The NEXT LAST R600 Rumours & Speculation Thread [Archive] - Page 22

Bouncing Zabaglione Bros.

12-May-2007, 11:02

Is this going to be R520: part 2?
First it was suppose to be amazing
Then it got delayed
Right before launch we heard from various sources about how it'll be give n' take between the x1800xtx and 7800GTX
Then this was rebuttled with "it's a driver issue. wait for better drivers!"
This was followed by the rumours of a faster part coming only a few months after
The x1800xtx was quickly dethroned and forgotten about with the release of the x1900 series

Does this sound about right to anyone else?

"wouda coulda shoulda" aside, if the R600 clearly does not show it's muscle by hanging with or outdoing the G80, it'll look quite foolish for Ati. They've had plenty of time and need to leave the excuses at the door. Monday we find out. Can't wait!

I think it's quite telling that there is no real XTX until R650 arrives at .65 in a couple more months. It seems that for whatever reason, AMD doesn't have a top-of-the-range product to go head-to-head with the competition, so they are having to try and push their second string product into top place. This explains the low price, and the less than screaming performance compared to G80.

I think that just like we didn't see the full realisation of R5x0 until R580, so we won't see the full realisation of R6x0 until R650 and I suspect it's much higher clocks. I also think that R650 will come sooner that we expect, because of that obvious gap at the XTX level of the AMD product range that is begging to be filled.

Galduta

12-May-2007, 11:45

A quick analysis on these scores reveals:

- a wrong copy/paste (FarCry, 1280&1920 8x AA)
- inconsistent scores (Oblivion, 2560 > 1920, to the point 8x AA > 4x AA in 2560)
- strange behaviors (F.E.A.R., 2560 8x AA shows the XT ahead while it shouldn't given the 4x and 16x AA numbers)

Well ... the "old" drivers - the Cd drivers - are a scary movie . 3d mark 2006 and 2005 not load , Aquamarck with the screen in black ... inconsistent scores in Stalker ,COD2 , poor perfomance in FEAR ... but well, the monday ...

pd /
Call of duty 2 v1.3 timedemo TECHPOWERUP
1600x1200
no AA and AFs: 66fps
FA@16x Q:38fps http://www.nextgpu.com/forum/Smileys/smilies_smf/crazy2.gif
FA@16x Q + AA@4x: 30fps http://www.nextgpu.com/forum/Smileys/smilies_smf/shocked.gif

1920x1200
no AA and AF: 58fps
FA@16x Q : 53.5fps http://www.nextgpu.com/forum/Smileys/smilies_smf/shocked.gif,
FA@16x@ AA@4x: 38

STALKER FA@16x Q:
bench1:
1280x1024:
8800GTS: min:17//med:44//max:84
2900XT: min:1//med:23//max:99 http://www.nextgpu.com/forum/Smileys/smilies_smf/2funny.gif

1600x1200:
8800GTS: min:13//med:35//max::88
2900XT: min:1//med:18//max::76

1920x1200:. http://www.nextgpu.com/forum/Smileys/smilies_smf/uglystupid2.gif
8800GTS: min:12//med:30//max::80
2900XT: min:1//med:26//max::104

etc etc . the 2900 is noisy in 3d , with good temperature in full - 70º - 8800 GTS 80º -.

w0mbat

12-May-2007, 12:26

Driver: 8.37-4, CPU: E6700, from europes biggest PC-Magazine ct

Oblivion, HDR on

1280 X 1024 noAA/noAF

2900 XT 49 fps
88 GTS 48 fps
88 GTX 48 fps

1280 X 1024 AA 4X / noAF

2900 XT 39 fps
88 GTS 46 fps
88 GTX 48 fps

1280 x1024 AA 8X / noAF

HD 2900 XT 17 fps
88 GTS 28 fps
88 GTX 39 fps

PREY

1600 x 1200 AA 8X / AF 16X

HD 2900 XT 43 fps
88 GTS 37 fps
88 GTX 50 fps

Windows Vista Direct X10 SDK

PIPEGS

2900XT 159 fps
88GTS 34 fps
88GTX 63 fps

CubemapGS, Car, Instancing

2900XT 23 fps
88GTS 9 fps
88GTX 11 fps

Cubemap, Car, Instancing

2900XT 18 fps
88GTS 10 fps
88GTX 11 fps

Love_In_Rio

12-May-2007, 12:29

What´s up with AA ?

w0mbat

12-May-2007, 12:33

What´s up with AA ?

I think its the drivers cause theres enough bandwidth.

IbaneZ

12-May-2007, 12:45

AF will probably hurt R600 more than AA.

AnarchX

12-May-2007, 13:02

The Benches were made with AAA/TRSAA. :wink:

trinibwoy

12-May-2007, 13:04

trinibwoy

12-May-2007, 13:10

PIPEGS

2900XT 159 fps
88GTS 34 fps
88GTX 63 fps

Hmmm the DX10 face-off should certainly be interesting. Looking forward to seeing what ATI did specifically to enhance DX10 performance.

vertex_shader

12-May-2007, 13:20

The Benches were made with AAA/TRSAA. :wink:

Why they not use EATM ?

AnarchX

12-May-2007, 13:22

Why they not use EATM ?

Still not official?

But here you see that the low texel-fillrate can broke R600's neck.

vertex_shader

12-May-2007, 13:26

Still not official?

But her you see that the low texel-fillrate can broke R600's neck.

Rumors are true its coming officialy with catalyst 7.5, what is few days away, so the beta/build version drivers has it too i think.

trinibwoy

12-May-2007, 13:27

A little video from the AMD/[H] party a couple of days ago.

http://tmc.kuyaglen.com/events/amdhardocp/hardocp_amd_preview.wmv

Looks like it was a pretty decent time although I thought more people would have shown up. I'm relieved to see that Kyle looks and acts like a regular guy. Funny how the AMD PR guy hasn't even gotten used to saying "HD" as yet - he started off saying "X" :lol:

AlexV

12-May-2007, 13:29

Hmmm the DX10 face-off should certainly be interesting. Looking forward to seeing what ATI did specifically to enhance DX10 performance.

They made buggy nVidia drivers :D The way that performance scales from GTS to GTX is odd...unless I`m mistaken(too lazy to look now), the GTX isn`t 2x shading power WRT GTS, even though that test suggests it. Very interesting.

ChrisRay

12-May-2007, 14:06

Looks like it was a pretty decent time although I thought more people would have shown up. I'm relieved to see that Kyle looks and acts like a regular guy. Funny how the AMD PR guy hasn't even gotten used to saying "HD" as yet - he started off saying "X" :lol:

Kyle isnt a bad guy to meet and talk too. He's pretty forward and reasonable to talk too. But he's definately got the same aura he does on the forums. Extreme confidence in himself. But not a bad guy at all.

Sobek

12-May-2007, 15:06

I know nothing's totally concrete...but i'm supremely unimpressed with the HD2900. Supremely. Un. Im. Pressed. Don't know what else to say...I truly expected more for all this time spent waiting. I'll still buy one though, perhaps two and go for Crossfire. But I was really expecting just a *tad* more :???:

PSU-failure

12-May-2007, 15:08

They made buggy nVidia drivers :D The way that performance scales from GTS to GTX is odd...unless I`m mistaken(too lazy to look now), the GTX isn`t 2x shading power WRT GTS, even though that test suggests it. Very interesting.

These numbers don't suggest that, but the possibility the G80 could be the direct successor of the NV30 due to the SIMD design and perhaps some other flaws.

Correct me if I'm wrong, but the GTS only has 6 out of the 8 G80's SIMD units, which reduces the flexibility even more, thus the 33% lower theorical mathematical power has a chance to become 50%.

If this is exact, nVidia will have won his bet by selling thousands of "NV30 like" DX10 GPUs before any use of the API is made.

INKster

12-May-2007, 15:13

These numbers don't suggest that, but the possibility the G80 could be the direct successor of the NV30 due to the SIMD design and perhaps some other flaws.

Correct me if I'm wrong, but the GTS only has 6 out of the 8 G80's SIMD units, which reduces the flexibility even more, thus the 33% lower theorical mathematical power has a chance to become 50%.

If this is exact, nVidia will have won his bet by selling thousands of "NV30 like" DX10 GPUs before any use of the API is made.

Comparing NV30 with G80 is like comparing a Pentium 4 with a Core 2.
Especially so when a "feature" suddenly turns into a "flaw", just because it didn't work on a previous and completely different main architecture design...
Ever heard the saying "What doesn't kill us makes us stronger" ? ;)

Skinner

12-May-2007, 15:29

If this is exact, nVidia will have won his bet by selling thousands of "NV30 like" DX10 GPUs before any use of the API is made.

Why do you think it's that bad in dx10?

PSU-failure

12-May-2007, 15:36

Comparing NV30 with G80 is like comparing a Pentium 4 with a Core 2.
Especially so when a "feature" suddenly turns into a "flaw", just because it didn't work on a previous and completely different main architecture design...
Ever heard the saying "What doesn't kill us makes us stronger" ? ;)
You know, NV30 had lots of wonderfull features... according to nVidia :twisted:

The SIMD design of the G80 made me think there was a risk performance-wise since day 1 as it isn't that flexible and requires alot of work and cache to be efficient if the instructions are not excessively similar. Removing 2 SIMD units out of 8 removes even more flexibility, leaving only 6 distinct instructions to be handled simultaneously and thus reducing efficiency.

Don't take all this as truth as I'm not stating this is the case, but these numbers suggest it.

aca

12-May-2007, 15:47

Comparing NV30 with G80 is like comparing a Pentium 4 with a Core 2.

Maybe the Core 2 could be compared to a Pentium 3 since it belongs to the same pedigree tree?

Andrew Lauritzen

12-May-2007, 15:48

PIPEGS
2900XT 159 fps
88GTX 63 fps

CubemapGS, Car, Instancing
2900XT 23 fps
88GTX 11 fps

Cubemap, Car, Instancing
2900XT 18 fps
88GTX 11 fps
Aha - this seems to be confirming my theory that if R600 is faster anywhere, it will be in geometry shading. Granted G80 still has a lot of space to improve in that area: a recent test that I did indicated that even having *any* geometry shader present (I used an "identity" geometry shader that just passed through the triangles) cut the performance of the shader in *half*.

It will certainly be interested to see... in any case I think the Cubemap examples are a bit contrived to begin with because arguably separately rendering the faces with a good frustum culling algorithm is going to be just as fast or faster except in the most degenerate cases.

What I really want to see is how well R600 runs cascades...

IbaneZ

12-May-2007, 16:05

Kyle isnt a bad guy to meet and talk too. He's pretty forward and reasonable to talk too. But he's definately got the same aura he does on the forums. Extreme confidence in himself. But not a bad guy at all.

He's a very handsome man too. :grin:

trinibwoy

12-May-2007, 16:07

The SIMD design of the G80 made me think there was a risk performance-wise since day 1 as it isn't that flexible and requires alot of work and cache to be efficient if the instructions are not excessively similar. Removing 2 SIMD units out of 8 removes even more flexibility, leaving only 6 distinct instructions to be handled simultaneously and thus reducing efficiency.

What about G80's SIMD design is inflexible, especially compared to R600's? Instructions need to be similiar? Only six instructions? Reducing efficiency? What? :???: :)

trinibwoy

12-May-2007, 16:09

Aha - this seems to be confirming my theory that if R600 is faster anywhere, it will be in geometry shading. Granted G80 still has a lot of space to improve in that area: a recent test that I did indicated that even having *any* geometry shader present (I used an "identity" geometry shader that just passed through the triangles) cut the performance of the shader in *half*.

Hold on, you created a geometry shader that did no work and it still halved performance? :shock:

INKster

12-May-2007, 16:10

Maybe the Core 2 could be compared to a Pentium 3 since it belongs to the same pedigree tree?

Well, by that order or thinking, we could trace much or AMD's K10/Barcelona's basic design back to the original Athlon and Athlon XP, right up to the similar L1 and L2 cache size and configuration...

You know, NV30 had lots of wonderfull features... according to nVidia :twisted:

The SIMD design of the G80 made me think there was a risk performance-wise since day 1 as it isn't that flexible and requires alot of work and cache to be efficient if the instructions are not excessively similar. Removing 2 SIMD units out of 8 removes even more flexibility, leaving only 6 distinct instructions to be handled simultaneously and thus reducing efficiency.

Don't take all this as truth as I'm not stating this is the case, but these numbers suggest it.

And what do the actual performance numbers suggest when you compare a G80 to a NV30 ?

trinibwoy

12-May-2007, 16:10

He's a very handsome man too. :grin:

Well I agree he certainly doesn't look like the typical socially castrated tech-geek but handsome? That's pushing it a little too far (j/k) :D Does look like he works out, although that could just be fat and bad lighting playing tricks on me.

PSU-failure

12-May-2007, 16:22

What about G80's SIMD design is inflexible, especially compared to R600's? Instructions need to be similiar? Only six instructions? Reducing efficiency? What? :???: :)
8 16-wide SIMD units are far less flexible than 128 independant SP.

Each unit has to execute 16 times the same Op to be efficient, leaving only 8 different branches, and even 6 in the case of the 8800GTS. To do that, you must reorder the instructions (out of order execution, heavy task) and store temporarily unuseable data.

Now, if we consider R600 to have 8 8-wide SIMD units (8 clusters of 5 scalars using the same Ops combination), there are 8 different "paths" too, but up to 40 different Ops could be executed at the same time, reducing the reordering task.
If R600 has 16 4-wide SIMD units, it's even more flexible and the fact each cluster could be used to treat a particular pixel adds a little more too.

Galduta

12-May-2007, 16:24

Well the 8.37

FEAR

1600x1200 FA@16x, FAA@4x SS
2900XT/8800GTS:
min:17/20
med:35/37
max:64/70

1920x1200 FA@16x, FAA@4x SS
2900XT/8800GTS:
min:16/16
med:30/32
max:55/57

2900XT - 8.37- 8800GTS - 158.19 -

1600x1200 FA@16x, FAA@4x no SShadows,
2900XT/8800GTS:
min:20/30
med:59/58
max:124/130

1920x1200 FA@16x, FAA@4x no SShadows,
2900XT/8800GTS:
min:19/27
med:51/51
max:111/104

STALKER 1600x1200 all máximun, AFMax y noAA - C2 3200

Timedemo1: Av: 37,55, Min: 19,79, Max: 123,84, Mid: 39,52
Timedemo2: Av: 60,14, Min: 37,60, Max: 291,21, Mid: 65,48

.8,36

Timedemo1: Average: 33,12, Min: 17,51, Max: 88,29, Mid: 35,51
Timedemo2: Average: 50,50, Min: 31,28, Max: 282,81, Mid: 56,67

With the 8.37 , disappear in the control panel the setting "high quality anisotropic " ;)....

Rys

12-May-2007, 16:27

Hold on, you created a geometry shader that did no work and it still halved performance? :shock:
Currently it seems to be that if any GS shader is scheduled, the hardware will decrease the number of threads running for the other types indiscriminately. That's the driver doing that, though (as far as I can tell), rather than something at the hardware level.

Julidz

12-May-2007, 16:29

could u explain the difference between Vec5 superscalar and Vec4 + 1 scalar architectures ?

silent_guy

12-May-2007, 16:30

8 16-wide SIMD units are far less flexible than 128 independant SP.

Each unit has to execute 16 times the same Op to be efficient, leaving only 8 different branches, and even 6 in the case of the 8800GTS. To do that, you must reorder the instructions (out of order execution, heavy task) and store temporarily unuseable data.

...etc etc

A little bit of knowledge is often worse than no knowledge at all...

dizietsma

12-May-2007, 16:33

Glad to see that Rys will be concentrating on the architecture etc and probably some theoretical limit tests and not yet game tests with such, it seems, immature drivers.

I have been reading this thread and the first thing I do not like is these reported big jumps in 3dmark with every single release. This reminds me of something Green did in the past ( no not AMD ) when their hardware was not up to the task. I like to see small jumps in 3dmark matched by small jumps in other games to go along with it, that shows me good progression. However I do not like big jumps in 3dmark with no big jumps in games, so I hope, when we see this thing tested, tha sites use old and new drivers for both games and 3dmark and see if they go up the same amount. Maybe I am just a bit cynical. I doubt very much AMD have resorted to some things that nvidia did to try and polish a turd in the past but it does strike me of damage limitation.

The second thing is the number of drivers released in such a short period of time. AND THE CARD IS MONTHS LATE. What would they have been like 8 months ago one has to wonder ?

I really think G80 caught ATI with their pants down, all that talk from Dr Evil, sorry Dr David Kirk :D , about how it was not a unified architecture and really just used a 4x2 pipeline arrangement made out of spaghetti .. ATI fell for it good and proper !

I bet David is sitting in his mountain top lair, stroking his white Persian cat and chuckling maniacally from behind his high backed swivel chair as I write.

IbaneZ

12-May-2007, 16:35

Well I agree he certainly doesn't look like the typical socially castrated tech-geek but handsome? That's pushing it a little too far (j/k) :D

:lol:

Jawed

12-May-2007, 16:46

I really think G80 caught ATI with their pants down, all that talk from Dr Evil, sorry Dr David Kirk :D , about how it was not a unified architecture and really just used a 4x2 pipeline arrangement made out of spaghetti .. ATI fell for it good and proper !
Eh?

G80's GS is worse than it should be (half-performance for a null shader :shock: ), theoretically, and you think that NVidia building a unified architecture is somehow surprising to ATI? Apparently the surprise is that it's not working at all well.

Or maybe that's not so much of a surprise. Take it whichever way you like :smile:

Jawed

trinibwoy

12-May-2007, 16:54

Currently it seems to be that if any GS shader is scheduled, the hardware will decrease the number of threads running for the other types indiscriminately. That's the driver doing that, though (as far as I can tell), rather than something at the hardware level.

Interesting. That sounds a bit lazy. I was under the impression that the number of threads was scaled on the fly based on register pressure. Would they do something like this to guarantee that sufficient register space was available for the GS threads?

trinibwoy

12-May-2007, 16:55

A little bit of knowledge is often worse than no knowledge at all...

Couldn't have said it better myself :lol:

PSU-failure, I think it's time you replaced that PSU ;)

fellix

12-May-2007, 16:58

Awful scheduling (or lack of, being er.. "emulated"), very latent state switching (poluting the pipeline), buffer/stack trashing . . . you name it. :roll:

Jawed

12-May-2007, 16:59

Interesting. That sounds a bit lazy. I was under the impression that the number of threads was scaled on the fly based on register pressure. Would they do something like this to guarantee that sufficient register space was available for the GS threads?
It may have something to do with the fact that a GS is supposed to be able to generate new vertices.

D3D10 requires the developer to specify an upper bound on the the number of vertices generated per input vertex. It might be that the driver is playing safe and always assuming the maximum...

Jawed

Razor1

12-May-2007, 17:00

nAo

12-May-2007, 17:01

Eh?

G80's GS is worse than it should be (half-performance for a null shader :shock: ), theoretically, and you think that NVidia building a unified architecture is somehow surprising to ATI? Apparently the surprise is that it's not working at all well.

You seem to imply that lack of GS power is related to having a unified architecture, while at the moment we really dont' have a clue about that. Too be honest we also really don't have much info about GS performance at all.

Jawed

12-May-2007, 17:04

You seem to imply that lack of GS power is related to having a unified architecture, while at the moment we really dont' have a clue about that. Too be honest we also really don't have much info about GS performance at all.
No we don't have much info. But Rys did say that load-balancing in this unified architecture goes to shit when GS work is scheduled.

Jawed

Razor1

12-May-2007, 17:04

well we do know vista drivers still don't give the same performance as xp drivers, so are the numbers that suprising for Dx10 at this point?

No we don't have much info. But Rys did say that load-balancing in this unified architecture goes to shit when GS work is scheduled.

because of drivers, what do drivers have to do with load balancing in the frist place?

nAo

12-May-2007, 17:10

Well, I wouldn't be suprised if drivers can tweak or even modifying load balancing policies, it was possible on more primitive architectures so I don't see why things should change now.
Regarding DX10 performance I'd like to see tested MUCH MORE features than geometry shaders, it's not going to be the end of the world if GS implementation is not uber fast now, there's time for that, as long as devs can start to use it.

dizietsma

12-May-2007, 17:10

Eh?

G80's GS is worse than it should be (half-performance for a null shader :shock: ), theoretically, and you think that NVidia building a unified architecture is somehow surprising to ATI? Apparently the surprise is that it's not working at all well.

Jawed

Well, if it is not surprising to ATi and also if the G80 is not performing as much as your beloved theoretical whumajits would say it should ( at least on a forum and not in the real world ) then it is

EVEN MORE SURPRISING THE R600 cannot whup a G80 after 8 months delay.

trinibwoy

12-May-2007, 17:10

No we don't have much info. But Rys did say that load-balancing in this unified architecture goes to shit when GS work is scheduled.

Jawed

Yeah that's very curious. It will be kinda retarded if G80's "DX10 performance" is hobbled by clunky handling of GS threads and not by underlying shortcomings of the hardware.

PSU-failure

12-May-2007, 17:11

A little bit of knowledge is often worse than no knowledge at all...
I'd like to know where the mistakes are, since you seem to have a great knowledge. :?:

It's easier to say someone is wrong than to explain why.

Oh, or perhaps that's just some fanboyism? Or my english being too frenchie... who knows. :???:

PurplePigeon

12-May-2007, 17:12

The second thing is the number of drivers released in such a short period of time. AND THE CARD IS MONTHS LATE. What would they have been like 8 months ago one has to wonder ?

Just writing as a typical software developer and not someone in any computer graphics related industry, rapid release of drivers at this point in time doesn't really bother me. Many companies try to be "agile" and have a goal of being able to generate stable software builds every day with the latest updates and fixes. It might even been seen as a positive thing that drivers can be released so quickly and rapidly (if based on reviewer feedback, say). Yes, one could also ask why they haven't made those fixes before... But, it's kind of a glass half-full or half-empty perspective.

trinibwoy

12-May-2007, 17:15

EVEN MORE SURPRISING THE R600 cannot whup a G80 after 8 months delay.

Forget the delay. R600's reported performance is far too low given the leaked specs. I'm looking forward to Rys' analysis, hopefully he can shed some light on what's going with the texturing hardware and identify some potential bottlenecks. If R600 really has only 16 bilinear int8 filtering units then it has no more texturing ability than the GTS. Of course, there could be some really malicious software issues at play but there's just so much raw power I just don't see that being a major factor.

Rys

12-May-2007, 17:19

because of drivers, what do drivers have to do with load balancing in the frist place?
The driver can pass down what amounts to a scheduling hint based on what it's about to send the hardware to chew on.

Jawed

12-May-2007, 17:20

Well, I wouldn't be suprised if drivers can tweak or even modifying load balancing policies, it was possible on more primitive architectures so I don't see why things should change now.
I agree. What Rys says implies he considers it a driver fault.

Regarding DX10 performance I'd like to see tested MUCH MORE features than geometry shaders,
I'd just like to see them tested, so far nought. No conception of the theoretical performance of G80's GS architecture and no idea how G80 is dealing with the variety of possibilities there.

it's not going to be the end of the world if GS implementation is not uber fast now, there's time for that, as long as devs can start to use it.
I guess we'll just have to see...

Jawed

trinibwoy

12-May-2007, 17:22

I'd like to know where the mistakes are, since you seem to have a great knowledge. :?: It's easier to say someone is wrong than to explain why.

It's also easier for someone to ask for an explanation than to go research the topic themselves and try to better their own understanding so that they can avoid making such silly statements in the future.

First of all, instruction re-ordering is handled by the shader compiler so your statement there has no merit, at least I've never heard of OOE in a GPU. And even if it did, I fail to see how that would be relevant to an R600 to G80 comparison. Let's assume for kicks that GPU's did OOE in hardware. How is R600's ALU configuration more amenable to this than G80's?

And that bit about storing temporary data for in-flight threads - that's the theme that all modern GPU's are built around. Not even sure how to respond to your comments about the number of instructions or branching since they really make no sense to me. Maybe somebody is willing to take a shot at it for ya.

Rys

12-May-2007, 17:25

Forget the delay. R600's reported performance is far too low given the leaked specs. I'm looking forward to Rys' analysis, hopefully he can shed some light on what's going with the texturing hardware and identify some potential bottlenecks. If R600 really has only 16 bilinear int8 filtering units then it has no more texturing ability than the GTS. Of course, there could be some really malicious software issues at play but there's just so much raw power I just don't see that being a major factor.
We can push peak filter rates (and for more than INT8 surfaces) out of R600 at this point (large and small textures too) with a new tester (w00t!), so it seems the driver and hardware is running freely in that respect, so it would depend on what the app is doing to poop on that somehow.

Don't look forward to it too much! I get nervous around this time that I'm not going to have the time to put everything in there. I've already cut some stuff (which we'll talk about at some point though)......

Which seems to say I should spend less time drinking tea and reading this thread, and more time hacking it up :lol:

trinibwoy

12-May-2007, 17:33

We can push peak filter rates (and for more than INT8 surfaces) out of R600 at this point (large and small textures too) with a new tester (w00t!), so it seems the driver and hardware is running freely in that respect, so it would depend on what the app is doing to poop on that somehow.

Are those peaks high peaks or low peaks? :smile:

Don't look forward to it too much!

Awwwww :cry:

Razor1

12-May-2007, 17:41

The driver can pass down what amounts to a scheduling hint based on what it's about to send the hardware to chew on.

interesting

Which seems to say I should spend less time drinking tea and reading this thread, and more time hacking it up :lol:

LOL

Jawed

12-May-2007, 17:44

It's easier to say someone is wrong than to explain why.

Unfortunately explaining is not simple.

You need to think of time-sliced batch scheduling as the primary mechanism, with the GPU operating on a set of tens or hundreds of batches (per cluster). Instruction parallelism is something that's pretty much hidden and not (in my opinion) relevant to a discussion of load-balancing and overall batch throughput.

This is a good starting point:

http://www.beyond3d.com/content/articles/4/8

Jawed

SugarCoat

12-May-2007, 18:01

I think it's quite telling that there is no real XTX until R650 arrives at .65 in a couple more months. It seems that for whatever reason, AMD doesn't have a top-of-the-range product to go head-to-head with the competition, so they are having to try and push their second string product into top place. This explains the low price, and the less than screaming performance compared to G80.

I think that just like we didn't see the full realisation of R5x0 until R580, so we won't see the full realisation of R6x0 until R650 and I suspect it's much higher clocks. I also think that R650 will come sooner that we expect, because of that obvious gap at the XTX level of the AMD product range that is begging to be filled.

The R580 wasnt a massive leap in real world performance over the X1800XT in any respect. I believe the biggest advantages came at ultra high resolutions but for the most part the R520 was a good strong core and the R580 was simply a refresh of that giving a speed boost. In this case, with the R600, its looking like the problem is with the core itself so if what we're waiting on is substantial clock increases at .65nm then thats not very hopeful considering nVidia can do the same thing.

The R520 also had a really good advantage, to me anyway, in improved high IQ performance, and substantially so over Geforce 7 parts, so if the IQ hasnt been improved yet again to something beyond that of the G80 even if its slower, then this card is a pass to me.

Still got the inhouse tech demos to look forward too!

tEd

12-May-2007, 18:20

All the bench leaks , i'm surprised no driver has been leaked yet

aeryon

12-May-2007, 19:12

All the bench leaks , i'm surprised no driver has been leaked yet

for what ? actually they last no more than 2 or 3 days before a new release comes out :lol:

Razor1

12-May-2007, 19:26

Rys, just had a question, if the threads in flight are reduced when GS is being used, that will effect everything else right? It will become a "systemic" problem?

3dcgi

12-May-2007, 19:34

Hold on, you created a geometry shader that did no work and it still halved performance? :shock:
Unless the driver/compiler can detect that the shader will do no work GS threads must still be run and memory must be allocated. This will be true of both G80 and R600.

Galduta

12-May-2007, 19:59

Very interesting .... R6 Las Vegas ,

R6: Las Vegas, maximun settings

2900XT/8800GTS:
1024x768
min:38/27
med:74/60
max:111/95

1280x960
min:26/18
med:53/42
max:83/68

1600x1200
min:19/13
med:37/30
max:70/48

Rys

12-May-2007, 20:02

Rys, just had a question, if the threads in flight are reduced when GS is being used, that will effect everything else right? It will become a "systemic" problem?
In that it'll affect the performance of the entire chip? It has the potential to, of course, if there aren't enough available threads to keep throughput up.

I think there are also cases (currently, and on G80) where a GS shader can have the thread count increasingly reduced to the point where only a relatively small number of them are running and possibly only on one cluster, because it's doing increased amounts of amplification.

I think worst cases like that are possible on dynamically load-balanced architecture which has fixed resources, though. I haven't tested a shader like that heavily though, nor on R600 yet.

mboeller

12-May-2007, 20:05

From what I have read so far I conclude that the AA of the HD 2900XT is still "broken" because without AA the HD 2900XT is significantly faster than the GTS but with AA the card is slower.

can someone confirm this?

IbaneZ

12-May-2007, 20:13

Very interesting .... R6 Las Vegas ,

Interesting indeed.

Why the hell doesn't R600 kick the living crap out of the 8800 GTS?

Please, someone has to remind the 320 stream processors that's it's showtime. Wakey wakey. :lol:

R300King!

12-May-2007, 20:29

I think this is a HD2900XT here. (http://64.233.179.104/translate_c?hl=en&langpair=de%7Cen&u=http://forums.hardwarezone.com/showthread.php%3Ft%3D1607936%26page%3D6%26pp%3D15)

http://i9.photobucket.com/albums/a52/V2-G/System/Teaser/archmark.jpg

FEAR
http://i9.photobucket.com/albums/a52/V2-G/System/Teaser/fear.jpg

CoH
http://i9.photobucket.com/albums/a52/V2-G/System/Teaser/CoH.jpg

Are these good scores? :D

Galduta

12-May-2007, 20:32

I think this is a HD2900XT here. (http://64.233.179.104/translate_c?hl=en&langpair=de%7Cen&u=http://forums.hardwarezone.com/showthread.php%3Ft%3D1607936%26page%3D6%26pp%3D15)

FEAR
http://i9.photobucket.com/albums/a52/V2-G/System/Teaser/fear.jpg

Are these good scores? :D

Fake :D ? This test is CPU limited

http://www.firingsquad.com/hardware/geforce_8800_gtx_gts_core_2_performance/page5.asp

http://www.firingsquad.com/hardware/intel_core_2_duo_e4300_review/images/fear800.gif

Andrew Lauritzen

12-May-2007, 20:39

D3D10 requires the developer to specify an upper bound on the the number of vertices generated per input vertex. It might be that the driver is playing safe and always assuming the maximum...
I specified a maximum of 3 vertices....

Honestly on one hand I understand that there are a lot of really hard things to do with respect to the geometry shader. It really does break the parallelism and can easily be coded to bring *any* card to a halt. That said, they must have thought that some of the easy cases could be accelerated efficiently or else they wouldn't have added it to the spec (assuming MS isn't just totally off in left field). Still, the current performance of the G80 GS leaves something to be desired, but I'm willing to conceded that it could be partially or entirely driver related at this point.

Still I won't be surprised if geometry shading is another dynamic branching wrt. NV40 vs R520...

R300King!

12-May-2007, 20:53

Fake :D ? This test is CPU limited

http://www.firingsquad.com/hardware/geforce_8800_gtx_gts_core_2_performance/page5.asp

http://www.firingsquad.com/hardware/intel_core_2_duo_e4300_review/images/fear800.gif

It doesn't seem like a joke or anything. He says its on Core2duo @2.4Ghz default. ..so, I donno. We'll see maybe Monday.

w0mbat

12-May-2007, 21:01

single card WR!
http://img46.imageshack.us/img46/4594/30kni1.jpg

tEd

12-May-2007, 21:03

jesus what's with the 3dmark scores , theres alot other stuff to show which would be mucho more interesting.

Skinner

12-May-2007, 21:15

aeryon

12-May-2007, 21:43

jesus what's with the 3dmark scores , theres alot other stuff to show which would be mucho more interesting.

through all time, humans like to compare their penis and see who gets the biggest...

... sorry, me run away :oops:

spidy

12-May-2007, 21:48

sorry, me run away :oops:
What? You run away when it comes to a penis comparison? Ah wait, you're french? That'd be a great excuse :lol:

R300King!

12-May-2007, 21:51

W0mbat, what was the previous single card 05 WR? 8800GTX, yes? What was it's score?

Jawed

12-May-2007, 22:00

That said, they must have thought that some of the easy cases could be accelerated efficiently or else they wouldn't have added it to the spec (assuming MS isn't just totally off in left field).
Wouldn't the vertex-creation cap be a feature of D3D10 requested by the IHVs?

Jawed

vertex_shader

12-May-2007, 22:21

W0mbat, what was the previous single card 05 WR? 8800GTX, yes? What was it's score?

When i good remember 8800ultra, extreme overcloked with LN2 cooling score 296xx point, this was the WR before in 3dmark2k5.

R300King!

12-May-2007, 22:37

When i good remember 8800ultra, extreme overcloked with LN2 cooling score 296xx point, this was the WR before in 3dmark2k5.

Thanks +1 rep for your research. ;)

Kinc also says (http://www.nordichardware.com/forum/next-vt8428.html?postdays=0&postorder=asc&start=25&sid=458c12f772b08c3b1af8ae7fab2a7f58) the HD2900XT can go 1200(core)/1100(mem) and more on LN. Bring on the cold. :)

nicolasb

12-May-2007, 22:51

P35 is also a 8X+8X not 16X+16X.No it isn't. P35 is x16/x4 - the same as P965 (not surprisingly, given that neither is intended for the high performance market). Proper CF support will hopefully come with the X38 version of Bearlake, which is rumoured to be x16/x16. (Although that won't be out till Q3).

trinibwoy

12-May-2007, 23:21

Thanks +1 rep for your research. ;)

Heeeeeey, don't bring that rep announcing thing over here. It's cheesy :p

Twinkie

12-May-2007, 23:30

Someone at rage3d already has the card, installed and ready to go.

http://www.rage3d.com/board/showthread.php?t=33890816

Bought it for$485 :roll:

Sound_Card

12-May-2007, 23:37

Someone at rage3d already has the card, installed and ready to go.

http://www.rage3d.com/board/showthread.php?t=33890816

Bought it for$485 :roll:

Whats with the eye roll? Have a problem with someone paying for something they want?:roll:

See, I gave one right back.:razz:

I'm impressed by his results. He's running Vista and one of the weaker Core2's(E6300), and still pulls a comfortable 10k+ in 3dmark 06.

Twinkie

12-May-2007, 23:39

The specs of the R600 which is pretty much confirmed.
[Moderator EDIT] Removed direct link to scan from magazine - a few lines may be "fair use" but an entire page is probably not. Please just quote your reference source[/EDIT]

R600

80nm
700million transistors
742 mhz Core
828 mhz memory
320 SP
16 TMUs
16 ROPs
512bit bus
512mb GDDR3
215W?
No HD video engine??

Twinkie

12-May-2007, 23:42

Whats with the eye roll? Have a problem with someone paying for something they want?:roll:

See, I gave one right back.:razz:

I'm impressed by his results. He's running Vista and one of the weaker Core2's(E6300), and still pulls a comfortable 10k+ in 3dmark 06.

He bought it for $485.. well im not sure about people at rage3d but i dont just throw away cash like that. He could have bought it for $399~$425 easily. At the price he paid for, GTX wouldve been a better buy.

But i cant say much since rage3d is a fansite for ATi. :grin:

grats to him, and ill be keeping an eye on that thread for real world game benches.

R300King!

12-May-2007, 23:56

Heeeeeey, don't bring that rep announcing thing over here. It's cheesy :p

Hey, I like cheese, I'll have it with my crackers. hehe ;)

fellix

13-May-2007, 00:05

Hmm, I wonder -- those mediocre hi-res/AA scores from R600 -- could the be due to lack of proper frame-buffer optimization. I mean... it looks to me, the successfully applied read/write sequencer adjustment to the old 256 (8*32-bit) interface is no more doing good here, now. Maybe lots of empty burst cycles are wasted. :roll:

trinibwoy

13-May-2007, 00:27

Russell

13-May-2007, 00:45

Yeah it's getting brained when AA is applied. Now we have to wait and see if it's an issue inherent to the hardware or if it can be addressed with a driver update.

R300King, now you got me craving crackers and cheese and I have neither :lol:

I'm optimistic that the AA problems are driver-related. It really makes no sense for it to be anything but. At least so I hope.

Fornowagain

13-May-2007, 00:46

Review, found by Topgear

http://extremeoverclock.org/his-hd-2900xt-review.html#more-307

http://img206.imageshack.us/img206/8607/6082642006ocsz5.jpg (http://imageshack.us)

Sound_Card

13-May-2007, 01:05

http://rage3d.com/board/showpost.php?p=1334937886&postcount=57

BECCO demo's us the new AA modes from R600.:wink:

Geeforcer

13-May-2007, 01:12

Am I the only one who can't see the difference between the three shots? I am not looking at the hands (which are at slightly different angles) but at all other edges?

Mcmlxxx IV

13-May-2007, 01:22

Am I the only one who can't see the difference between the three shots? I am not looking at the hands (which are at slightly different angles) but at all other edges?

I don't see much (any?) difference either. That is Doom 3, I think, and it is possible the new AA modes are not activated for OGL yet. Didn't it take ATI a while to put HQAF in for OGL (or was that AAA)?

Kaotik

13-May-2007, 01:25

Am I the only one who can't see the difference between the three shots? I am not looking at the hands (which are at slightly different angles) but at all other edges?

You did realize to click on the "click here for full size" thing, right?

Fornowagain

13-May-2007, 01:28

Am I the only one who can't see the difference between the three shots? I am not looking at the hands (which are at slightly different angles) but at all other edges?
I was just cycling through them on a HD projector 96" image and I can't see any major difference. Maybe the res is too low?

Geeforcer

13-May-2007, 01:31

You did realize to click on the "click here for full size" thing, right?

Right. Can't see the the difference.

Moloch

13-May-2007, 01:33

Well open all 3 images in 3 different tabs and go back and forth... there is no diff at all that I can see, not even the crates.

dnavas

13-May-2007, 01:36

You did realize to click on the "click here for full size" thing, right?

Well, I don't see any diffs either, but, I'm not the best judge of these things.
I do wonder why it says "16xHQAF" and then the "High Quality" box isn't checked (1st image only).

I'm pretty impressed by the higher number of math units in R600. I was rather expecting more from the G80 in that respect, so I'm happy to see ATI put some pressure on NV there. Not sure what to think of these AA modes yet, though. I hope we get a Chris Ray - style article about them at some point!

-Dave

Mcmlxxx IV

13-May-2007, 01:41

The new AA modes are probably not in for OGL yet.

Fornowagain

13-May-2007, 01:46

Well open all 3 images in 3 different tabs and go back and forth... there is no diff at all that I can see, not even the crates.
[/URL][URL="http://img183.imageshack.us/my.php?image=image1bt1.jpg"]http://img183.imageshack.us/img183/3734/image1bt1.th.jpg (http://imageshack.us)

SugarCoat

13-May-2007, 01:59

Even if they are working Doom 3 is problably about as terrible a game to pick to show off AA as you can get short of a game that doesnt have AA at all. Should of used BF2 or HL2 or Oblivion etc...

Why is it every time one of these people who gets the cards early pops up they have NO GAMES installed. I swear its a conspiracy. If that was me you'd see a clustered desktop with about twenty game shortcuts.

Arun

13-May-2007, 02:03

BRiT

13-May-2007, 02:10

Why is it every time one of these people who gets the cards early pops up they have NO GAMES installed. I swear its a conspiracy. If that was me you'd see a clustered desktop with about twenty game shortcuts.

Perhaps they're organized and use a "Games" toolbar ? I dont know why anyone would have any shortcuts on their desktop at all.

I do agree, what is it with those who get the cards before they're officially out and yet have no games at all? Perhaps they saved up all their money to buy the card so they couldn't buy any games?

^eMpTy^

13-May-2007, 02:25

The images are identical...even zoomed in you can't see so much as a single pixel different except for the hands being off a little:

http://www.gpureview.com/stuff/iq/iq.php

Geeforcer

13-May-2007, 02:28

Some history
...
Huddy says R600 (http://www.beyond3d.com/forum/showthread.php?t=31049)
...

You know, I was just re-reading this (year old) thread. Few things that struck me were few "OMG it was in front of us all along!" moments on the very first page. In addition, I was impressed how many things Fudo eventually got right (admittedly, after 2-6 tries): 700-750MHz clock, 64 unified shaders, 2007 launch, etc.

btwango

13-May-2007, 03:19

In addition, I was impressed how many things Fudo eventually got right (admittedly, after 2-6 tries): 700-750MHz clock, 64 unified shaders, 2007 launch, etc.

Infinite monkey theorem

Fornowagain

13-May-2007, 03:48

Borrowed these two images from here (http://www.rage3d.com/board/showpost.php?p=1334937899&postcount=61). Same settings under the single frame 3dmark06 rendering options. Jpeg compression makes them more than a bit limited, but thought I'd put them up anyway. NV drivers 158.16

Both: 3DMark06 1280x1024 8xAA 16xAF

2900XT left. 8800GTX right.

http://img120.imageshack.us/img120/271/xt1ae8.th.jpg (http://img120.imageshack.us/my.php?image=xt1ae8.jpg)http://img297.imageshack.us/img297/3573/gtx1xv2.th.jpg (http://img297.imageshack.us/my.php?image=gtx1xv2.jpg)

http://img153.imageshack.us/img153/2991/xt2xw4.th.jpg (http://img153.imageshack.us/my.php?image=xt2xw4.jpg)http://img120.imageshack.us/img120/1743/gtx2br1.th.jpg (http://img120.imageshack.us/my.php?image=gtx2br1.jpg)

Unknown Soldier

13-May-2007, 03:56

Is it my imagination or is the GTX image sharper?

You can see it on the dragon and the Airship.

Also the cables(under the sun) seem shaper on the GTX in the snow benchie.

US

Skrying

13-May-2007, 03:57

In those the HD 2900XT handles AA much better than the GTX.

Geeforcer

13-May-2007, 04:00

Humm. The dragon is blurry yet the hangar roof in the bottom right is "noisy".

Banko

13-May-2007, 04:02

In those the HD 2900XT handles AA much better than the GTX.
It seems to be blurring the image more then anything.

Unknown Soldier

13-May-2007, 04:07

Ye.. the HD2900XT image is blurry and the GTX is sharper.

US

Twinkie

13-May-2007, 04:11

Is it my imagination or is the GTX image sharper?

You can see it on the dragon and the Airship.

Also the cables(under the sun) seem shaper on the GTX in the snow benchie.

US

Thats what i'm seeing too. On the airship, you can see the lines on the GTX shot while that gets blurred out on the XT shot.

silent_guy

13-May-2007, 04:15

Is it my imagination or is the GTX image sharper?
You can see it on the dragon and the Airship.
I don't think it's your imagination: the skin of the dragon loses a lot of detail. Ouch.

Also the cables(under the sun) seem shaper on the GTX in the snow benchie.
For cables and for edges, the blurring obviously removes the harsh straircasing. Not bad. But in my opinion, it doesn't weigh against the loss of sharpness.

Geeforcer

13-May-2007, 04:21

What the heck is going on here though???

http://forum.beyond3d.com/attachment.php?attachmentid=263&d=1179026450

Skrying

13-May-2007, 04:31

The HD 2900XT image is slightly blurrier. Its really hard to tell, compression is REALLY destroying the images in that instance to me.

ninjaspa

13-May-2007, 04:32

they look identical to me

ninjaspa

13-May-2007, 04:42

they look identical to me

on second look, in the dragon comparisons it looks like the nose pointer on the blimp in the xt shot is sharper than in the gtx shot. the blimp canvas itself is sharper too. aside from that i think the biggest difference is that the gtx files are about 3 and 5 times larger than the xt files. i mean, they're just jpegs so it beats me

Martin Eddy

13-May-2007, 04:42

they look identical to me

Good, I'm not the only one that can't see a damn difference.

LLB

13-May-2007, 04:43

I agree with Geeforcer, the roof in the lower right corner of the arctic screen shot is much cleaner on the GTX than the 2900XT.

LLB

caffeinated

13-May-2007, 04:45

This single thread constitutes approximately 3% of all posts on this forum.

I've been reading this thread now for 2 days trying to catch up with what is going on after a few weeks not staying up with it. I figure by about midnight on Sunday I'll probably be caught up enough to be in sync with the release - provided there isn't some burst of new information which causes a slew of posts and puts me even further behind. Sigh. I'm on page 129, 40 posts per page.

Don't mind me. :lol:

Skrying

13-May-2007, 04:47

on second look, in the dragon comparisons it looks like the nose pointer on the blimp in the xt shot is sharper than in the gtx shot. the blimp canvas itself is sharper too. aside from that i think the biggest difference is that the gtx files are about 3 and 5 times larger than the xt files. i mean, they're just jpegs so it beats me

Look at the edges on the dragon, the XT is clean the GTX jagged. Same on the snow one, look at the wire on the far left, clean on XT jagged on GTX. The pole towards the left in the snow one, XT is clean once again jagged on GTX.

It is also very weird that the file size is much greater on the GTX. Which means the XT could just be blurrier and actually not be any better. Which could explain the clean edges and slightly blurry dragon for the XT...

Sound_Card

13-May-2007, 04:53

HD 2900XT performance in STALKER. 1280x1020

http://i162.photobucket.com/albums/t246/becco222/2_fraps_log.jpg

http://i162.photobucket.com/albums/t246/becco222/1_option.jpg

http://i162.photobucket.com/albums/t246/becco222/6_stalker.jpg

I hate BECCO so much right now...:cry:

ninjaspa

13-May-2007, 04:55

http://img.photobucket.com/albums/v194/jasperdelarge/blimpcomp.jpg

idk, ultimately we're still drawing on jpegs, and yeah up close, particularly on the xt in this instance the compressions is dirtier, yet overall appears sharper. who's to say which is better honestly, idk. do you call the xt sharper, or the gtx more refined?

Silent_Buddha

13-May-2007, 05:06

If the R600 picture is 3-5 times smaller, wouldn't that imply that there is more JPG compression? Which in turn might indicate that even if actual rendered screen is identical the R600 picture would still be slightly blurrier?

This is why I hate comparing shots taken by one guy in forum to shots taken by another guy in a forum.

As to the shots, the one on the R600 looks ever so slightly blurrier in some places, yet on the other hand appears to show more detail in other places. The nose of the blimp for example.

I'm just going to file it as interesting to watch, and still waiting for real reviews and comparisons.

Regards,
SB

Unknown Soldier

13-May-2007, 05:13

ah yes, I see the GTX is 622KB while the XT pics is 123KB .. so it could account for the loss.

US

ninjaspa

13-May-2007, 05:15

yeah, it might just help if we had some small bitmap comparisons. i'm sure the professional analysts will have this, but we're all tired of waiting on this arbitrary nda to lift and having to resort to all manner of enthusiast benching which we appreciate all the same.
anywho, about this nda ending on the 14th, can we expect sites to update at midnight eastern time or what? anyone know which official tech sites stay most current?

Geeforcer

13-May-2007, 05:19

As to the shots, the one on the R600 looks ever so slightly blurrier in some places, yet on the other hand appears to show more detail in other places. The nose of the blimp for example.

I don't know if it's "ever so slightly":

http://img143.imageshack.us/img143/7688/gtxxtlw8.th.jpg (http://img143.imageshack.us/my.php?image=gtxxtlw8.jpg)

The XT shot is messed up, probably mostly by compression - you can see the jpeg artifacts.

ChrisRay

13-May-2007, 05:35

I think people may end up finding that the blur is not just a compression thing.

Pete

13-May-2007, 05:46

Is this the tent filter in action? I notice BECCA had it enabled in earlier sshots, but he didn't say one way or the other with the 3DM06 ones.

tEd

13-May-2007, 05:51

Looks very blurred , which i'm a little surprised. Though being no expert shouldn't the correct weightening of the samples with a tent filter minimze the blur effect.

Silent_Buddha

13-May-2007, 06:02

I would imagine a wide tent would have a blurrier effect than the narrow tent. And thus far I haven't actually seen a shot of the 12x setting with the narrow tent. He supposedly did that with the Doom 3 screens, but honestly I see no difference in those so I'm rather doubting it's actually being applied in Doom 3.

There have been shots of the 16x setting with the wide tent and that does tend to be noticably blurrier. Enough so that if that actually proves to the be case in actual review coming up, it wouldn't be a setting I'd probably use.

Regards,
SB

Colourless

13-May-2007, 06:03

Ow, the XT images are really badly jpeg artifacted and the GTX ones aren't. It's not a fair comparison at all.

ninjaspa

13-May-2007, 07:31

everyone look at the file sizes. one is over 600kb and the other is over just 100k

Shtal

13-May-2007, 07:47

So far HD-2900XT disappointed me, especially 6 months late and no answer to G80-GTX.
R600XT 320 stream processors could not win over 128 stream on G80-GTX.
Hmmm.... probably because 16TMU's and 16ROP's

Last hope driver improvement! :)

Sc4freak

13-May-2007, 08:09

I know JPEG artifacts when I see them. The 2900XT shot, when zoomed in, is far blurrier due to massive JPEG artifacting. The reason you don't notice it that much when zoomed out is because that's what JPEG is designed to do.

rwolf

13-May-2007, 08:12

I think people may end up finding that the blur is not just a compression thing.

:rolls eyes:

chavvdarrr

13-May-2007, 08:57

offtopic, but it would be great if there was a way to see what JPEG quality setting was used when saving a shot :(
Wild guess: 80 for XT shot and 90-100 for 8800

vertex_shader

13-May-2007, 09:12

offtopic, but it would be great if there was a way to see what JPEG quality setting was used when saving a shot :(
Wild guess: 80 for XT shot and 90-100 for 8800

I try it, original .bmp size 3.932.216 byte, than i convert to .jpg, quality 65% file size 126.013 byte (xt picture size 126.125byte), quality 98% file size 635.662 byte (gtx picture size 643.841byte).

Dalton Sleeper

13-May-2007, 09:48

The last pic must be more compressed, you clearly see those squares on XT, not a single one on GTX.

satein

13-May-2007, 10:06

I would like to ask about this explanation made by Kinc on Nordichardware.com as he posted here...
http://www.nordichardware.com/forum/-vp21040.html#21040 (http://www.nordichardware.com/forum/-vp21040.html#21040)

This is what he said...

Im sorry to say that there wont be any faster R600 card then the HD 2900XT. Since ATI didnt know until to late that their 512bit interface and 1024bit ringbus did not work well with GDDR4. 1100MHz GDDR4 is about the same bandwidth as 830MHz GDDR3. The bandwidth at stock is over 100GB/s wich allready is more then enough so there is no need to worry about the bandwidth :wink:

It sounds as if the 512bit interface and 1024bit internal ringbus cannot be ultilized at its potential with the GDDR4. Here it is my assumption, might it be possible that the GDDR4 performance get satuarated?

I am just a noob reading on this board and thus having got no good basic about how memory and bus will be ultilized. But I just would like to ask if this might be a reason happened that even AMD/ATi themself haven't expected before hand (yes, GDDR4 spec would sound awlsome in the paper anyway)...

Regards,

bigtabs

13-May-2007, 10:06

So, we may find that the 1GB version that we think will arrive in June/July will have GDDR3? Interesting.

Is there a particular time B3D is aiming for to get their first piece out on the R600?

BlizzardOne

13-May-2007, 10:43

1100MHz GDDR4 is about the same bandwidth as 830MHz GDDR3.

:???: since when has the 'generation' of memory made a difference in bandwidth at a given speed..?

Galduta

13-May-2007, 10:49

HL2: LOST COAST FAA@4x , FA@16x , HDR@ful l 8,37

2900XT/8800GTS:
1600x1200
med:130/126

1920x1200
med:113/112

Galduta

13-May-2007, 10:53

Splinter Cell: Double Agent AA@4x in game , FA@16x in CCC.

And this tweak

X:\Program files\Ubisoft\Tom Clancy's Splinter Cell Double Agent\SCDA-Offline\System

[Graphics]
bEnableHDR=True
bTrilinearFiltering=True
bSpecularLighting=True
bVerticalSync=False
bHighQualityVisualFX=True

[WinDrv.WindowsClient]
WindowedViewportX=1920
WindowedViewportY=1080
FullscreenViewportX=1920
FullscreenViewportY=1080
TextureDetail=UltraHigh
SkinDetail=UltraHigh
Brightness=0.000000
Contrast=0.000000
Gamma=0.000000
MipFactor=1.000000
UseJoystick=True
CaptureMouse=True
StartupFullscreen=True
ScreenFlashes=True
NoLighting=False
DeadZoneXYZ=True
DeadZoneRUV=False
InvertVertical=False
ScaleXYZ=1.000000
ScaleRUV=1.000000
MinDesiredFrameRate=10.000000
Decals=True
NoDynamicLights=False
SkinDetail=UltraHigh
TextureDetail=UltraHigh
NoFractalAnim=False
Decals=True

This route

http://www.mediafire.com/?cnymd5mdihd

Splinter Cell: Double Agent AA@4x in game , FA@16x in CCC.

2900XT/8800GTS:
1280x1024
min:24/--
med:38/--
max:57/--

1600x1200
min:18/--
med:28/--
max:42/

no AA, FA@16x in CCC , all maximun

1600x1200
min:40/--
med:57/--:shock:
max:87/--

And the 8800 GTX and 1900 XTX in this route ..

http://i4.photobucket.com/albums/y117/jonelo/sc4i1600.jpg

IbaneZ

13-May-2007, 10:57

Nice, keep the gaming benchies coming. Tired of 3dmark by now. :wink:

Only 27 hours left. :shock:

Love_In_Rio

13-May-2007, 11:00

It seems the bus bandwith is 128 bits instead 512...

Arnold Beckenbauer

13-May-2007, 11:05

HL2: LOST COAST FAA@4x , FA@16x , HDR@ful l 8,37

2900XT/8800GTS:
1600x1200
med:130/126

1920x1200
med:113/112

HL2: LC? Is it the integrated VST (video stress test) or custom timedemo? If VST, there is (unfortunetly) no HDR lightning a la VALVe.

Galduta

13-May-2007, 11:14

HL2: LC? Is it the integrated VST (video stress test) or custom timedemo? If VST, there is (unfortunetly) no HDR lightning a la VALVe.

Is the HardwareOC Benchtools Bench , the 1.6 version (the 2.0 not have lost coast time demo )

edit / with HardwareOC Benchtools Bench 2.2 version (http://www.hocbench.com/hl2.html)

HL2: LOST COAST & HardwareOC Tool v2.2v NO AA , NO AF HDR@full y all máximun.
2900XT/8800GTS:
1600x1200
med:137/--

1920x1200
med:125/--

HL2: LOST COAST & HardwareOC Tool v2.2 AA@4x y FA@16x, HDR@full y all máximun
2900XT/8800GTS:
1600x1200
med:76/--:huh:

1920x1200
med:64/--:huh:

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

Oblivion 8,37

OBLIVION, all maximun , FAA@4x y FA@16x , HDR

-2900XT/8800GTS:

Forest

1920x1200
12/15

1600x1200
16/18

1280x1024
19/23

This route

http://img120.imageshack.us/img120/7038/ov1080ugncustombw9.jpg

http://i4.photobucket.com/albums/y117/jonelo/ov1650.jpg

Forest . night , raining

1920x1200
10/10

1600x1200
13/12

1280x1024
16/15

Razor1

13-May-2007, 12:06

Looks very blurred , which i'm a little surprised. Though being no expert shouldn't the correct weightening of the samples with a tent filter minimze the blur effect.

its not that easy :wink: , wieghting isn't an accurate process unless done by hand, I can't even imagine a smart alogrithm, how complex it would have to be to really work well.

Yes the XT does blur, anything above x8 has a slight to large blur effect.

trinibwoy

13-May-2007, 12:19

This is going to be interesting to watch. If I understand the process correctly this blurring should be easily noticeable with lower screen resolutions and/or higher-resolution textures where neighbouring pixels can have sufficiently different colors that any blending could really throw things off. I've never used Quincunx but I vaguely remember there being a camp that stood by it and others who denounced it as crap. Something tells me history will repeat itself :)

Anybody post CSAA vs CFAA shots yet?

ChrisRay

13-May-2007, 12:45

K.I.L.L.E.R was the only one I remember liking quincunx ;)

seahawk

13-May-2007, 12:50

X2900XT looks better nad better with real game benches. Oblivion is most likely just a driver bug. Given its price point the card is one sweet offer.

Fornowagain

13-May-2007, 12:56

I have the original bitmaps, put them here, hopefully the XT owner will do the same.

http://www.rage3d.com/board/showthre...3890816&page=3 (http://www.rage3d.com/board/showthread.php?t=33890816&page=3)

http://www.sharebigfile.com/file/167947/gtx1-bmp.html
http://www.sharebigfile.com/file/167952/gtx2-bmp.html

Anybody know a better host?

Or png

http://img50.imageshack.us/img50/4821/gtx1wt7.th.png (http://img50.imageshack.us/my.php?image=gtx1wt7.png)
http://img520.imageshack.us/img520/8040/gtx2sp4.th.png (http://img520.imageshack.us/my.php?image=gtx2sp4.png)

Fornowagain

13-May-2007, 13:38

2900XT users guide.

http://rapidshare.com/files/31038555/Radeon_HD2900_Users_Guide.pdf.html

http://img153.imageshack.us/img153/2116/image1vl7.jpg (http://imageshack.us)

One port for the HDMI.

IbaneZ

13-May-2007, 13:50

X2900XT looks better nad better with real game benches. Oblivion is most likely just a driver bug. Given its price point the card is one sweet offer.

Yep. The only thing that worries me is the heat and noise rumour. :???:

http://www.xtremesystems.org/forums/attachment.php?attachmentid=58965&stc=1&d=1179052867

:lol:

Hands off!

Julidz

13-May-2007, 14:17

True ?? by Fudzilla

Radeon HD 2900XT user guide revealed
Written by Fuad Abazovic
Sunday, 13 May 2007 10:17

70 pages to confirm a 750W minimum

Two weeks ago we told you that AMD recommends a 750W but that the Radeon HD 2900XT will work on a high end 400 to 500W as well.

As AMD is a legal company they said that with any 750W the card should work as many cheap 500W will fail to power the card. You can read the original story here.

In this seventy pages document you can read about the power connectors and pay attention on the page 6 as AMD suggests you to upgrade the AGP drivers before you install this AGP card. Well done.

And the document is here.

Kaotik

13-May-2007, 14:19

Julidz

13-May-2007, 14:26

Could be true, as in, even the worst 750W's can do it - for a fact I do know that 430W Antec NeoHE is enough for HD2900, 4x HDD and A64 X2 6000+ though.

where did u see that ?

IbaneZ

13-May-2007, 14:37

for a fact I do know that 430W Antec NeoHE is enough for HD2900, 4x HDD and A64 X2 6000+ though.

That's awesome. There's still hope for my old friend then, the 480W Hiper. :smile:

Kaotik

13-May-2007, 14:46

where did u see that ?

Posted already earlier in this thread, I don't know if either has posted it in english anywhere, but they did on biggest finnish board

Arnold Beckenbauer

13-May-2007, 14:51

2900XT users guide.

http://rapidshare.com/files/31038555/Radeon_HD2900_Users_Guide.pdf.html

http://img153.imageshack.us/img153/2116/image1vl7.jpg (http://imageshack.us)

One port for the HDMI.

Page 51:
ATI CrossFire™ Requirements
The basic requirements of an ATI Radeon™ HD 2900 Series ATI
CrossFire™ system are:
• Two ATI Radeon™ HD 2900 Series native CrossFire™
graphics cards
• Two internal ATI CrossFire™ bridge interconnects
• An ATI CrossFire™-ready motherboard
• We recommend a 750-watt power supply or better with two 2x3-
pin and two 2x4-pin PCIe™ power connectors.
For an up-to-date listing of ATI CrossFire™ certified motherboards,
see
ati.amd.com/products/certification
For an up-to-date listing of certified power supplies, see:
ati.amd.com/CertifiedPSU
Installing ATI CrossFire™ Graphics Cards
The following set of instructions assumes that one ATI Radeon™ HD
2900 Series graphics card has already been successfully installed in the
primary slot (slot zero), and that the secondary slot (slot one) is empty. If
you have not already done so, see the Getting Started Guide that came
with the graphics card for installation instructions.
http://img512.imageshack.us/img512/7630/blindln9.th.jpg (http://img512.imageshack.us/my.php?image=blindln9.jpg)

Kaotik

13-May-2007, 15:02

Ah, so that 750W recommendation is for crossfire, not single card

ninjaspa

13-May-2007, 15:09

what about the amp requirements? do they even say? i think one of the main problems with just recommending a psu in tech requirements for gpu manufacturers is that they only suggest wattage and not amperage. this is why there is the whole good 400w'er better than a lousy 600w'er argument.

from what i've read nvidia recommends 30amps on 12V to comfortably run the 8800gtx.

of course, amps output is only 1 of the variables in psu fidelity [like efficiency] but at least it draws a more accurate picture for requirements. i'm no psu expert, but i know enough that wattage listings are increasingly vague in the over saturated market of power supply vendors.

and at this rate, it makes me wonder what nvidia's W recommendation was for their 8 series.

INKster

13-May-2007, 15:15

and at this rate, it makes me wonder what nvidia's W recommendation was for their 8 series.

8800 GTS - 26 amps, 450W
8800 GTX - 30 amps, 500W

vertex_shader

13-May-2007, 15:16

Ah, so that 750W recommendation is for crossfire, not single card

Fruitzilla forget 2 weeks ago write its recommended for CF, another fruitzilla blow :lol:

Julidz

13-May-2007, 15:16

so , another fake from fudzilla as they say that the guide shows a 750w minimum for a single card

Razor1

13-May-2007, 15:29

Check page 1 (page 9 of the pdf)

It didn't mention crossfire on that page

• AMD Athlon® or Intel® Pentium® 4

• 512MB of system memory; 1GB or more for best
performance.
• Optical drive for installation software (CD-ROM or
DVD-ROM drive).
• A 750 Watt or better power supply with two 2x3-pin
PCIe™ power connectors. For optimal performance,
we recommend requires one 2x3-pin and one 2x4-pin
PCIe™ power connector. For a list of certified power
supplies, please see ati.amd.com/certifiedpsu

Also looks exactly like the screenshot Fuad took too, so he didn't read the entire manual or he did it for pure sensationalism, I pick the later.

Unknown Soldier

13-May-2007, 15:32

thegrommit

13-May-2007, 15:35

what about the amp requirements? do they even say? i think one of the main problems with just recommending a psu in tech requirements for gpu manufacturers is that they only suggest wattage and not amperage. this is why there is the whole good 400w'er better than a lousy 600w'er argument.

from what i've read nvidia recommends 30amps on 12V to comfortably run the 8800gtx.

Assuming the user manual is correct, a single card requires one 2x3 + one 2x4 PCIe power connector, so total 12V draw would be 75W+100W+75W (from the motherboard) = 250W / 21A.

vertex_shader

13-May-2007, 15:37

Vrzone broke the NDA as usual :wink:
Link (http://www.vr-zone.com/?i=4946&s=1)

Unknown Soldier

13-May-2007, 15:40

Assuming the user manual is correct, a single card requires one 2x3 + one 2x4 PCIe power connector, so total 12V draw would be 75W+100W+75W (from the motherboard) = 250W / 21A.

Read again:

• A 750 Watt or better power supply with two 2x3-pin PCIe™ power connectors. For optimal performance, we recommend requires one 2x3-pin and one 2x4-pin PCIe™ power connector.

That means you can use two 2x3-pin PCIe™ power connectors. Only for optimal performance should you use one 2x3-pin and one 2x4-pin PCIe™ power connector.

US

tEd

13-May-2007, 15:45

Vrzone broke the NDA as usual :wink:
Link (http://www.vr-zone.com/?i=4946&s=1)

Maybe they put the review on a server with a 56kbit connection on purpose!?

Unknown Soldier

13-May-2007, 15:49

either that or it's taking huge hits. ;)

US

INKster

13-May-2007, 15:50

lol ... no one pick up that he said AGP card? :D

Fuad talking FUD(or should that be crap) again.

I think he was being sarcastic. ;)

Unknown Soldier

13-May-2007, 15:52

2900XT users guide.

http://rapidshare.com/files/31038555/Radeon_HD2900_Users_Guide.pdf.html

I have a feeling that the guide is doctored.

I don't think it's real.

US

Unknown Soldier

13-May-2007, 15:55

I think he was being sarcastic. ;)

Nope, the guide clearly says that the AGP motherboard drivers need to be updated(pg14-16).

I think Fuad fell for a fake ATI 2900XT Guide. I mean the "guide" even has 6 year old Intel 440BX/LX mobo chipsets. Comeon, I think it's definitely fake.. or if it's real then AMD need to get real quickly.

US

IbaneZ

13-May-2007, 15:55

Vrzone broke the NDA as usual :wink:
Link (http://www.vr-zone.com/?i=4946&s=1)

Told you so. :lol:

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

Too bad it's so damn slow at the moment. :mad:

Rebel44

13-May-2007, 15:58

Maybe they put the review on a server with a 56kbit connection on purpose!?

Iagree - 2 minutes to load every page is ridiculous:evil:

tEd

13-May-2007, 16:00

Iagree - 2 minutes to load every page is ridiculous:evil:

Well you're lucky then, you get it loaded at least

Rebel44

13-May-2007, 16:02

Well you're lucky then, you get it loaded at least

well it was with only 1 picture out of 5 or 6:wink:

Unknown Soldier

13-May-2007, 16:04

I got them all. :) Number 44(HD2400) looks weird(haven't seen this one before).

http://img114.imageshack.us/img114/5893/44sr6.th.jpg (http://img114.imageshack.us/my.php?image=44sr6.jpg)

Not to mention 45

http://img53.imageshack.us/img53/5186/45zd2.th.jpg (http://img53.imageshack.us/my.php?image=45zd2.jpg)

US

Julidz

13-May-2007, 16:09

do u know where i can get or execute the CubeMapGS ??

vertex_shader

13-May-2007, 16:10

Whats wrong with the card? i see nothing weird :smile:

Coz

13-May-2007, 16:10

http://www.vr-zone.com/?i=4946&s=9

Power Consumption of Total System Idle/Load

PowerColor HD X2900XT..........245/365 watts
Inno3D 8800GTX 575/900MHz..........250/348 watts
ASUS EN8800GTS 640MB 513/792MHz..........240/300 watts
EVGA 8800GTS 320MB Superclocked 576/850MHz..........235/276 watts
ASUS EN1950XTX..........195/325 watts

Unknown Soldier

13-May-2007, 16:12

Whats wrong with the card? i see nothing weird :smile:

Sorry, had 46 in there .. meant 44. The black cover is weird and I hadn't seen this card(as in the pic) in other previews before.

I've see 45, but it looks just damn ugly. :)

US

thegrommit

13-May-2007, 16:15

Read again:

That means you can use two 2x3-pin PCIe™ power connectors. Only for optimal performance should you use one 2x3-pin and one 2x4-pin PCIe™ power connector.

US

Read again (http://forum.beyond3d.com/showpost.php?p=985860&postcount=5411). A single card only requires one 2x3 + one 2x4 connector. The images on the vr-zone review clearly show this.

Galduta

13-May-2007, 16:18

S.T.A.L.K.E.R. v1.3, All setting maximun, AF in game . 8.37 Drivers

2900XT//8800GTS
bench1
1280x1024
med:~47fps~44fps
1600x1200
med:~39fps~35fps
1920x1200
med:~34fps~30fps

bench2
1280x1024
med:~73fps~64fps
1600x1200
med:~58fps~52fps
1920x1200
med:~52fps~44fps

An improvement of a 20% in this game with the 8.37

vertex_shader

13-May-2007, 16:20

Unknown Soldier

13-May-2007, 16:21

Read again (http://forum.beyond3d.com/showpost.php?p=985860&postcount=5411). A single card only requires one 2x3 + one 2x4 connector. The images on the vr-zone review clearly show this.

The card implements a 6-pin PCIE power connector with another 8-pin power connector to feed the card. A minimum of 2 x 6-pin power connector is needed to run the card, similar to rival NVIDIA's 8800GTX.

http://www.vr-zone.com/?i=4946&s=6

It's been known for ages that the card will run on 2x6-pin connectors only.

If you want to OVERCLOCK, then you NEED 1x6-pin and 1x8-pin.

US

Julidz

13-May-2007, 16:24

do u know where i can get or execute the CubeMapGS ??

:?:

IbaneZ

13-May-2007, 16:26

Taking a setting at 1600x1200 with 16xAF, I saw a major increase in performance, particularly Company Of Heroes and Quake 4. Performance went up by 11% on COH and 42% on Quake 4! This shows that the drivers is still very raw on this card, with just a minor driver revision boosting up performance that much, it gives us quite a lot of hope for a fair bit of improvement to come. Let's hope for that!

In many non Anti-Aliasing, High Definition game settings, you have seen the X2900XT push ahead of the performance of it's closest competitor, the GeForce 8800GTS 640MB, sometimes by quite a large margin, sometimes falling behind or ahead by a small percentage. In a select few games, the GTS is slightly faster, and vice versa. When Anti-Aliasing is turned on, the X2900XT showed that it carries it off with great efficiency in games that the drivers are optimized for, performing significantly better than the GTS; while the AA efficiency is piss-poor in some games due to the raw driver which has not fully blossomed to take advantage of ATi's new GPU technology. Just take a look at how performance has boosted from Drivers 8.36 to 8.37, that shows the potential in performance growth... a whole lot of it to reap.

It is slightly off tradition that the GPU company's flagship product sails off not to meet the flagship of it's competitor, but one target lower. Then again, the lower we go down the price pyramid, the bigger the audience, more people with the budget to spend. I'd say that there is no clear winner between the 8800 GTS and X2900XT, the GTS displayed more consistent performance behavior while the X2900XT fluctuates around due to the in-matured driver. I would say that despite the heat thrown out by the GPU, the X2900XT overclocks better than the 8800GTS by 8-10%, but that's putting out a lot more heat and drawing more power than it already consumes. So this is something potential XT buyers should take note of, the heat produced by the card is no small amount, nor is the power consumed by it - more than 60w over the GTS. What you would be investing in is a higher potential of upcoming performance boosts (including the latest pre-Alpha 8.37.4.2_47323 Catalyst just released 3 days before this review) and full HDCP support with integrated audio controller. And of course the new programmable Tessellation technology which we will probably not see support in games until much later.

Not the fastest video card in the market for sure, but definitely holds it's own at it's current price-point. We only hope that supply will be adequate and not lead to an indirect increase in prices due to short supply. We hope to see some interesting implementations from various card partners as well, be it overclocked specifications, or improved coolers.

Shaminos conclusion, if you don't want to sit around and wait for the page to load. :smile:

The drivers seem to need more work, but other than that, the card looks like a pretty fun toy.

INKster

13-May-2007, 16:28

Still, far from a GTX killer (let alone an Ultra).
It will all come down to a price-war between the GTS and the XT, just like i suspected.
"Very nice". :D

Jawed

13-May-2007, 16:29

:?:
D3D10 SDK I think.

Jawed

Unknown Soldier

13-May-2007, 16:31

That was expected from what we've seen in the last few weeks.

US

thegrommit

13-May-2007, 16:37

http://www.vr-zone.com/?i=4946&s=6

It's been known for ages that the card will run on 2x6-pin connectors only.

If you want to OVERCLOCK, then you NEED 1x6-pin and 1x8-pin.

US

Ahh, I see your point now. I was basing my post on the manual and the vr-zone review.

Regardless, given the huge disparity in power draw, I think it's time to pick up a overclocked GTS 320.

Kaotik

13-May-2007, 16:37

http://www.vr-zone.com/?i=4946&s=6

It's been known for ages that the card will run on 2x6-pin connectors only.

If you want to OVERCLOCK, then you NEED 1x6-pin and 1x8-pin.

US

Nope, if you want to use Overdrive you need 1x6pin + 1x8pin, you can still use other software (like the clocktool AMD released) to overclock the card even if you had 2x 6pin

bigtabs

13-May-2007, 16:43

From a guy who has one (http://www.rage3d.com/board/showpost.php?p=1334938307&postcount=81):

the card has 6pin and 8pin "slot"
- default: you should plug 2x6-pins (yup, 6-pin cable from PSU CAN be plugged to R600's 8pin "slot" .... it's OK and Normal)
- OCing: 1x6-pin + 1x8-pin

satein

13-May-2007, 16:45

Ahh, I see your point now. I was basing my post on the manual and the vr-zone review.

Regardless, given the huge disparity in power draw, I think it's time to pick up a overclocked GTS 320.

I dont' think HD2900xt draw much power than that of 8800GTS (640) according to that VR-Zone testing... it is a bit different by 5w at default setting.... His conclusion sounds "off" by talking that it will consume more power than GTS at general use (not by overclocking the card).

Anyway, GTS 320 would be a bit lower power draw as it is 320MB memory on the card (yes, the power should draw less than that of GTS 640 too)...

Razor1

13-May-2007, 16:51

I dont' think HD2900xt draw much power than that of 8800GTS (640) according to that VR-Zone testing... it is a bit different by 5w at default setting.... His conclusion sounds "off" by talking that it will consume more power than GTS at general use (not by overclocking the card).

Anyway, GTS 320 would be a bit lower power draw as it is 320MB memory on the card (yes, the power should draw less than that of GTS 640 too)...

Under load the xt is 65 watts more then the gts and 20 watts more then the gtx.

Think ya looked at the wrong column :wink:

Fornowagain

13-May-2007, 16:52

If you want to OVERCLOCK, then you NEED 1x6-pin and 1x8-pin.

US
Not quite, if you want to overclock with the Overdrive you need the 8 pin. Other apps still work.

ChrisRay

13-May-2007, 17:08

http://www.tweaktown.com/articles/1100/ati_radeon_hd_2900_xt_amd_s_long_awaited_r600_dx10 _gpu_arrives/index.html

IbaneZ

13-May-2007, 17:11

http://www.tweaktown.com/articles/1100/ati_radeon_hd_2900_xt_amd_s_long_awaited_r600_dx10 _gpu_arrives/index.html

Wohoo! :grin:

Rys

13-May-2007, 17:19

http://www.tweaktown.com/articles/1100/ati_radeon_hd_2900_xt_amd_s_long_awaited_r600_dx10 _gpu_arrives/index.html
Sweet!

Rebel44

13-May-2007, 17:24

http://www.tweaktown.com/articles/1100/ati_radeon_hd_2900_xt_amd_s_long_awaited_r600_dx10 _gpu_arrives/index.html

Looks like nobody cares about NDA:lol: :twisted:

trinibwoy

13-May-2007, 17:27

That vr-zone battlefield shot shows the best evidence of CFAA blurring thus far IMO.

http://www.vr-zone.com/?i=4946&s=16

silent_guy

13-May-2007, 17:29

Still, far from a GTX killer (let alone an Ultra).
It will all come down to a price-war between the GTS and the XT, just like i suspected.
"Very nice". :D
Nice for the customers, indeed. Not so nice for AMD: margins on this things must be (once again) way below expectations.

Brian118

13-May-2007, 17:29

Does anyone think that even with the most optimized drivers, the HD 2900xt can catch up with the 8800gtx? I feel like I just wasted 6 months waiting for this stupid card just to find out that it is super hot, takes up more power than the gtx, and recommends a new power supply to overclock it, which means I would probably have to get new cooling also. So once I have my XT on par with the GTX, I have already spent the amount of money it costs just to buy a GTX.

IbaneZ

13-May-2007, 17:30

That vr-zone battlefield shot shows the best evidence of CFAA blurring thus far IMO.

http://www.vr-zone.com/?i=4946&s=16

NV winning the IQ test.. Oh boy, the world has gone crazy. :razz:

spidy

13-May-2007, 17:35

That battlefield image quality comparison is pretty disastrous for the 2900XT. Far too blurry to be called "a good Anti-Aliasing".

Sound_Card

13-May-2007, 17:36

The latest 8.37.4.2_47323 drivers is supposed to implement a new intelligent algorithm that increases FPS while applying similar image quality when running Adaptive Anti-Aliasing. In Oblivion, performance several times faster than previous drivers using the new adaptive AA algorithm was claimed to have been acheived. New optimizations for HDR applications in general resulted in a 5-30% increase in performance.

The 8.37.4.2_47323 is actually a pre-alpha driver, but it includes a preview of new 12xAA and 24xAA modes. These modes use an advanced edge detection filter that delivers edge quality while eliminating blurring.

Inresting.

ChrisRay

13-May-2007, 17:37

Too bad none go too deep into AA IQ yet. I guess we are looking at the bad boy NDA breakers at this point.

Chris

Arty

13-May-2007, 17:39

Galduta

13-May-2007, 17:41

Does anyone think that even with the most optimized drivers, the HD 2900xt can catch up with the 8800gtx? .

In R6 Las Vegas and SC DA is better ..;) . The 2900 have a big bug/problem with Af /AA

http://images.tweaktown.com/imagebank/radeonhd2900xtreview_test010.gif

AA 4X free :mrgreen: EDIT / is the 8.361 driver but is the CD driver

Geeforcer

13-May-2007, 17:44

Quick comments about a few things.

Did TweakTown mention hardware problems as a reason for delay or is it just my imagination?
Generally underperforms it specs... I guess we'll have a while to figure out what's going on. In the mean time, I have to wonder what the margins on that thing are.
OC picutres that some people were passing off as "air and stock volts!" are actually cascade cooler + voltmod.
Blurvision 2.0! Yey!?

Brian118

13-May-2007, 17:44

But even with the physical hardware, is it possible for the XT to be at least SOMEWHAT near the GTX?

Sobek

13-May-2007, 17:44

From a guy who has one (http://www.rage3d.com/board/showpost.php?p=1334938307&postcount=81):

Hmm...does it matter how the second 6 pin goes into the 8 pin slot? ie; any particular side the spare two pins should be on?

Or does it just not matter?

*edit* or can it only really go in one way in particular, like an ATX connection.

Rebel44

13-May-2007, 17:47

I can understand AMD postponing the launch because of hardware related problems but they've had months to work on the drivers and yet the drivers look so unfinished that its not even funny. In some cases its slower than the X1950 .. :|

Yeah I also expected that drivers will be much better:???: - now I want comparison in Vista 64 because thats my OS

Sound_Card

13-May-2007, 17:47

From VR zone.

Well, it does seem that there is an Image Quality Improvement on the X2900XT cards as compared to previous generation X1950 cards. The X2900 outshines the GeForce 8800 in Anti-Aliasing Quality slightly, while the Anisotropic Filtering on the 8800 still seems to be a little more accurate compared to the X2900XT.

Tables seem to have flipped compared to last gen.

Jawed

13-May-2007, 17:51

I can understand AMD postponing the launch because of hardware related problems but they've had months to work on the drivers and yet the drivers look so unfinished that its not even funny. In some cases its slower than the X1950 .. :|
It looks like people who buy HD2xxx GPUs will have a long wait ahead of them for decent gaming performance. The shortfalls in performance are shocking really.

Then there's the worry of the delays caused by the official Catalyst schedule as it causes months of lead-time for features/bug-fixes.

Jawed

Geeforcer

13-May-2007, 17:53

But even with the physical hardware, is it possible for the XT to be at least SOMEWHAT near the GTX?

On paper, looking only at superficial specs, XT would seem more capable in several areas, no?

320x750 Vs 128x1.35
86Gb/s vs 102 Gb/s

Etc.

IbaneZ

13-May-2007, 18:07

http://www.ocworkbench.com/2007/asus/EAH2900XT/g1.htm

Asus 2900XT review.

Fornowagain

13-May-2007, 18:08

Another one (http://www.ocworkbench.com/2007/asus/EAH2900XT/g1.htm), oops /\ beat my by a few seconds

http://img486.imageshack.us/img486/7082/004copywx8.jpg (http://imageshack.us)
http://img486.imageshack.us/img486/1125/005copyta9.jpg (http://imageshack.us)

http://img113.imageshack.us/img113/3954/3dmark06ml4.jpg (http://imageshack.us)
http://img514.imageshack.us/img514/5131/fearlg7.jpg (http://imageshack.us)

Conclusion

ASUS has delivered the first graphics card based on the reference design of the ATi HD RADEON 2900XT. Although they look alike, the package has more in house applications bundled and the game Stalker is given as part of the package.

In terms of performance, this card is run on a standard 965P board and found to be performing well in areas of synthetic benchmarks. When it comes to real games, running them with AA/AF at high resolutions, the card gave some better results when compared to the NV's 8800GTX. This is probably due to a 512 bits memory controller instead of 384bits on 8800GTX.

On the other hand, this card requires a better PSU to power it. ATI will be putting up a recommended list of PSU on its site. As it is based on a 80nm manufacturing process, the GPU is quite hot. Although we are unable to detect the actualy GPU temperature, measurements made around the card in the previous page shows that it is estimated to be around 50-60 deg on load.

For the HDTV buffs, this card simplifies the cabling or requirement of an addtional graphics card for sound output. The bundled DVI-HDMI connector is able to deliver the audio through the DVI-HDMI interface to your LCD TV.

During our test, DOOM3 failed to run with driver 8.37, we have to go back to ver 8.361 provided to get it working. Drivers are most likely to be in the works to rectify some bugs with 3Dmark05, 3Dmark06 and some games. We would expect the final driver to get all these fixed.

Overall, the EAH2900XT should most likely be priced around the price range of USD 399 (for OEM cards). That is cheaper than the 8800GTX. If you are one that goes for the best image quality, the ASUS EAH2900XT based on Radeon HD 2900XT would be a good choice not only for your 3D games at high resolutions, but also for HDTV guys who wants the ease to playback HD DVD and Blu Ray in the near future.

ChrisRay

13-May-2007, 18:08

VR-Zone mirror.

http://homepage.ntlworld.com/gerald.marley/index.htm

*Edit* Delete this as you see fit. If their review site wasnt lagging like crazy ><

turtle

13-May-2007, 18:09

On paper, looking only at superficial specs, XT would seem more capable in several areas, no?

320x750 Vs 128x1.35
86Gb/s vs 102 Gb/s

Etc.

It's really 256/384 depending on how you look at it. The MUL devoted to special function I suppose makes the 256 number more accurate, although who knows how much of the 320 is used for SF.

jamis

13-May-2007, 18:16

On paper, looking only at superficial specs, XT would seem more capable in several areas, no?

This part caught my eye on VR-zone:
The Vertex Shading capability of the X2900XT on the other hand is ridiculously strong. Here you see it even beating the GTX by 76%! More than 200% the performance of the X1950XTX.

http://www.vr-zone.com/?i=4946&s=14

It's possibly wasting much more of it's shader units on vertex shading compared to G80 when running DX9. (Or anything < DX10)

Headstone

13-May-2007, 18:19

There do seem to be some good jumps between the 8.361 and the 8.37 drivers based on the 2 different reviews (the latter 2). Does anyone here know if the 8.38 will be even that much better? If so then the perfomance will be much closer to the GTX speeds but at GTS pricing.

Dave Baumann

13-May-2007, 18:21

It's possibly wasting much more of it's shader units on vertex shading compared to G80 when running DX9. (Or anything < DX10)

I'm sorry, but :lol:

:razz:

trinibwoy

13-May-2007, 18:23

It's possibly wasting much more of it's shader units on vertex shading compared to G80 when running DX9. (Or anything < DX10)

The synthetic vertex shader tests are reflecting what we expect given the relative shader power of both cards. Note that vertex data probably fits very nicely with R600's vec5 configuration and without bandwidth/texturing/scheduling to worry about it probably gets pretty good utilization.

So in this particular case, it's simply performing as it should :)

And although Dave has already laughed at you, it might be worth noting that these architectures will balance the workload of vertex/pixel/geometry shaders so what you're suggesting is probably not happening :) It's not even possible since you'll quickly run out of vertices to shade anyway!

ninjaspa

13-May-2007, 18:23

seems to win as much as it loses against the gts. i'll take that as a fair trade-off considering the drivers have marked room for improvement in just going from 8.36 - 8.37.
aint .38 supposed to be out? i mean it's obvious when it loses, it loses bad like when it falls short of last gen ati card. if this isnt a driver issue, i wouldn't know what is.
meanwhile, the upside is when it beats the gts there are enough instances which it is closer to gtx performance than anything else.
at the moment [and i'd still like to see better reviews], it looks like a deal to me. hopefully providing they continue to tweak the drivers like they did for quake 4 results, it leans toward steal for me since i def want that black box and i could definitely use the sound output for my hdtv.

jamis

13-May-2007, 18:24

The synthetic vertex shader tests are reflecting what we expect given the relative shader power of both cards. Note that vertex data probably fits very nicely with R600's vec5 configuration and without bandwidth/texturing/scheduling to worry about it probably gets pretty good utilization.

So in this particular case, it's simply performing as it should :)
In that case, no hopes for improvements then with DX10.:grin:

trinibwoy

13-May-2007, 18:25

In that case, no hopes for improvements then with DX10.:grin:

Why not? The current performance issues can be due to a lot of things. I'm not sure how 3dmark's vertex test sheds any light on the situation at all.

jamis

13-May-2007, 18:31

I'm not sure how 3dmark's vertex test sheds any light on the situation at all.
Well I was just thinking that as they are unified architectures and running in fixed mode in DX9 with no load balancing, so maybe R600 has a larger junk of it's shader units assigned for vertex shading than G80 when doing DX9. So in that case things might be very different with DX10 and load balancing.
But if not, well...

trinibwoy

13-May-2007, 18:32

Well I was just thinking that as they are unified architectures and running in fixed mode in DX9 with no load balancing, so maybe R600 has a larger junk of it's shader units assigned for vertex shading than G80 when doing DX9. So in that case things might be very different with DX10 and load balancing.
But if not, well...

Unified hardware doesn't run in fixed mode in DX9. That myth was dispelled months ago :smile: Nothing about DX9 forces the hardware to assign fixed ratios of pixel and vertex shaders.

jamis

13-May-2007, 18:33

Unified hardware doesn't run in fixed mode in DX9. That myth was dispelled months ago :smile: Nothing about DX9 forces the hardware to assign fixed ratios of pixel and vertex shaders.
LOL, havn't been following that carefully.:lol:

Edit:
Note that vertex data probably fits very nicely with R600's vec5
So it does very well in vertex and possibly geometry shading but just sucks at pixel shading then?

Geeforcer

13-May-2007, 18:42

HWZ: With the merger, it is understandable that certain products have been realigned, but could we have an update on the R600 delays?

Henri Richard: The R600 will be out in the second quarter. The reason we decided to delay the launch was that we wanted to have a complete DX10-enabled solution top to bottom. A lot of people wrote that the reason it is delayed is because of a problem with the silicon, but there is no problem with the silicon. We are demonstrating it. We can ship it today. But if you think about it, looking at where the market is at, the volumes are going to be in the R610 and R630, so it makes sense for us to do a one time launch of the entire family of DX10 enabled products. That meant delaying the R600 for a few weeks, but frankly it doesn't make a difference in the life cycle of the product and talking with our customers and partners, they felt that it would make a bigger impact with one full launch. So we decided to do that.

Also increasingly in particular with Vista, as we've seen with the competition. It doesn't matter if you're shipping the silicon if the drivers are not stable. There is nothing more frustrating than having bought a new graphics card and having your system crash repeatedly because the drivers are not ready. Although we today, even by Microsoft standards, have the best and most stable drivers in the entire industry. The few weeks will give us even more time to continue improving the drivers. Again, the decision lies in the fact that we will have a top to bottom DX10 offering with drivers that will have a very very high level of stability and the only difference is a few weeks. So it seems to make a lot of sense to do it that way.

So... yeah.

trinibwoy

13-May-2007, 18:45

So it does very well in vertex and possibly geometry shading but just sucks at pixel shading then?

Lol, no we have no evidence that it sucks in pixel shading.

trinibwoy

13-May-2007, 18:51

Unless the driver/compiler can detect that the shader will do no work GS threads must still be run and memory must be allocated. This will be true of both G80 and R600.

So this could just be a register pressure issue? Hmmmm reminds of a certain ill-fated architecture.....

ChrisRay

13-May-2007, 18:53

This part caught my eye on VR-zone:

http://www.vr-zone.com/?i=4946&s=14

It's possibly wasting much more of it's shader units on vertex shading compared to G80 when running DX9. (Or anything < DX10)

Err shouldnt that be dynamically loaded?

jamis

13-May-2007, 19:00

Err shouldnt that be dynamically loaded?
You are late, it's allready over and done with.:lol:

Silent_Buddha

13-May-2007, 19:08

Just a few things I noted that bug me from the VR-Zone piece.

When they tested Anistropic Filtering, for some reason they decided to also enable "best" AA modes. Problem is, that 16x CFAA uses a wide tent that blurs the image. Thus, they aren't really judging Anistropic Filtering at all, but are judging something totally different.

So at this point I still haven't a clue how R600's AF compares to G80's AF. Grrrr. Hopefully reviews on Monday will do a better job.

Likewise, it sounds like 12x CFAA (8x + Narrow Tent) and 24x CFAA (8x + Edge Detect) are enabled in 8.38. For someone like me, and I'd assume most others on this board, those have the greatest potential of bringing improved AA quality to the table.

24x CFAA sounds the most intriguing. What exactly does ATI mean by edge detect. I can't imagine they mean something as simple as polygon edges, as that would just how current AA is done on both vendors cards, right?

So does this mean it actually has an algorythm that analyzes a texture or entire image and find instances where there are lines/edges and only applies a tent filter to those case. And otherwise doesn't apply narrow or wide tent to the rest of the image/textures?

So in effect it would be trying to anti-alias anything that appears to be a line or an edge on the screen even if it isn't an actual edge of an object. While leaving everything else as it is. If so, this sounds very intriguing. However, I shudder to think of the performance impact something like this would have.

Regards,
SB

nicolasb

13-May-2007, 19:25

24x CFAA sounds the most intriguing. What exactly does ATI mean by edge detect. I can't imagine they mean something as simple as polygon edges, as that would just how current AA is done on both vendors cards, right?

So does this mean it actually has an algorythm that analyzes a texture or entire image and find instances where there are lines/edges and only applies a tent filter to those case. And otherwise doesn't apply narrow or wide tent to the rest of the image/textures?Maybe the AA is angle-dependent? Aliasing is particularly noticeable on nearly-horizontal or nearly-vertical lines, and much less noticeable at 45 degrees. So perhaps if a polygon edge is close to vertical or horizontal it turns up the AA, and then turns it back down for diagonal lines?

Techno+

13-May-2007, 19:34

theinq has just posted the nvidia document comparing G80 to R600

Link (http://www.theinquirer.net/default.aspx?article=39578)

Natoma

13-May-2007, 19:39

So... yeah.

He didn't say anything about performance. Only stability. Two different things. :wink:

Bouncing Zabaglione Bros.

13-May-2007, 19:43

theinq has just posted the nvidia document comparing G80 to R600

Link (http://www.theinquirer.net/default.aspx?article=39578)

Be interesting to see the AMD rebuttal that was sent out a few days ago.

Fornowagain

13-May-2007, 19:44

Part three

http://it-review.net/index.php?option=com_content&task=view&id=1335&Itemid=1

bdmosky

13-May-2007, 19:51

Honestly though, that article Nvidia is sending out isn't nearly as brutal as they've done in the past.

AnarchX

13-May-2007, 19:52

Nvidia's FUD presentation revealed
http://www.theinquirer.net/default.aspx?article=39578

:lol:

ups.. I am to late, but the post by Techno+ was not good visible.

Nebula

13-May-2007, 19:57

Nvidia's FUD presentation revealed
http://www.theinquirer.net/default.aspx?article=39578

:lol:

ups.. I am to late, but the post by Techno+ was not good visible.

Interesting and quite honest to. The only bizzare thing was the Inq writers comments! :lol:

RobertR1

13-May-2007, 20:06

So far:

Nvidia has better AF
Nvidia has better AA (sharper picture > less edges). Ati blurring the pic doesn't make it better IQ for me.
The 8800GTX remains untouched
The 8800GTS will trade back n' forth with the x2900xt
All hope lies on Drivers from Ati.

From what I've seen so far, I'd rate this product a 6/10. Late, worse IQ and does not clearly beat the competitor at the same price point, let alone their flagship card.

Before I get called a nvidia fanboy, my video card purchases since I got into the enthusiast PC hobby include: x800xtpe, x1800xtx, x1900xtx (currently).

If there were a lots of DX10 games coming out in the new few months, I'd say that'd be a wildcard but by the time DX10 games do become the norm, these cards will be atlest 1-2 generations behind.