculating on the architecture of a card that very few have had a chance to test (and those people are under NDA).

Kombatant

27-Apr-2007, 18:52

That's not what the sorce for those pictures says: Radeon HD 2900 XT显卡的最终工作频率定在740/825MHz（核心/显存）。It's in Chinese. That should do it, I believe it :razz: :cool:

trinibwoy

27-Apr-2007, 18:53

I thought it was 8x64?

I think the rumour is that the memory controller is configured as 8x64. But that could mean two modules per controller - not necessarily 64-bit modules.

Jawed

27-Apr-2007, 18:54

I thought it was 8x64?
R5xx introduced the 32-bit per channel memory controller. Why would ATI go backwards?

Plus, these memory chips are 32-bit. Look at 8800GTX, it takes 12 chips to provide its 384-bit bus. It merely gangs pairs of chips to create 64-bit memory channels.

Jawed

gunblade

27-Apr-2007, 18:58

GDDR3/4/5 only come in x32 configuration and gDDR2 comes in only x16 configuration.

jimmyjames123

27-Apr-2007, 19:03

But would it work for all games?

It might not work on all games, but would that matter? With dual cards, one could get higher image quality using the special AA modes, and in most cases get superior frame rates as well.

NVIDIA just doesn't have a cost-effective solution to answer Crossfire HD 2900 XT 1024MB RAM total. Two 8800 GTS in SLI would most likely be slower, and two 8800 GTX in SLI would have 512MB RAM more in comparison which would make it much more expensive. NV has to move to a 512 bit bus to get more parity on RAM costs in this high end segment.

_xxx_

27-Apr-2007, 19:08

SugarCoat

27-Apr-2007, 19:08

It might not work on all games, but would that matter? With dual cards, one could get higher image quality using the special AA modes, and in most cases get superior frame rates as well.

NVIDIA just doesn't have a cost-effective solution to answer Crossfire HD 2900 XT 1028MB RAM total. Two 8800 GTS in SLI would most likely be slower, and two 8800 GTX in SLI would have 512MB RAM more in comparison which would make it much more expensive. NV has to move to a 512 bit bus to get more parity on RAM costs in this high end segment.

And whats your basis for thinking ram costs would be so devastating that nVidia/AIBs would be helpless? You also arent factoring in the ratio of current SLI ready boards compared to CrossFire ones, theres a huge disparity. I think the idea of taking mainstream to a dual card solution is odd at best, 1 card against 1 card is the way its always been done, regardless of any type of bundle.

jimmyjames123

27-Apr-2007, 19:11

Looks like 8800 GTX prices will have to come down if ATI will market a crossfire HD 2900 XT 1024MB for a similar MSRP. So while the crossfire solution will outperform the GTX, NV would have a cost advantage in comparison and could price their GTX much lower than they did at launch. However, the Ultra model doesn't seem to serve much purpose, unless they price it much lower than rumored MSRP.

jimmyjames123

27-Apr-2007, 19:17

And whats your basis for thinking ram costs would be so devastating that nVidia/AIBs would be helpless? You also arent factoring in the ratio of current SLI ready boards compared to CrossFire ones, theres a huge disparity. I think the idea of taking mainstream to a dual card solution is odd at best, 1 card against 1 card is the way its always been done, regardless of any type of bundle.

I don't think NV/AIB's will be helpless. They will simply have to reduce the high margins that they are currently earning on the 8800 GTX. No doubt, the GTX should have a notable cost advantage over a 1024MB crossfire solution, so the GTX can easily carve out a nitch at MSRP lower than the current $649 and much lower than a 1024MB crossfire solution. The crossfire would be a compelling high end solution though. It would be unlikely that the Ultra would outperform it. However, it's true that SLI/Crossfire is not for everyone, and those people would be attracted to a single Ultra.

Note that the 640MB 8800 GTS can now be found for as little as $359 online (after $30 rebate) from newegg, so the price war is already underway. Two 640MB 8800 GTS in SLI would be a pretty sweet combo for ~$700, and would surely give the XT crossfire a better run for it's money than a single GTX (in the situations where SLI actually significantly boosts performance).

Maybe it is possible that ATI had little choice other than to market a crossfire solution in their high end segment. The R600 seems to have much less flexibility than the G80 in terms of coming up with a wide variety of cards with different stream processors, memory bus size, shader clock domains, etc. NV's G8x architecture seems incredibly flexible in that regard. They just need a 512 bit bus solution in order to have all combinations covered against ATI cards.

turtle

27-Apr-2007, 19:17

R5xx introduced the 32-bit per channel memory controller. Why would ATI go backwards?

Plus, these memory chips are 32-bit. Look at 8800GTX, it takes 12 chips to provide its 384-bit bus. It merely gangs pairs of chips to create 64-bit memory channels.

Jawed

Right on, that makes sense. I imagine the ring bus will look very interesting with that configuration. Sounds very congested.

mao5

27-Apr-2007, 19:22

That's not what the sorce for those pictures says: Radeon HD 2900 XT显卡的最终工作频率定在740/825MHz（核心/显存）。

I'm sorry for this problem, I didn't realized it was a 1GB XTX.

mao5

27-Apr-2007, 19:25

At last, we have an explaination, as I'm saying the xtx DT used is an old engeering sample, the real xtx 1GB card which can be get by NDA is the below one, Dailytech totally messed up

http://www.chiphell.com/attachments/month_0704/20070428_e771f730b96681515265iEl2FoDRp8gJ.jpg

mao5

27-Apr-2007, 19:26

Ye but that card looks bigger than 9.5" and you can clearly see the extra memory on the back(1024MB) so it's not an XT.

US

That's correct, it's 1GB XTX

Mize

27-Apr-2007, 19:26

mao5

27-Apr-2007, 19:28

It's in Chinese. That should do it, I believe it :razz: :cool:

I wrote these chinese words, but at that time, I didn't realized it was a 1GB xtx, I thought it was a 512 XT. Sorry for that problem.

Bob

27-Apr-2007, 19:30

Being able to offer a high performance 1028MB total RAM crossfire solution (where NVIDIA right now can only offer 1280MB or 1536MB total RAM on GTS and GTX) is certainly a notable price-to-performance advantage for ATI in terms of cost of RAM, and that is tied to the fact that their card also uses a 512 bit bus. I said weeks ago that NVIDIA would need a 512 bit bus in order to better compete on a cost basis (in terms of on-board RAM cost).
You're assuming that 12 chips @ 128 MB each is more expensive than 16 chips @ 64 MB each. That's not necessarily the case, even ignoring the board design complexity difference.

Plus, you may want to check your powers of 2.

Jawed

27-Apr-2007, 19:31

Right on, that makes sense. I imagine the ring bus will look very interesting with that configuration. Sounds very congested.
The mapping between ring stops and memory channels isn't 1:1, so who knows what configuration R600 will use. 4 ring stops with 4 memory channels each seems pretty likely though.

A lot depends on cache architecture. Does each shader unit (one quarter of R600) have a walloping big L2 cache for all its clients (TMUs, vertex fetch, ROPs, etc.) or does each client type have its own L2 cache? If the latter, then you might argue that more ring stops would make sense.

Jawed

Geeforcer

27-Apr-2007, 19:32

Mao5, why do you think it's an XTX? (Unless you know for sure at this point)

mao5

27-Apr-2007, 19:36

Why do we think DT's was different? Might it simply have had a different (extend past the PCB) cooler? The power connects are about right if you consider the DT shots show a cooler extending past the end of the board.

Let comparing pic does the job, left is DT version,right is NDA verion, both are 1GB XTX.

http://www.chiphell.com/attachments/month_0704/20070428_a9e35c03f1a7446f865bhTe3jX7KAEUs.jpg

jimmyjames123

27-Apr-2007, 19:37

You're assuming that 12 chips @ 128 MB each is more expensive than 16 chips @ 64 MB each. That's not necessarily the case, even ignoring the board design complexity difference.

Plus, you may want to check your powers of 2.

If a lower quantity of 128MB chips is cheaper than a higher quantity of 64MB chips, then why would ATI not use four 128MB chips on their 512MB XT card (is their some type of techical reason for using 64MB chips based on their architecture design)? That said if their 1024MB RAM on XT crossfire is really not cheaper than 1536MB RAM on GTX SLI, then ATI would be just destroying their margins, and it would be more like a desperate move on their part (unless their performance in Crossfire ends up being significantly faster than SLI priced equivalents). One way or another, NV is in the driver's seat right now and NV seems to have a more flexible architecture too.

Oops, I don't know why I kept typing 1028MB, I'll fix that.

Star_Hunter

27-Apr-2007, 19:40

Anyone know if these Crysis benchmark scores are real?

http://www.hardwareanalysis.com/content/topic/66310/

Cuthalu

27-Apr-2007, 19:40

Edit: Nothing.

Geeforcer

27-Apr-2007, 19:42

Let comparing pic does the job, left is DT version,right is NDA verion, both are 1GB XTX.

The left is OEM version. The right is a retail version. I am quite mistified as to why you are sure the right one is a 1GB XTX, considerign that the back of that card looks exactly the same as the back of 512MB 2900 XT DailyTech tested.

mao5

27-Apr-2007, 19:42

How so? How did you came up with that conclusion?

extra memory on the back(1024MB)

http://www.chiphell.com/attachments/month_0704/20070428_4c5251a15600149fc902UhoNyb5pSjfi.jpg

Geeforcer

27-Apr-2007, 19:45

Did you miss the whole "16 chips for 512MB bus" things?

mao5

27-Apr-2007, 19:45

The left is OEM version. The right is a retail version. I am quite mistified as to why you are sure the right one is a 1GB XTX, considerign that the back of that card looks exactly the same as the back of 512MB 2900 XT DailyTech tested.

remember this pic?

http://www.chiphell.com/attachments/month_0704/20070428_f432bddf0ad6864bc797HS3EuwgO1Ele.jpg

Mize

27-Apr-2007, 19:46

Let comparing pic does the job, left is DT version,right is NDA verion, both are 1GB XTX.

Thanks. Missed that DT shot.

trinibwoy

27-Apr-2007, 19:48

extra memory on the back(1024MB)

As explained earlier in the thread, the XT probably has memory on the back as well.

Geeforcer, why do you think the left pic is OEM? The left PCB looks significantly shorter than the one on the right.

Geeforcer

27-Apr-2007, 19:48

extra memory on the back(1024MB)

http://www.chiphell.com/attachments/month_0704/20070428_4c5251a15600149fc902UhoNyb5pSjfi.jpg

Now compare that with the back of 512MB 2900 XTX (the card on the bottom):
http://images.dailytech.com/nimage/4560_large_r600xtx_2.jpg
See the difference? Yeah, me neither.

vertex_shader

27-Apr-2007, 19:48

Let comparing pic does the job, left is DT version,right is NDA verion, both are 1GB XTX.

http://www.chiphell.com/attachments/month_0704/20070428_a9e35c03f1a7446f865bhTe3jX7KAEUs.jpg

Left card in the picture is a naked OEM version.

Hmm, in the right picture the PCB looks ~1.2inch longer.

Geeforcer

27-Apr-2007, 19:49

As explained earlier in the thread, the XT probably has memory on the back as well.

Geeforcer, why do you think the left pic is OEM? The left PCB looks significantly shorter than the one on the right.

It has to be be shorter in order for the huge rear cooler to fit. Look at the DT picture above.

trinibwoy

27-Apr-2007, 19:49

Aaaargh, will you guys please create smaller versions of those images? You're screwing up the page!!

mao5

27-Apr-2007, 19:50

Maybe I'm wrong,but pic analysis is a good start.

trinibwoy

27-Apr-2007, 19:50

It has to be be shorter in order for the huge rear cooler to fit. Look at the DT picture above.

Ah, I see what you're saying - PCB on OEM is shorter to allow for the fan off the edge. But why would the retail PCB be longer? More power circuitry required?

mao5

27-Apr-2007, 19:52

Left card in the picture is a naked OEM version.

Hmm, in the right picture the PCB looks ~1.2inch longer.

left is a 1GB XTX OEM verion sell on pcinlife.com several week ago and I think it's the card also used by DT. Its chip marked with 0704 (Feb, 2007), R600 chip respinned for several times since Feb, 2007. DT should show us a naked chip shoot.

BlizzardOne

27-Apr-2007, 19:54

Aaaargh, will you guys please create smaller versions of those images? You're screwing up the page!!

fits perfectly on my dell 24"! :razz: :lol:

that aside.. how many different versions are there floating around now?? reckon this is AMD's tactic - confuse the hell out of everybody, then floor them with some impressive numbers? :lol:

vertex_shader

27-Apr-2007, 19:55

Now compare that with the back of 512MB 2900 XTX (the card on the bottom):
http://images.dailytech.com/nimage/4560_large_r600xtx_2.jpg
See the difference? Yeah, me neither.

Looks like dailytech have a dodgy card :lol:
Check the PCB lenght in the OEM version, is noticable smaller than the retail version 9.5 inch, rumors suggest that the OEM have the same PCB size, i think thats a dev sample.

mao5

27-Apr-2007, 19:55

Aaaargh, will you guys please create smaller versions of those images? You're screwing up the page!!

I'm really sorry.

Mize

27-Apr-2007, 19:56

It's a bit sick that this thread is bumping about once ever six seconds.

Mize

27-Apr-2007, 19:57

Aaaargh, will you guys please create smaller versions of those images? You're screwing up the page!!

Opera -> View -> Fit to Width.

Cuthalu

27-Apr-2007, 19:58

extra memory on the back(1024MB)
]

I don't see any pictures of Dailytech's 2900xtx showing ram.

Geeforcer

27-Apr-2007, 20:00

Ah, I see what you're saying - PCB on OEM is shorter to allow for the fan off the edge. But why would the retail PCB be longer? More power circuitry required?

Because you CAN space things out if you don't have to worry about PCB+fan length. If you look at the picture, you'll see that some of the power regulators have been moves to the rear of the retail card.

trinibwoy

27-Apr-2007, 20:05

fits perfectly on my dell 24"! :razz: :lol:

Yeah I know....my resolution at home is 1920x1200. But unfortunately I'm stuck at at 1280x1024 at work :(

Razor1

27-Apr-2007, 20:05

Mao can you get closer pics of the GPU? so we can see whats written on it. Or they don't want to take the paste off :grin:

Geeforcer

27-Apr-2007, 20:10

Looks like dailytech have a dodgy card :lol:
Check the PCB lenght in the OEM version, is noticable smaller than the retail version 9.5 inch, rumors suggest that the OEM have the same PCB size, i think thats a dev sample.

Let's do the math. If retail PCB is 9.5 inches long, and the OEM is 12.5 inches long, then the whole rear fan assembly would have to be 3 inches. 3 inches ~ 7.5 cm. That would mean that ATI uses the... err....always popular... 75mm fan, they would still not have enough room for the housing or the rear handle. And the actuall fan on the OEM version looks like a 80-92mm unit.

Conclusion: OEM version HAS to use a shorter PCB then retail one.

vertex_shader

27-Apr-2007, 20:11

http://www.chiphell.com/attachments/month_0704/20070428_4c5251a15600149fc902UhoNyb5pSjfi.jpg

I read hynix in the ram, hynix not have any gddr4 memory in the website (http://www.hynix.com/eng/)

SugarCoat

27-Apr-2007, 20:12

Let's do the math. If retail PCB is 9.5 inches long, and the OEM is 12.5 inches long, then the whole rear fan assembly would have to be 3 inches. 3 inches ~ 7.5 cm. That would mean that ATI uses the... err....always popular... 75mm fan, they would still not have enough room for the housing or the rear handle. And the actuall fan on the OEM version looks like a 80-92mm unit.

Conclusion: OEM version HAS to use a shorter PCB then retail one.

i dont remember that being the case and personally i dont think that makes any sense from a production standpoint. They'll problably be the exact same. Im sure there were multiple revisions, but the final production models should be damn near identical once you remove the HSF.

mao5

27-Apr-2007, 20:18

Mao can you get closer pics of the GPU? so we can see whats written on it. Or they don't want to take the paste off :grin:

I compare the oem GPU and NDA GPU, any difference is the golden mark on the top edge on NDA GPU, and 0704 mark on oem GPU, so I really want to see the naked xtx chip pic of DT version.

http://www.chiphell.com/attachments/month_0704/20070428_6b3f38d04bb6b78ef5a60rh1CkRdImkz.jpg

mao5

27-Apr-2007, 20:20

http://www.chiphell.com/attachments/month_0704/20070428_4c5251a15600149fc902UhoNyb5pSjfi.jpg

I read hynix in the ram, hynix not have any gddr4 memory in the website (http://www.hynix.com/eng/)

if this's true, then I'm wrong, it's a 512MB XT

Razor1

27-Apr-2007, 20:24

http://www.hynix.com/eng/02_products/04_graphics/01_04_00.jsp

don't know if it was released yet or not.....

mao5

27-Apr-2007, 20:30

http://www.hynix.com/eng/02_products/04_graphics/01_04_00.jsp

don't know if it was released yet or not.....

http://translate.google.com/translate?u=http%3A%2F%2Fwww.hardsalon.com%2F2005% 2F12-30%2F09410640643.html&langpair=zh%7Cen&hl=en&newwindow=1&ie=UTF-8&oe=UTF-8&prev=%2Flanguage_tools

Hynnix GDDR4 volum production started

Geeforcer

27-Apr-2007, 20:30

i dont remember that being the case and personally i dont think that makes any sense from a production standpoint. They'll problably be the exact same. Im sure there were multiple revisions, but the final production models should be damn near identical once you remove the HSF.

Again, it's simple math. Look at the size of the fan attachment on the OEM card and compare it's length with a width of the PCB. To me, it looks longer then PCB is wider by at least 1cm. That would put the length of the cooler at ~9.5 cm. That's almost 4 inches there alone.

jamis

27-Apr-2007, 20:33

I read hynix in the ram, hynix not have any gddr4 memory in the website (http://www.hynix.com/eng/)
Dunno about their website but I think they've had GDDR4 memory for about a year now.

mao5

27-Apr-2007, 20:35

so right now we can not tell if the picture on the right is a NDA XTX.

vertex_shader

27-Apr-2007, 20:38

Dunno about their website but I think they've had GDDR4 memory for about a year now.

Yes they have, but i never heard/read anything about mass production hynix gddr4 rams.

I can't make the numbers readable in the ram chips :sad:

Geeforcer

27-Apr-2007, 20:40

Well, if you have really good eyes, all you need is to read the first 4 characters on the model number. HY 5F is GDDR4 and HY 5R is GDDR3.

AlexV

27-Apr-2007, 20:40

left is a 1GB XTX OEM verion sell on pcinlife.com several week ago and I think it's the card also used by DT. Its chip marked with 0704 (Feb, 2007), R600 chip respinned for several times since Feb, 2007. DT should show us a naked chip shoot.

There have been SEVERAL respins from Feb, 2007?Really now?And how long do these respin things last, I wonder...like, you call the Fab and they have it done by:a)mid-morning;b)afternoon;c)late evening?I`m gleefully twiddling my thumbs in order to hear about this respin stuff, seems it`s fairly easy and quickish to do.

Unknown Soldier

27-Apr-2007, 20:42

So maybe it's gonna be a 512MB GDDR4 card.

Hey, since they can't fit 32 16MBx32bit chips on a card it makes sense.

US

vertex_shader

27-Apr-2007, 20:44

Well, if you have really good eyes, all you need is to read the first 4 characters on the model number. HY 5F is GDDR4 and HY 5R is GDDR3.

I try figure out the speed (right to left last number is the temperature, than the 2 number is the speed), i see 25, but thats not possible, 25 means 2.5ns memory at 400mhz :smile:

SugarCoat

27-Apr-2007, 20:50

Again, it's simple math. Look at the size of the fan attachment on the OEM card and compare it's length with a width of the PCB. To me, it looks longer then PCB is wider by at least 1cm. That would put the length of the cooler at ~9.5 cm. That's almost 4 inches there alone.

as i said, multiple revisions.

Geeforcer

27-Apr-2007, 20:57

So you are saying that retail cards will be shorter than 9.5 inches?

pjbliverpool

27-Apr-2007, 21:05

I'm not sure what the 47.5GP/s rate is meant to be? Fragment rate seems likely, 64*742.

Jawed

It will be samples, 4 per ROP at 742Mhz = 47.49 Gsamples.

mao5

27-Apr-2007, 21:09

So maybe it's gonna be a 512MB GDDR4 card.

Hey, since they can't fit 32 16MBx32bit chips on a card it makes sense.

US

it make sense, it's a 512MB GDDR4 XTX

Geeforcer

27-Apr-2007, 21:51

There is a 512MB XTX now?

EasyRaider

27-Apr-2007, 21:59

There was from the beginning. Actually, didn't the 256 MB show up later?

Jawed

27-Apr-2007, 22:09

It will be samples, 4 per ROP at 742Mhz = 47.49 Gsamples.
Regardless of that slide, what you've said is likely to be true! as 4xAA per loop seems extremely likely.

Since this is seemingly an official AMD slide, the terminology should be meaningful. Being blunt, if AMD meant samples per second, the slide would say so.

Since it says pixels per second, but separately says 16 render back-end (another way of saying ROP), "pixels per second" isn't referring to ROP colour fillrate.

---

This makes me wonder if R600's ALUs can process 64 pixels (or vertices or primitives) at any one time but they perform 5 different (or the same) scalar instructions on each of those 64 pixels.

---

It's just bloody annoying. So near yet so far. I would expect such a slide to state the AA sample rate, so the fact it doesn't do so is suspicious in itself.

So, I'm still confused.

Jawed

Geeforcer

27-Apr-2007, 22:12

So, to recap: The card was originally labeled as 2900 XT. Then, it became 1GB XTX because OMG Snakes on the PLa.... I mean, chips on the back. Then, when it was pointed out that RAM on the back in of itself != 1GB, it became a 2900 XTX 512MB. All the while the card in pictures looks identical to 2900 XT picture on DailyTech. Am I missing anything?

Arty

27-Apr-2007, 22:18

Am I missing anything?
The snakes! :lol:

I think we need to keep our cool as the day is not far.

ERK

27-Apr-2007, 22:38

one slide of $399 Radeon HD 2900XT confirms 16 texture units.

http://r800.blogspot.com/2007/04/radeon-hd-2900xt.html

This goes without saying, but...

The 475GFLOPS number ~confirms 1 MADD per SP, which we've pretty much known since the TFLOP box.

ERK

Unknown Soldier

27-Apr-2007, 22:58

Geeforcer

27-Apr-2007, 23:07

I never said it can't be, just that it doesn't have to be; as opposed to people who said it can't be 512MB due to "OMFGSnakes!!!!"

Arnold Beckenbauer

27-Apr-2007, 23:50

This goes without saying, but...

The 475GFLOPS number ~confirms 1 MADD per SP, which we've pretty much known since the TFLOP box.

ERK

Or 64 Vec4+1 ALUs with 5 MADDs and you get 475 GLOP/s (740 MHz).
R520's eight VS ALUs are "5D" - Vec4+1 and groupped in one "block" (SIMD).
The next idea:
• Lets pretend that R600 has work blocks of 17 shaders and look at the consequences…
320/17= error
320/(16*(4*Vec1 ALUs)+16*Vec1 ALUs) = 4
Crazy, isn't it?

Kaotik

28-Apr-2007, 00:19

If you believe the current rumours, May 2nd is a paper launch, with actual retail availability in mid-May.

http://www.theinquirer.net/default.aspx?article=39223

I'm drunk etc, but this is something I still can't understand.
Cell phones get reviews even months before retail availability
Intel demonstrated Core2's months before retail availability
AMD demonstrated Barcelona months before retail availability
Few years back video cards were previewed, even reviewed before retail availability or even official launch
Yet no-one ever talks about "paperlaunches" when it comes to those, but now lifting NDA before actual "launch" 14th, it's a paperlaunch? :???: