FX and PS 1.4, DX9 tests?

[Doomtrooper]

I would say:

NVIDIA developed the NV30 with long/complex shaders in mind. Unoptimized drivers can kill even the fastest hardware, the driver team is in my opionion at least as important as the hardware guys, just because of the complexity of the drivers. So I would say give them a month (at least) or until the first commercial DX9 app is out (3d-mark 2003 / AquaMark) to optimize their drivers and then compare the two cards again...

Thomas

I would prefer your benchmark Thomas, optimizations to synthetic benchmarks never relate to game peformance improvements...look at the Detonator Database..almost all games flat...3Dmark + 30%.

 

[tb]

Okay, I just did a test with ShaderMark 1.6a on my 9700 Pro on a system very close to Brents. (AXP 2700+ vs AXP 2800+)

System specs:

Asus A7N8X (nforce2)
AXP 2700+ (166fsb)
DCDDR333 512MB
WinXP SP1
Cat 3.0a

Edit. I deleted the results: Not all the tests were displayed correctly - or at all. But since the numbers seem in line with others it looks like a bug. Thomas, any ideas?

Could you post some screenshots? Have you copied all files and folders from ShaderMark into the Treasure Chest directory ? Something overclocked?

 

[noko]

I didn't have that problem, all test ran without any image problems.

 

[LeStoffer]

Okay, I just did a test with ShaderMark 1.6a on my 9700 Pro on a system very close to Brents. (AXP 2700+ vs AXP 2800+)

System specs:

Asus A7N8X (nforce2)
AXP 2700+ (166fsb)
DCDDR333 512MB
WinXP SP1
Cat 3.0a

Edit. I deleted the results: Not all the tests were displayed correctly - or at all. But since the numbers seem in line with others it looks like a bug. Thomas, any ideas?

Could you post some screenshots? Have you copied all files and folders from ShaderMark into the Treasure Chest directory ? Something overclocked?

Odd thing is that the DX8 works without problems. But with the DX9 tests (both PS 1.1-1.4 and PS 2.0) the screen is just go blank at some of the tests. (Things are copied the way they should and nothing is overclocked).

Here is the textfile from the DX9 PS 2.0 test:

ShaderMark v1.6a - DX9 2.0 Pixel Shader Benchmark - ToMMTi-Systems (http://www.tommti-systems.com)

video mode / device info
(1024x768), X8R8G8B8 (D24X8)
HAL (pure hw vp): RADEON 9700/9500 SERIES
benchmark info
mip filter reflections: DX9

shaders:

919.70 fps

shaders:

920.35 fps

shaders:

920.44 fps

shaders:

919.50 fps

shaders:
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
228.75 fps

shaders:
PS 2.0 - Bumped Diffuse + Specular

383.16 fps

shaders:
PS 2.0 - Bumped Diffuse + Specular
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
208.56 fps

shaders:
PS 2.0 - Bumped Diffuse and Specular Lighting with per pixel Specular Exponent

422.89 fps

shaders:
PS 2.0 - Bumped Diffuse and Specular Lighting with per pixel Specular Exponent
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
219.97 fps

shaders:
PS 2.0 - Per Pixel Anisotropic Lighting

524.09 fps

shaders:
PS 2.0 - Per Pixel Anisotropic Lighting
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
245.17 fps

shaders:
PS 2.0 - Per Pixel Bumped Anisotropic Lighting plus Diffuse

366.59 fps

shaders:
PS 2.0 - Per Pixel Bumped Anisotropic Lighting plus Diffuse
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
203.26 fps

shaders:
PS 2.0 - Cubic Environment Bumped Reflections

259.81 fps

shaders:
PS 2.0 - Cubic Environment Bumped Reflections
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
167.82 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Independently Colored Reflections

145.36 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Independently Colored Reflections
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
109.69 fps

shaders:
PS 2.0 - Ghost Shader

340.87 fps

shaders:
PS 2.0 - Ghost Shader
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
159.70 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Tinted Reflections with per pixel Fresnel Term

280.02 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Tinted Reflections with per pixel Fresnel Term
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
174.24 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Independently Colored Reflections

257.68 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Independently Colored Reflections
PS 2.0 - Bumped Diffuse Lighting with per pixel intensity falloff
165.12 fps

shaders:
PS 2.0 - 4 Lights/Pass Diffuse Bump Mapping
PS 2.0 - 4 Lights/Pass Diffuse Bump Mapping
100.35 fps

shaders:
PS 2.0 - 2 Spot Lights
PS 2.0 - 2 Spot Lights
78.57 fps

shaders:
PS 2.0 - Cubic Environment Bumped Diffuse and Independently Colored Reflections
PS 2.0 - 4 Lights/Pass Diffuse Bump Mapping
99.99 fps

shaders:
PS 2.0 - Cubic Environment Diffuse Light and Tinted Refractions
PS 2.0 - 4 Lights/Pass Diffuse Bump Mapping
106.97 fps

 

[tb]

Hmm, seems like not all shaders could be found/compiled at runtime. Could be a dx9 issue, bacause dx8 work fine. Have you installed the final dx9 release? (try dxdiag) Have you enabled the debug runtime in the directx control panel ?

Thomas

 

[LeStoffer]

Have you installed the final dx9 release? (try dxdiag) Have you enabled the debug runtime in the directx control panel ?

Thomas

Final release, yes, but where do I enable/disable the debug runtime? (I have no directx control panel) Do I need to install the full DX9 SDK? (ATI's DX9 demos runs great BTW).

 

[Reverend]

I would prefer your benchmark Thomas, optimizations to synthetic benchmarks never relate to game peformance improvements...look at the Detonator Database..almost all games flat...3Dmark + 30%.

Sometimes, though, we need to study the optimizations. If it is in shading performance, it is very relevant (and interesting) at this site. Some here aren't interested in "3DMarks" but the specific 3DMark tests. Using games as a point of interest in participating in some sites may not be worth someone's while if they're only ever interested in available games. Beyond3D goes beyond interest in currently available games.

 

[Reverend]

Joe and Russ, please ignore each others' posts as often as you can.

 

[Joe DeFuria]

Russ,

What do you say we both just gang up on Rev?!

 

[Padman]

Is nobody surprised by the fact that GFFX gets absolutely raped by radeon9000 in PS1.4?

Ran the tests again with DirectX 8 and now the 9000 is faster in allmost all the tests. this is even on a slower CPU.

My way of thinking is ATI learnt a lot from the 8500 shaders (PS1.4 being "real" pixelshaders vs. the register combiners of PS1.3 and below) and that this headstart in the programmability race is paying of for them now.

 

[Luminescent]

Seems plausible Padman.

 

[tb]

Have you installed the final dx9 release? (try dxdiag) Have you enabled the debug runtime in the directx control panel ?

Thomas

Final release, yes, but where do I enable/disable the debug runtime? (I have no directx control panel) Do I need to install the full DX9 SDK? (ATI's DX9 demos runs great BTW).

No you don't. If you havn't installed the SDK, you could not use the debug runtime, so there is no need for this specific control panel. Iam running out of ideas...

 

[LeStoffer]

I am running out of ideas...

Damn...

 

[tb]

ShaderMark 1.7

Hi!

I've released version 1.7.
- several bug fixes in the dx8 (ati's code wasn't 100% correct) / dx9 shaders and in the dx8 exe
- support for lower precision 2.0 shaders (partial precision)
- only one dx9 exe, you can switch through the shaders via the 1,2,3 keys

"The partial precision hint (represented as _pp in the assembly) can be used by the application to indicate to the device that the operation can be performed and the result stored at a lower precision (at least s10e5). This is a hint and many implementations might ignore it."

Thomas

 

[Althornin]

Alright!
we need someone (stares at brent ) to run some benchies on the GFFX with the _pp hint.

 

[Typedef Enum]

Actually, a new approach should be adopted by anybody with an FX in hand...

Downclock the thing to normal FX levels. It's no longer an option, IMHO, to benchmark a card that won't be available in retail channels. These things should be clocked @ 400/800...

 

[mr]

Actually, a new approach should be adopted by anybody with an FX in hand...

Downclock the thing to normal FX levels. It's no longer an option, IMHO, to benchmark a card that won't be available in retail channels. These things should be clocked @ 400/800...

I disagree.
IMO people with a 5800 Ultra should bench the card at default clockspeeds as it was built by nvidia and as people who preordered will (presumably) receive it. AFAIK nobody has received a sample of a 5800, so nobody knows if or how the 5800 will differ from the Ultra (and a downclocked Ultra), be it in a favourable way or not.

 

[Joe DeFuria]

I disagree.
IMO people with a 5800 Ultra should bench the card at default clockspeeds as it was built by nvidia and as people who preordered will (presumably) receive it.

Any "consumer" who has of course received his Ultra, has every right to bench it at whatever settings and publicize it. However, web review sites are different. The validity of using "Ultra" hardware depends entirely on the nature of the review: Is it just a review of the tech and general impressions of the card, or is it presented as an article to help the consumer to decide which card to buy? Most reviews are geared toward the latter structure. (They contain "shootout" bechmarks, comments on price, etc.) Those types of reviews should only exist for products that you can actually order and buy.

Assuming the rumor is true, and you can no longer ORDER a 5800 Ultra....there's no point in deciding if you want one. And benchmarking it is purely an exercise in "what if." It provides little more value than what "overclocking" benchmarks do.

AFAIK nobody has received a sample of a 5800, so nobody knows if or how the 5800 will differ from the Ultra (and a downclocked Ultra), be it in a favourable way or not.

Then it's up to nvidia to supply web-sites with actual 5800 (non-ultra) p/review samples.

 

[Joe DeFuria]

Addition...

Take a look at the recent Carmack .plan file. (Which is kind of a "shootout" specifically for Doom3).

Any guess on whether he has an Ultra or non Ultra?

How different do you suppose his .plan file "analysis" would be if he only had access to a 400/400 card, rather than the 500/500 Ultra....Are his .plan file comments even relevant any more?

I would like to see him update his .plan file based on the 400/400 FX...and give his thoughts on performance of cards that more than a handful of people can actually buy...

 

[LeStoffer]

Any guess on whether he has an Ultra or non Ultra?

He made a comment on the loud noise = ultra.