8800GTX Shadermark results

[oeLangOetan]

Doesnt make sense you'd have to adjust Cat AI for that to work at optimal, unless it wasnt using AFR before and that forced it on. Seems to be acting like the Doom 3 Engine shader replacement/memory optimization to me. Even a couple of the single card tests got a noticable kick.

Cat AI on high foces AFR and texture filtering optimisations.
I see no reason to believe there are shader replacements going on. In normal games the single card tests show bigger gains with CatAI on high.

 

[ChrisRay]

Doesnt make sense you'd have to adjust Cat AI for that to work at optimal, unless it wasnt using AFR before and that forced it on. Seems to be acting like the Doom 3 Engine shader replacement/memory optimization to me. Even a couple of the single card tests got a noticable kick.

It is my understanding that setting Cat A.I too "on" when using Crossfire turns on their profile system for AFR rendering. Some of those Crossfire scores are increased by over 130%, and I doubt that has anything to do with shader replacement.

 

[Skinner]

Some of those Crossfire scores are increased by over 130%, and I doubt that has anything to do with shader replacement.

Well, just for the fun, change "FEARXP" to "FEAR" with a single card (ATIXxxx), you well be amazed

 

[russo121]

Everyone's saying the GPU is limited by the CPU, but, is the nvidia driver written to take advantage of the quad core from intel (or even dualcore)?

 

[BlizzardOne]

Everyone's saying the GPU is limited by the CPU, but, is the nvidia driver written to take advantage of the quad core from intel (or even dualcore)?

Didn't nVidia write dual-core 'optimizations' or something into their drivers a number of moons ago? Not sure if it was just dual-core, or multi-core though.. perhaps somebody else remembers better

And as an aside, does anyone know how much impact, if any, the CPU will have on the ShaderMark results?

 

[fellix]

Nvidia is using multi-threaded code only for the OGL path in their driver, AFAIK -- it offloads the driver calls to a separate thread, which isn't a big issue as the driver overhead for this API is much less than D3D, anyway.

On the topic, the good and not so old R580 is still showing some respectable numbers in here.
But just wait to see the SLi-ed GTX graphzilla numbers.

 

[russo121]

Nvidia is using multi-threaded code only for the OGL path in their driver, AFAIK -- it offloads the driver calls to a separate thread.

On the topic, the good and not so old R580 is still showing some respectable numbers in here. But just wait to see the dual GTX graphzilla.

Can we say going into multicore will not help in DX? Only increasing clock speed?

 

[KimB]

No, multicore helps in D3D too. Just not as much right now. nVidia's current drivers are optimized for multicore in both D3D and OpenGL.

 

[fellix]

Glad to read that [for D3D] -- now all it's left is to feel it... somehow.

 

[KimB]

Glad to read that [for D3D] -- now all it's left is to feel it... somehow.

Well, if you want to try, you can always disable the optimizations and see what the difference is. Instructions are in the driver release notes, on page 25 in the PDF for the latest version (93.71).

 

[alexsok]

So, for the laymen among us, how do the results rank against the latest-and-greatest of current-gen cards?

 

[Pete]

Check BlizzardOne's handy chart a page back. You can't miss it, just look for the colors that aren't shades of blue.

 

[Moloch]

On the topic, the good and not so old R580 is still showing some respectable numbers in here.
But just wait to see the SLi-ed GTX graphzilla numbers.

Hardly, the 8800gtx is like close to twice as fast .
Only looks bad for a per clock deal, as isnt the 8800gtx supposed to have like 128 alus or something like that?

 

[swaaye]

I wonder if these coming cards will have ATI and NV on a similar level of shader performance. R580 was stupidly ahead of G71. Sure to be fun!

 

[KimB]

I wonder if these coming cards will have ATI and NV on a similar level of shader performance. R580 was stupidly ahead of G71. Sure to be fun!

It was? Where? I only remember it being significantly ahead when dealing with floating-point textures and dynamic branching. I don't think it was ahead that much in general shader performance.

 

[trinibwoy]

I wonder if these coming cards will have ATI and NV on a similar level of shader performance. R580 was stupidly ahead of G71. Sure to be fun!

Not according to the B3D Rightmark numbers. Most people understandably bought into the "triple the shader powar!!" hype but G71 held its own typical shader usage. R580's threading and latency hiding mechanisms probably contributed significantly to performance.

 

[swaaye]

Well reading the GPGPU stuff convinced me that G71 was not R580's equal at all in shader math. But hey, I guess that could be as you said and that its advantages are just those specific tasks. Its advantages didn't really show up much in games, that's for sure.

On the other hand, hopefully we'll see something more potent at the mid-range than the sad RV530 and disappointing RV560.

 

[KimB]

GPGPU is typically heavy on floating point I/O, and that really hurt nVidia's performance. It doesn't look like an issue on the G8x now, though.

 

[Geeforcer]

Mike over at [H]orums has added Archmark and Fillrate tester results:

Fillrate Tester
--------------------------
Display adapter: NVIDIA GeForce 8800 GTX
Driver version: 6.14.10.9689
Display mode: 1600x1200 R5G6B5 65Hz
Z-Buffer format: D24S8
--------------------------

FFP - Pure fillrate - 11342.501953M pixels/sec
FFP - Z pixel rate - 70649.945313M pixels/sec
FFP - Single texture - 11223.582031M pixels/sec
FFP - Dual texture - 8979.156250M pixels/sec
FFP - Triple texture - 5981.024414M pixels/sec
FFP - Quad texture - 4480.708008M pixels/sec
PS 1.1 - Simple - 11297.077148M pixels/sec
PS 1.4 - Simple - 11317.381836M pixels/sec
PS 2.0 - Simple - 11324.293945M pixels/sec
PS 2.0 PP - Simple - 11293.700195M pixels/sec
PS 2.0 - Longer - 9287.584961M pixels/sec
PS 2.0 PP - Longer - 9293.107422M pixels/sec
PS 2.0 - Longer 4 Registers - 9481.595703M pixels/sec
PS 2.0 PP - Longer 4 Registers - 9472.769531M pixels/sec
PS 2.0 - Per Pixel Lighting - 3815.257813M pixels/sec
PS 2.0 PP - Per Pixel Lighting - 3798.813232M pixels/sec

Archmark

ArchMark 0.50
Driver GeForce 8800 GTX/PCI/SSE2/3DNOW! v2.1.0
Resolution 1024x768 @ 59.97Hz
Method Flush

Fillrate
--32 bits---------------------------------------
Mode R8G8B8A8 Z24 S8
Col 13.719 GPix/s
Z 69.236 GPix/s
ColZ 13.718 GPix/s
ZPassColZ 13.381 GPix/s
ZCullLEqual 142.161 GPix/s
ZCullGEqual 80.114 GPix/s
ZCullEqual 142.202 GPix/s
S 54.562 GPix/s
SCull 54.521 GPix/s
----stencil test passed-------------------------
S 37.430 GPix/s
ZFailS 29.489 GPix/s
------z test passed (LEQUAL)--------------------
S 35.988 GPix/s
ZS 29.503 GPix/s
Col 13.630 GPix/s
ColZ 12.900 GPix/s
ColS 11.800 GPix/s
ColZS 11.103 GPix/s



--16 bits---------------------------------------
Error no pixel format


Bandwidth
Mode R8G8B8A8 Z24 S8
--available to buffer clears--------------------
All 119.774 GB/s
Color 82.869 GB/s
ZAndStencil 212.239 GB/s
Z 119.279 GB/s
Stencil 81.311 GB/s

Draw 101.835 GB/s
BurnedByRAMDAC 188.663 MB/s
Physical 102.024 GB/s

Geometry
Mode R5G6B5A0 Z0 S0
--Plain vertices--------------------------------
Fan 549.612 MTris/s
List 191.424 MTris/s
Clip 191.423 MTris/s

--Vertex shading speed--------------------------
LightD1 153.106 MTris/s
LightP1 153.103 MTris/s
LightP8 153.103 MTris/s


Texturing
Mode R5G6B5A0 Z0 S0
--Textured fillrate-----------------------------
----Bilinear filter-----------------------------
1 10.406 GPix/s
2 7.152 GPix/s
3 5.611 GPix/s
4 4.548 GPix/s

----Trilinear filter----------------------------
1 10.469 GPix/s
2 7.186 GPix/s
3 5.638 GPix/s
4 4.571 GPix/s



Readback
Mode R8G8B8A8 Z24 S8
--Whole buffer----------------------------------
R8G8B8A8 190.578 MPix/s
B8G8R8A8 336.463 MPix/s
R8G8B8 218.928 MPix/s
B8G8R8 227.864 MPix/s
Zuint 192.135 MPix/s
Zfloat 195.067 MPix/s
S8 349.963 MPix/s

--32x32 region----------------------------------
R8G8B8A8 56.190 MPix/s
B8G8R8A8 62.124 MPix/s
R8G8B8 56.581 MPix/s
B8G8R8 57.295 MPix/s
Zuint 57.156 MPix/s
Zfloat 58.188 MPix/s
S8 59.711 MPix/s


Texture cache
Mode R8G8B8A8 Z24 S8
RGBA 256B
DXT5 128 kiB

Tiling
Mode R8G8B8A8 Z24 S8
--preferred block alignment---------------------
----updating all buffers------------------------
Width 8
Height 2

----in color buffer-----------------------------
Width 8
Height 2

----in depth buffer-----------------------------
Width 16
Height 4

----in stencil buffer---------------------------
Width 16
Height 4

 

[Geeforcer]

For the reference, R520/R580/G70/G71 results.