Shadermark

Lezmaka · May 28, 2003

I had been wondering if the improvements seen in shadermark were because of something similar to 3DMark03, replacing shaders with ones that work better with NV3x.

I saw that [H] has a review of the "new" 5600 Ultra. Supposedly, the only things that have changed are the packaging and a 50MHz increase in core and memory clocks to 400/400. And of course, new drivers.

I saw that the Shadermark scores for the new 5600 Ultra were pretty darn close to the 9600 Pro, when in the past, it was nowhere close.

So I went back and looked at the initial 9600 Pro review, which used 43.45 for the 5600 Ultra.

Here's a comparison of the scores (too lazy to write out the individual test names, sorry)...

Code:

  New     Old
 44.03	 43.45
------  ------
378.79	367.15
307.69	267.64
136.26	116.24
131.02	29.58
98.52	 20.88
139.13	24.67
107.79	19.86
132.61	24.96
103.66	20.08
138.93	26.58
107.68	21.13
116.19	23.26
93.31	 18.92
96.16	 23.05
80.82	 18.90
84.73	 20.20
71.91	 16.85
81.53	 19.66
67.85	 16.52
92.53	 19.50
77.39	 16.40
98.17	 20.78
80.81	 17.30
52.87	 10.73
35.51	 9.39
48.42	 10.20
50.95	 10.62

Since the difference is an increased clockspeed (chip packaging shouldn't have an effect on performance right?) somethings sure seems wrong to me.

Here's links to the info...
Old 5600U Scores - http://www.hardocp.com/article.html?art=NDU4LDY=
New 5600U Scores - http://www.hardocp.com/article.html?art=NDc4LDc=

Not meaning to start another cheat thread that might end up getting heated like all the others, just curious

tEd · May 28, 2003

This certainly needs more investigation , i hope tommy can take a look at it.

Doomtrooper · May 28, 2003

If a review site would just compare the high dynamic range DX9 Demo between the two sets, that would help alot..if PS 2.0 speed is improved that much, it would show ACROSS the board.

FUDie · May 28, 2003

Doomtrooper said:
If a review site would just compare the high dynamic range DX9 Demo between the two sets, that would help alot..if PS 2.0 speed is improved that much, it would show ACROSS the board.

I think it would be more interesting to change the shaders slightly but leave the instruction count the same and see how performance compares between the new and old drivers. This would show whether the improvements were application specific or general improvements to the driver.

-FUDie

McElvis · May 28, 2003

Well www.hardware.fr has a review of the new 5600 Ultra today, using Det 44.03 drivers.

They test the High Dynamic Range test and Humus Mandelbrot rendering program. Both were still very slow. No data for an older driver set though...

HDR (1024*768 32 bits, AA 4x) -
5600 Ultra - 1.78 fps
ATI 9600 Pro - 10.5 fps
ATI 9500 Pro - 13.9 fps

Mandelbrot (1024*768 32 bits) -
5600 Ultra - 9 fps
ATI 9500 Pro - 45 fps
ATI 9600 Pro -32 fps

Link to a translated version - http://216.239.37.104/translate_c?h.../articles/468/page4.html&prev=/language_tools

tEd · May 29, 2003

McElvis said:
Well www.hardware.fr has a review of the new 5600 Ultra today, using Det 44.03 drivers.

They test the High Dynamic Range test and Humus Mandelbrot rendering program. Both were still very slow. No data for an older driver set though...

HDR (1024*768 32 bits, AA 4x) -
5600 Ultra - 1.78 fps
ATI 9600 Pro - 10.5 fps
ATI 9500 Pro - 13.9 fps

Mandelbrot (1024*768 32 bits) -
5600 Ultra - 9 fps
ATI 9500 Pro - 45 fps
ATI 9600 Pro -32 fps

Link to a translated version - http://216.239.37.104/translate_c?h...icles/468/page4.html&prev=/language_tools

thx for the link , interesting

Doomtrooper · May 29, 2003

Finally..and thanks McElvis and http://www.hardware.fr/

Now, where is the optimizing thread...

jjayb · May 29, 2003

So, this is what we have. The 44.03 drivers doubled the p.s. 2 performance of the fx cards on 3dmark03 and shadermark. They have not doubled the performance of any of the other p.s. 2 programs. IE: Rightmark, the mandlebrot demo, or the high dynamic range demo. We've since found out that the 3dmark03 scores were doubled by shader replacement. Once 3dmark03 was patched to defeat application detection the shader performance went back down to the low numbers they were at in earlier drivers. I assume it would be safe to say the shadermark scores have been doubled the same way.

I can hear it now though, "but none of these are actual games".

Lezmaka · May 29, 2003

Or more than doubled in most of the Shadermark scores (as much as 5x faster)

Since I've never run Shadermark, is there any way to take screenshots to compare?

Ante P · May 29, 2003

Lezmaka said:
Or more than doubled in most of the Shadermark scores (as much as 5x faster)

Since I've never run Shadermark, is there any way to take screenshots to compare?

I've done before/after screenies and also compared it to R3x0, Shadermark looks identical in all cases so there's at least no visual degredation

I also ran Humus mandel brot demo on my 5800 just now and it has nowhere near as crappy IQ as hardware.fr's shots

downloading the rthdribl demo now

update:
just ran rthdribl and indeed there's a precision problem and performance is simply horrible. jumps between 10-12 fps

as for Humus mandel brot demo it runs at 53 fps

Hellbinder · May 29, 2003

Being that the hardware is the same, and it still has half to less then half the Execution units. I do not see any way that the Scores can increase without a reduction in Quality.

Could someone please explain this to me?

Ante P · May 29, 2003

Hellbinder[CE said:
]Being that the hardware is the same, and it still has half to less then half the Execution units. I do not see any way that the Scores can increase without a reduction in Quality.

Could someone please explain this to me?

Shadermark was originally created with DX8.1 in mind so perhaps it simply doesn't benefit very much from increased range/precision?

in any case just look at dawn, even when forcing FX12 for all shaders it looks almost identical to forcing FP32 on all shaders

Miksu · May 29, 2003

Doomtrooper said:
Finally..and thanks McElvis and http://www.hardware.fr/

I found the review and especially the test results very good, though I don't understand that language

They used IL-2, Freelancer etc. and we really don't see test-results from these kind of games very often. What I found interesting was the fact that in 1024x768 4xaa & 8xaniso the only games where FX 5600U was faster than Radeon 9600Pro were Serious Sam2, UT2003 and the new C&C.

Typedef Enum · May 29, 2003

Those images really say a lot, IMHO.

I mean, if you haven't been sold on the idea that nVidia sacrifices image quality for performance before...sure seems pretty obvious to me.

The funny thing is that under those 2 tests, the FX gets its ass slapped. Image the difference if the quality was on par with ATI!

Marc · May 29, 2003

Ante P said:
Lezmaka said:

Or more than doubled in most of the Shadermark scores (as much as 5x faster)

Since I've never run Shadermark, is there any way to take screenshots to compare?

Click to expand...

I also ran Humus mandel brot demo on my 5800 just now and it has nowhere near as crappy IQ as hardware.fr's shots

Hi

You have to zoom in in the humus demo to see the problem

Ante P · May 29, 2003

Marc said:
Ante P said:

Lezmaka said:

Or more than doubled in most of the Shadermark scores (as much as 5x faster)

Since I've never run Shadermark, is there any way to take screenshots to compare?

Click to expand...

I also ran Humus mandel brot demo on my 5800 just now and it has nowhere near as crappy IQ as hardware.fr's shots

Click to expand...

Hi

You have to zoom in in the humus demo to see the problem

rofl, hehe yeah mate, I might be a bit slow but I'm not that stupid

Marc · May 29, 2003

lol sorry

can you provide me a screenshot ? i personally have the same problem on 5800.
Here is the zoom view on 5600 (it's the same on my 5800)
http://www.hardware.fr/medias/photos_news/00/06/IMG0006350_1.jpg
Here is the zoom area for your information (with the white)

Ante P · May 29, 2003

Yes I have the redness but I don't have the "blocky closeups" so to speak (and yeah I tried zooming in the same area as they did)

Doomtrooper · May 29, 2003

Whats different ?? Same Hardware, same driver..just eyes I guess.

Too bad there was no way to RMA eyes.

T2k · May 29, 2003

jjayb said:
So, this is what we have. The 44.03 drivers doubled the p.s. 2 performance of the fx cards on 3dmark03 and shadermark. They have not doubled the performance of any of the other p.s. 2 programs. IE: Rightmark, the mandlebrot demo, or the high dynamic range demo. We've since found out that the 3dmark03 scores were doubled by shader replacement. Once 3dmark03 was patched to defeat application detection the shader performance went back down to the low numbers they were at in earlier drivers. I assume it would be safe to say the shadermark scores have been doubled the same way.

Do we have more test prog? 'cause it seems very strange... I can't believe it... maybe it'll turn out to be the 2nd Mark scandal?

I can hear it now though, "but none of these are actual games".

Me too...

EDIT: typos

Shadermark

Lezmaka

tEd

Casual Member

Doomtrooper

FUDie

McElvis

tEd

Casual Member

Doomtrooper

jjayb

Lezmaka

Ante P

Hellbinder

Ante P

Miksu

Typedef Enum

Marc

Ante P

Marc

Ante P

Doomtrooper

T2k

Similar threads