Wait for news from Unwinder (nvworld.ru)

[chavvdarrr]

GF4600
3DMark Score 12257
Game 1 - Car Chase - Low Detail 179.6 fps
Game 1 - Car Chase - High Detail 62.8 fps
Game 2 - Dragothic - Low Detail 226.7 fps
Game 2 - Dragothic - High Detail 123.2 fps
Game 3 - Lobby - Low Detail 157.2 fps
Game 3 - Lobby - High Detail 70.5 fps
Game 4 - Nature 74.6 fps
Fill Rate (Single-Texturing) 1040.3 MTexels/s
Fill Rate (Multi-Texturing) 2305.2 MTexels/s
High Polygon Count (1 Light) 55.3 MTriangles/s
High Polygon Count (8 Lights) 12.6 MTriangles/s
Environment Bump Mapping 185.2 fps
DOT3 Bump Mapping 157.3 fps
Vertex Shader 98.3 fps
Pixel Shader 118.9 fps
Advanced Pixel Shader 102.6 fps
Point Sprites 32.2 MSprites/s

44.03 after using AntiDetector.rts:

3DMark Score 11014
Game 1 - Car Chase - Low Detail 179.0 fps
Game 1 - Car Chase - High Detail 62.4 fps
Game 2 - Dragothic - Low Detail 193.9 fps
Game 2 - Dragothic - High Detail 112.7 fps
Game 3 - Lobby - Low Detail 153.1 fps
Game 3 - Lobby - High Detail 69.9 fps
Game 4 - Nature 42.8 fps
Fill Rate (Single-Texturing) 1043.6 MTexels/s
Fill Rate (Multi-Texturing) 2301.5 MTexels/s
High Polygon Count (1 Light) 55.8 MTriangles/s
High Polygon Count (8 Lights) 12.5 MTriangles/s
Environment Bump Mapping 185.1 fps
DOT3 Bump Mapping 157.7 fps
Vertex Shader 98.2 fps
Pixel Shader 118.7 fps
Advanced Pixel Shader 90.1 fps
Point Sprites 32.0 MSprites/s
..................
........
looking in Catalyst's now .....

Any comments? What everyone knew - cheats not only in last version of 3dMark - is now confirmed ....

[Arun]

Remember the original Detonator 40 launch?
They claimed improved shading performance, and many people were amazed by the huge Nature increase.

Now we know where it comes from, I guess - it's pretty much back to the Pre-Detonator 40 scores. :(
Cheating is nothing new at NVIDIA Corporation, it seems...


Uttar

[Entropy]

Any comments? What everyone knew - cheats not only in last version of 3dMark - is now confirmed ....

Confirmation is a step up from suspicion though, even if few are surprised.
Thanks.

Entropy

[Sharkfood]

"Cheating" has been around in the Detonator's since the 12.xx releases- it's just there was too much of a fanbase to actually dig in and do any research at the time- any offerings of evidence were quickly discounted and hot air/windy retorts are all that ever came of them.

On topic of this current 2001 issue, the basics of research still need to be performed. Noting a ~10% delta in performance by running a benchmark two consecutive times isn't exactly earth shattering- and I'd also say is pretty close to the margin of error of 3dmark2001se (hell, a fresh reboot and re-run can often times effect the total score by 1000). The Nature test is of the most interest given it's huge delta.

What needs to happen with this scenario is the basics of homework- screenshots, comparisons, isolation of what (if any) visual differences may occur.. etc.etc. One can only raise a red flag if adequate controls are in place and if the differences can be isolated.

[mczak]

Any comments? What everyone knew - cheats not only in last version of 3dMark - is now confirmed ....
I've already said I want an anti-cheat 3dmark01se version...
What does AntiDetector.rts do? Does it prevent shader detection?
Though it looks only Game 4 is, umm, optimized a lot. The other optimizations don't seem to provide much benefits (aps and dragothic about 10%, the rest of the scores look identical). Do the optimizations change image quality?

[Typedef Enum]

Is there a URL somewhere? I looked on the Russian site, and didn't see anything that jumped out...but then again, my knowledge of Russian is pretty limited :)

[jb]

I am sure this is no real suprise. We all know that when only 3dmarks scores of the nature test shot up nV was going some optimizations there. But its nice to have some data now to back it up. I wonder if all of those who "defened" sides on this 3dmark fasico because 3dmark2003 was "lame", "a bad benchmark", ect will do now that we know that even 3dmark2001, every ones pride and joy, was subjected to a simular process. Also notice that both AJ and Worm have not been posting as of late.

Things that make you go "Hmmmmm"..... :wink:

[Dave Baumann]

wrt 3DMark2001 I'd be surprised if it was only NV.

[digitalwanderer]

I am sure this is no real suprise. We all know that when only 3dmarks scores of the nature test shot up nV was going some optimizations there. But its nice to have some data now to back it up. I wonder if all of those who "defened" sides on this 3dmark fasico because 3dmark2003 was "lame", "a bad benchmark", ect will do now that we know that even 3dmark2001, every ones pride and joy, was subjected to a simular process. Also notice that both AJ and Worm have not been posting as of late.

Things that make you go "Hmmmmm"..... :wink:
WHEEEEEEE!!!!!!!

Thanks, I hadn't thought of how difficult it's gonna be for places like [t]ardOCP to wiggle around this one. :)

[Grall]

Depending on the driver versions, I've had my GF3 do scores in the low 30s to mid-50s in 2k1 nature test. Naturally, pun not intended, I hoped these score improvements were due to GENERAL optimizations in the driver, not GT4-SPECIFIC optimizations.

Naturally, it would be very disappointed to find I've been pinning my hopes of increased scores to nothing but an illusion. ...Especially since the general attitude towards Nvidia standards has been very high in the past.


*G*

[Bambers]

interesting but not hugely suprising, i always though the 70-80 fps scores were a little too high.

Still, the 8500 went from 20ish at launch and is now about 45-50+ might be interesting to see if any of that is iffy.

[Hellbinder]

interesting but not hugely suprising, i always though the 70-80 fps scores were a little too high.

Still, the 8500 went from 20ish at launch and is now about 45-50+ might be interesting to see if any of that is iffy.

Are you sure???

I remember that GT4 was always a strong point for the 8500 because it had a better Vertex engine to start with. their scores were also good on the Hardware lights tests.

[Joe DeFuria]

]I remember that GT4 was always a strong point for the 8500 because it had a better Vertex engine to start with. their scores were also good on the Hardware lights tests.

I do remember a relatively earily 8500 driver build that added "non trivial" performance increases in some shading tests. Don't recall if it was GT4 or the synthetic shading tests, or both.

[Blackwind]

I am sure this is no real suprise. We all know that when only 3dmarks scores of the nature test shot up nV was going some optimizations there. But its nice to have some data now to back it up. I wonder if all of those who "defened" sides on this 3dmark fasico because 3dmark2003 was "lame", "a bad benchmark", ect will do now that we know that even 3dmark2001, every ones pride and joy, was subjected to a simular process. Also notice that both AJ and Worm have not been posting as of late.

Things that make you go "Hmmmmm"..... :wink:
WHEEEEEEE!!!!!!!

Thanks, I hadn't thought of how difficult it's gonna be for places like [t]ardOCP to wiggle around this one. :)

Seeing that they don't have anything to "have to wiggle around," pretty easy. Issues pointed out with Futuremark and its use pre-dates 3DMark03.

[Bolloxoid]

wrt 3DMark2001 I'd be surprised if it was only NV.

Well, Trident's tricks have already been exposed so we know it's not only Nvidia.

[just me]

I must caution the use of the word "cheat" as it applies to 3DM2K1.

FM has stated (in the "Help" file & the audit.pdf) what a "cheat" is in 3DM03, but I fail to see the word "cheat" used in any literature released by MO/FM concerning 3DM2K1 usage.

Carry on, but CYA (cover your a$$) in the terminology used. :wink:

.02 info post,

[jb]

wrt 3DMark2001 I'd be surprised if it was only NV.

Yeap I agree others to be shown "optimizing" most likely.

I guess my point was a lot of people said that since they did not like 3dmark2k3 then its ok for this to happen. I tried to show them that logic is flawed.

[chavvdarrr]

Actually it was posted on ixbt forum:
http://forum.ixbt.com/0010/044212-30.html

- Detonators have 73-element table of "recognized" applications
- ~50 "recognized" shaders
He simply wrote rivatuner-patch hich will "block" detection routines in the driver.
IMHO just a matter of time to see how 8500 fps in GT4 jumped ahead...

[Typedef Enum]

Does anybody have some David Kirk quotes handy? I'm sure there's somebody out there that keeps track of all the BS this guy has stated over time...

I'm specifically referring to the "we don't optimize for specific titles" quote...I want to re-read that quote, and see if it was carefully worded or not.

[martrox]

Does anybody have some David Kirk quotes handy? I'm sure there's somebody out there that keeps track of all the BS this guy has stated over time...

I'm specifically referring to the "we don't optimize for specific titles" quote...I want to re-read that quote, and see if it was carefully worded or not.

Think his excuse might be that he was smoking something hallucinogenic? :wink:

[cellarboy]

interesting but not hugely suprising, i always though the 70-80 fps scores were a little too high.

Still, the 8500 went from 20ish at launch and is now about 45-50+ might be interesting to see if any of that is iffy.

The 8500 drivers at release were terrible under XP. They were lurching around like a drunken sailor on my system until ATI released a fix for 'systems with large amount of memory' in January '02. Performance was awesome across the board after that.

[whql]

http://www.nvnews.net/articles/david_kirk_interview.shtml+David+Kirk+optimize&hl=en&ie=UTF-8

David: I’ve heard this mentioned before, and I haven’t really dug into it too much, since it’s not a very realistic real-world example. The lit triangles test in 3Dmark2001 renders a lot of offscreen (invisible) triangles, which results in a lot of what I would call “meaningless computation”. So, I’m not sure exactly what capability is being measured in the test. It’s not visible. Often, benchmarks test Software or Hardware code paths that are not general and are not commonly used. We at NVIDIA don’t make it a practice to optimize our pipeline for specific benchmarks - we want to provide high quality and high performance on a wide variety of useful and entertaining applications, rather than just getting a good score. Ask yourself (or, better yet, ask ATI) why Radeon 8500 performs well on this one test, and poorly on many other 3DMark2001 tests.

http://firingsquad.gamers.com/hardware/kirkint/page3.asp

It's very easy for people to target synthetic benchmarks and optimize their drivers and hardware for them. That does not add anything to real world performance. You need to optimize your drivers to make game engines and API features go faster, not just specific game titles.

[CorwinB]

Think his excuse might be that he was smoking something hallucinogenic?

Pass it along, David...
http://www.smiley-faces.com/smilies/rasta.gif

73 elements and 50 shaders, hey ?
/me thinks the drivers guys at Nvidia really deserve a raise...

[tEd]

how many of the ~50 shaders are for real games? How many shaders does shadermark have?? :twisted:

[Pete]

Any updates?

[chavvdarrr]

Well first results for R300 (no results for R200 drivers yet)
Unwinder:
Catalyst 3.4

3DMark Score 11676
Game 1 - Car Chase - Low Detail 167.6 fps
Game 1 - Car Chase - High Detail 61.3 fps
Game 2 - Dragothic - Low Detail 206.8 fps
Game 2 - Dragothic - High Detail 119.7 fps
Game 3 - Lobby - Low Detail 157.2 fps
Game 3 - Lobby - High Detail 71.1 fps
Game 4 - Nature 66.0 fps
Fill Rate (Single-Texturing) 961.3 MTexels/s
Fill Rate (Multi-Texturing) 2160.8 MTexels/s
High Polygon Count (1 Light) 54.8 MTriangles/s
High Polygon Count (8 Lights) 13.6 MTriangles/s
Environment Bump Mapping 129.5 fps
DOT3 Bump Mapping 128.0 fps
Vertex Shader 149.6 fps
Pixel Shader 199.6 fps
Advanced Pixel Shader 149.4 fps
Point Sprites 27.0 MSprites/s

Ñatalyst 3.4 + ATIAntiDetector

3DMark Score 11291
Game 1 - Car Chase - Low Detail 165.4 fps
Game 1 - Car Chase - High Detail 63.2 fps
Game 2 - Dragothic - Low Detail 207.5 fps
Game 2 - Dragothic - High Detail 119.3 fps
Game 3 - Lobby - Low Detail 155.5 fps
Game 3 - Lobby - High Detail 70.8 fps
Game 4 - Nature 47.1 fps
Fill Rate (Single-Texturing) 963.4 MTexels/s
Fill Rate (Multi-Texturing) 2154.2 MTexels/s
High Polygon Count (1 Light) 54.0 MTriangles/s
High Polygon Count (8 Lights) 13.7 MTriangles/s
Environment Bump Mapping 129.2 fps
DOT3 Bump Mapping 128.0 fps
Vertex Shader 149.1 fps
Pixel Shader 199.4 fps
Advanced Pixel Shader 149.6 fps
Point Sprites 27.0 MSprites/s

AT least 3 pixel shaders are detected (1@1.1 and 2@2.0), texture detections (thats how GT4 in 3dmark2001 is detected) ....
Full texture patterns& shader code are in the driver, but although easy to see what they do, its very hard to make FULL anti-detect patch.....
.... what ATi make with these 2 p.sh. 2.0 is hardly "shifting" ......
....

any anti-ATi comments? :)

[RussSchultz]

For shame for shame. ;)

I'm not suprised. All the companies seem to do it. There's been a long history of 'cheating' in benchmarks from just about any of the companies, all the way back to the VGA days.

[jvd]

chavvdarrr ati already admit to it . They stated it will be gone from the next driver . The same thing that caused the diffrence in 3dmark 2003 is causing that. At the end of the day at least ati owned up to it .

[Evildeus]

Hu? There's no pb with GT2-3 on 3Dmark03.
chavvdarrr ati already admit to it . They stated it will be gone from the next driver . The same thing that caused the diffrence in 3dmark 2003 is causing that. At the end of the day at least ati owned up to it .

[mczak]

Interesting that both ATI & Nvidia got a large improvement in Nature with optimizations (though Nvidia did the better job :x ) without sacrificing image quality (or did they?). I'm wondering what exactly they did - shader optimizations probably won't cut it (as those are relatively simple PS 1.1 shaders) and GT4 is usually quite bandwidth limited (due to the alpha-tests).

[jvd]

Hu? There's no pb with GT2-3 on 3Dmark03.
chavvdarrr ati already admit to it . They stated it will be gone from the next driver . The same thing that caused the diffrence in 3dmark 2003 is causing that. At the end of the day at least ati owned up to it .

actually one of the test drops 13% with out the optimizations .

[Althornin]

Damn shame.
Personally, i iwll wait till i hear what the cheats do to IQ before i decide which is worse, but i dont like either of them.
I also wonder which of them cheater "first"?
IE, was it a case of "they cheated, we had to to keep up" (which i dont buy, you can always just expose them, of course, they did that this time, and look! ATI and 3dmark are getting all the bad press, so obviously, its not always the best path, eh?)?

[StealthHawk]

Does anyone know if this anti-detection patch script will ever be released to the public? I am most interested in seeing the results of 3dmark2001 with a gfFX card, and not a gf4Ti4600. And of course, I'd like to see some extensive game testing.

[K.I.L.E.R]

Does anyone know if this anti-detection patch script will ever be released to the public? I am most interested in seeing the results of 3dmark2001 with a gfFX card, and not a gf4Ti4600. And of course, I'd like to see some extensive game testing.

Hey SH, I would also like to see the FX perform with Unwinder's anti-detection patch script.

[jvd]

Damn shame.
Personally, i iwll wait till i hear what the cheats do to IQ before i decide which is worse, but i dont like either of them.
I also wonder which of them cheater "first"?
IE, was it a case of "they cheated, we had to to keep up" (which i dont buy, you can always just expose them, of course, they did that this time, and look! ATI and 3dmark are getting all the bad press, so obviously, its not always the best path, eh?)?

the ati thing doesn't do anything for iq . But there is a diffrence with it on. They found a way to speed up thier rendering . I dunno i will look for where i read it and post back asap

[Evildeus]

There's no drop in GT2/3 in 3Dmark03 which use PS1.4 contrary to GT4 which uses PS2.0. So there's some optimisation for PS1.4 in 3DMark01 and not in 3Dmark03.
Hu? There's no pb with GT2-3 on 3Dmark03.
chavvdarrr ati already admit to it . They stated it will be gone from the next driver . The same thing that caused the diffrence in 3dmark 2003 is causing that. At the end of the day at least ati owned up to it .

actually one of the test drops 13% with out the optimizations .

[FUDie]

There's no drop in GT2/3 in 3Dmark03 which use PS1.4 contrary to GT4 which uses PS2.0. So there's some optimisation for PS1.4 in 3DMark01 and not in 3Dmark03.
PS 1.4 is not used in any of the game tests in 3D Mark 2001. The advanced pixel shader test in 3D Mark 2001 uses PS 1.4, if supported, but it has no bearing on your score.

-FUDie

[Evildeus]

Well it's not true, it's used in the advanced pixel shader. But you are right i confused with it ;). Thx you for correcting me.

Anyway, the "optimisation" on nature doesn't seem to be the same as the one on GT4 on 3DMark03

[just me]

Anyone consider that ATi optimized the PS1.0 into PS2.0 (or PS1.4 for 8500/9100) for the higher FPS w/o loss of IQ? Could simply be the ATi cards use the highest PS they support & I don't see anything wrong with that in 3DM2K1. I hope they do it for my games too! :D

Mountain out of a molehill, IMHO. :wink:

.02,

[Doomtrooper]

Yes ATI is probably doing more than re-ordering in 3Dmark 2001, they are possibly even replacing the shader with a superior PS 1.4 version since Futuremark never included PS 1.4 in scoring during the benchmarks 2 years of existence...a really big joke.

Is image quality reduced, I would say absolutley not...and I personally have no problem with optimizations that don't reduce quality, as those examples could also be used in games.

[RussSchultz]

since Futuremark never included PS 1.4 in scoring during the benchmarks 2 years of existence...a really big joke.

The point of 3dmark is to have a consistant test. You can't change the contents of the test without invalidating the scores already taken.

This is why they came up with 3dmark03, instead of simply updating 3dmark01.

Yes, I agree, the changes done by ATI probably make much sense on modern day cards, however its not what was being tested.

[micron]

Forgive my lack of intelligence, but let me see if I'm following this correctly.
GT4 in 3DMark 2001SE is calling for PS1.0, but the Catalyst drivers are delivering another? which shader are they delivering exactly? 1.1? 1.4? both?

[RussSchultz]

Forgive my lack of intelligence, but let me see if I'm following this correctly.
GT4 in 3DMark 2001SE is calling for PS1.0, but the Catalyst drivers are delivering another? which shader are they delivering exactly? 1.1? 1.4? both?
Dunno the exact details, but that seems to be the gist of it.

[Doomtrooper]

since Futuremark never included PS 1.4 in scoring during the benchmarks 2 years of existence...a really big joke.

The point of 3dmark is to have a consistant test. You can't change the contents of the test without invalidating the scores already taken.

This is why they came up with 3dmark03, instead of simply updating 3dmark01.

Yes, I agree, the changes done by ATI probably make much sense on modern day cards, however its not what was being tested.

I hope the same mentality is used with FX12/FP16 _pp hint 'patch' , as the same mentality came be applied.

In fact absolutley the same, except PS 1.4 would not be lower in IQ.

[micron]

Forgive my lack of intelligence, but let me see if I'm following this correctly.
GT4 in 3DMark 2001SE is calling for PS1.0, but the Catalyst drivers are delivering another? which shader are they delivering exactly? 1.1? 1.4? both?
Dunno the exact details, but that seems to be the gist of it.
Thank you for replying Russ....

[just me]

Forgive my lack of intelligence, but let me see if I'm following this correctly.
GT4 in 3DMark 2001SE is calling for PS1.0, but the Catalyst drivers are delivering another? which shader are they delivering exactly? 1.1? 1.4? both?


PS1.0 is used in 2K1 GT4. Changing to 1.1, 1.2 or 1.3 wouldn't be a performance improvement. PS1.4 &/or PS2.0 would tho'. *If* they use PS1.4 for 8500/9100 & PS2.0 or PS1.4 for 9500+, that *could* explain the results (IMO).

It *could* simply be that the drivers call for the highest PS supported. I find nothing wrong w/that even in a benchmark as it gives a true 'real world' optimization that would be applied 'across the board' & wouldn't be 'application specific'.

I don't know if that is what *is* happening, but it seems that might be what is happening. :wink:

HTH & it's just my .01 :wink:

[RussSchultz]

You can't "call for the highest level pixel shader". (well, not when 3dmark2001 was written)

Unless you're using a high level shading language (like Cg, or the HLSL in DirectX, or GSLang, etc), you code your shaders using 'assembly'.

If a driver is recognizing a particular shader PS1.0 assembly sequence and replacing it with an optimized PS1.4 assembly sequence, its "cheating".

Even though, in the real world, you'd likely find developers using that optimized PS1.4 sequence, it isn't what the benchmark tool is asking for.

Just like the partial precision hints that some people think that Futuremark should have implemented, it doesn't change that fact that they didn't. Going behind the benchmark's back and doing it there isn't the right answer.

[StealthHawk]

Does anyone know if nvidia's optimizations in 3dmark2001 yielded in a drop in IQ?

[just me]

You can't "call for the highest level pixel shader". (well, not when 3dmark2001 was written)

Unless you're using a high level shading language (like Cg, or the HLSL in DirectX, or GSLang, etc), you code your shaders using 'assembly'.

If a driver is recognizing a particular shader PS1.0 assembly sequence and replacing it with an optimized PS1.4 assembly sequence, its "cheating".

Even though, in the real world, you'd likely find developers using that optimized PS1.4 sequence, it isn't what the benchmark tool is asking for.

Just like the partial precision hints that some people think that Futuremark should have implemented, it doesn't change that fact that they didn't. Going behind the benchmark's back and doing it there isn't the right answer.

'Call', 'code', 'assemble' ... whatever, you get the gist.

This is 2003. 3DM2K1 is old. It was written for PS1.0 DX8 (max) for the score. Running anything over a GF3 'Classic' is a *cheat*. :wink:

Allowing a driver to 'real world' (not application specific, but for everything) optimize 2K1 for the newer cards isn't wrong > it is SOO right, IMHO. What better way for a consumer to see the difference in the new hardware & it's supported API's? 8) The caveat being: as long as it doesn't lower IQ. 8)


PS: The topic is 3DMark 2001 NOT 3DM03. There is a HUGE difference. The main being > MO never said optimizing is a *cheat*. :wink:

[Althornin]

gah, i cant believe this.
It is still a CHEAT.
However, there are shades of grey.
IF indeed all ATI has done is rework some shaders so the output is mathematically identical but it executes faster on ATI hardware, then that is STILL A CHEAT in the context of 3dmark01.

Is it less wrong than a cheat that lowers IQ?
Hell yes.

[micron]

IF indeed all ATI has done is rework some shaders so the output is mathematically identical but it executes faster on ATI hardware, then that is STILL A CHEAT in the context of 3dmark01.
Did they not use a means of app detection to achieve the shader switcharoo?(or test detection)

[micron]

Does anyone know if nvidia's optimizations in 3dmark2001 yielded in a drop in IQ?
Replacing PS 1.0 with 1.4, or 2.0 should not drop the IQ.

[StealthHawk]

Does anyone know if nvidia's optimizations in 3dmark2001 yielded in a drop in IQ?
Replacing PS 1.0 with 1.4, or 2.0 should not drop the IQ.

Well, no it shouldn't.

But AFAIK there is no indication that is what nvidia has done. For R300 cards, ATI has replaced some PS1.0 shaders with PS2.0 shaders. I haven't read anywhere that nvidia was also converting PS1.0 shaders to PS2.0/PS1.4 shaders.

edit: corrections.

[micron]

GT 4 is calling for PS 1.0.....

[StealthHawk]

GT 4 is calling for PS 1.0.....

Ok, whatever. The point still stands.

Where was it proven that nvidia was converting PS1.0 programs into PS1.4 or PS2.0 programs?

[micron]

GT 4 is calling for PS 1.0.....

Ok, whatever. The point still stands.

Where was it proven that nvidia was converting PS1.0 programs into PS1.4 or PS2.0 programs?Does anyone know exactly which pixelshaders Nvidia is using in GT 4?(other then 1.0)?