New 3DMark03 patch out - nVidia still cheating

[digitalwanderer]

Originally posted by Voudoun
It's my lot to be ignored obviously. I did it on page 6, half-way down.

Voudoun

Thanks Voudoun, I do remember reading that now...I spaced that GF4ti's couldn't do GT4 too so it shouldn't really impact their scores much.

Thanks for taking the time to point it out to me again, I'll try not to space in the future. (Yeah, next time I plan on "flaking"... )

[Uttar]

Originally posted by adlai7
What I find interesting about the Beyond3d article, is that the test that lost the most performance, GT4 had no image differences between build 330 and 340. So what did change between the builds?

Well...

1) NVIDIA might have done some calculations on the CPU, which would actually DECREASE performance in a real game: the CPU is not the bottleneck in 3DMark03, but it often is in real games.

2) There were some minor IQ differences on the water. Yes, it was minor, but FP16 most likely was used. The not-so-high IQ differences compared to before might be that on the NV30/NV31/NV34, NVIDIA used FX12 I bet.

And regarding 1) - who knows what NVIDIA is doing on the CPU, or precalculated in textures... Maybe a lot, or maybe nothing!


Uttar

BTW, Voudoun: In what GT are you losing performance? GT2 and GT3 I assume? Or maybe just one of the two?

[AnteP]

Originally posted by DMA
Well, the ATI baised sites are gonna use it again in their next reviews for sure

The perfect benchmark suite for the ATI baised reviewer:

Mafia
Tomb Raider
BF 1942
3DMark-03 Build 340
Max Payne 2
UT2003

Of course you run all the tests at high res with AA/AF kicked up as high as possible. (IMPORTANT: Not one single test, except 3DM-03 of course, can be run without at least 4xAA/8xAF! NV might do better then.)

And throw in a good old Q3 bench just to give NV one win, cause we don't wanna seem baised right?

ROFL!

[ivzk]

My example of the problem at hand!


I spend 10 hours studying for an exam. It's the exact same exam everyone in my class is writing. As I go down the list of questions, I'm struggling a bit trying to remember all the answers. I look over at the guy sitting next to me who is just flying through the exam. This guy has been a f***up since day one so I know he's cheating. I end up with an A and so does he, all be it through way different methods.

By the logic of some of the posters here it doesn't matter that the guy cheated. All that matters is the fact that he got an A. Same result, different means!

Yeah right! Try expaining that to your dean!

Peace!

[Voudoun]

Originally posted by digitalwanderer
Thanks Voudoun, I do remember reading that now...I spaced that GF4ti's couldn't do GT4 too so it shouldn't really impact their scores much.

Thanks for taking the time to point it out to me again, I'll try not to space in the future. (Yeah, next time I plan on "flaking"... )

No worries. I wasn't entirely serious with my which I could have made clearer. I've been post cold and throat infection. How are you doing now that you've had some time to recover?

My contribution didn't turn out to be very helpful I know, but it did provide confirmation, if any were needed, where the problem lies. The more this goes on, the more I want to replace this Ti4600.

Voudoun

[Razor04]

I got it from the Inq...but still pretty interesting. Seems to me the NV PR department is in full swing making up excuses for their partners so they don't switch to ATI. Please make note of the bull in bold.

http://www.xbitlabs.com/news/video/d...112031947.html

Originally posted by X-Bit Labs
An official representative from a well-known manufacturer of graphics cards based on NVIDIA GeForce graphics processor today accused Futuremark Corporation, the developer of 3DMark software – an industrial benchmarks for graphics cards – of disabling certain functions available in NVIDIA ForceWare 52.16 drivers.

Futuremark Corporation announced yesterday the immediate availability of its v340 patch to the 3DMark03 benchmark suite. According to Tero Sarkkinen, Executive Vice President of Sales and Marketing at Futuremark Corporation, “the new version is published to make sure that our customers can get an objective 3DMark03 performance comparison with the latest hardware and drivers. Our customers will be able to perform apples-to-apples performance comparisons between the various IHVs’ graphics cards.”

Earlier this year Futuremark Corporation unveiled 3DMark03 specific optimizations in drivers from the world’s leading graphics processor companies, such as NVIDIA Corporation and ATI Technologies. 3DMark03 specific optimizations in drivers are against run rules of 3DMark03, because they invalidate the performance measurement results and thus make it impossible to compare performances of different hardware. In an attempt to disable benchmark specific optimizations, Futuremark implemented a number of changes in its 3DMark03 benchmark to avoid its recognition by GPU drivers that deploy application-specific optimizations.

As a response to Futuremark findings, ATI Technologies promised to remove its 3DMark03 application specific optimization from its CATALYST drivers and reportedly has fulfilled its promise. NVIDIA Corporation has never made official comments in regards application specific-optimizations for 3DMark03.

In October 2003 the world’s largest GPU manufacturer – Santa Clara, California-based NVIDIA Corporation – introduced its new Unified Compiler specifically integrated in its ForceWare 52.16 drivers that optimize Pixel Shader code for NVIDIA GeForce FX architecture to improve performance of graphics cards powered by NVIDIA’s latest GPUs in variety of demanding applications.

NVIDIA said that compiler technology tunes DirectX 9.0 execution on GeForce FX GPUs, and can be used to correct any similar conflict that arises with future APIs. NVIDIA indicated that the Unified Compiler is an automatic tuning tool that optimizes Pixel Shader performance on all applications, not just on specific ones.

According to reports from various web-sites dedicated to graphics cards benchmarking, NVIDIA GeForce FX 5950 Ultra scores roughly 15% lower in the newly patched version of 3DMark03 compared to previous build.

However, it appears that some technical specialists have something to say against the new patch. Hans-Wolfram Tismer, a Managing Director for Gainward Europe GmbH said today: “According to my information patch 340 disables the GPU compiler. The compiler has to run on the CPU instead resulting in code harder to digest and taking away 20% of the performance. This may not reflect gaming performance and may point in the wrong direction. To me 3DMark03 may look less and less suitable to be used for benchmarking. By the end of the day the end users are running games and applications, not 3DMark03.”

Due to some reasons, 3DMark03 in-depth testing of this web-site revealed no performance penalty in 3DMark03 Pixel Shader 2.0 performance test, but uncovered a 26.5% speed drop in Vertex Shader test.

According to benchmarks available around the World Wide Web, ATI RADEON hardware using the latest drivers performs equally in all versions of 3DMark03.

Representatives from NVIDIA and Futuremark did not return enquiries at press time.

[ChrisRay]

Originally posted by Uttar
Well...

1) NVIDIA might have done some calculations on the CPU, which would actually DECREASE performance in a real game: the CPU is not the bottleneck in 3DMark03, but it often is in real games.

2) There were some minor IQ differences on the water. Yes, it was minor, but FP16 most likely was used. The not-so-high IQ differences compared to before might be that on the NV30/NV31/NV34, NVIDIA used FX12 I bet.

And regarding 1) - who knows what NVIDIA is doing on the CPU, or precalculated in textures... Maybe a lot, or maybe nothing!


Uttar

BTW, Voudoun: In what GT are you losing performance? GT2 and GT3 I assume? Or maybe just one of the two?

Well it seems Game Test 2/3 are Vertex Shader Limited.. So it's possible for the CPU to emulate vertex shaders with lower performance loss...

[Voudoun]

Originally posted by Uttar

BTW, Voudoun: In what GT are you losing performance? GT2 and GT3 I assume? Or maybe just one of the two?

To be honest when I saw I'd only lost 65 points (I think) I didn't check as I felt it could have been acceptable error margin difference. I'll give it another go and see. I've got suspicions, but I'll run it again to confirm.

Voudoun

[digitalwanderer]

Originally posted by Voudoun
I've been post cold and throat infection. How are you doing now that you've had some time to recover?

Sorry to hear about the illness and hope you're past the worst of it and recover quickly, bad health SUCKS!

I'm doing faboo. I had my first cardiologist appt Monday (I ditched the last one after waiting an hour and a half, I HATE waiting for doctors! ) and the Doc was bloody delighted with my progress. I was given a clean bill-o-health and fully-recovered status with no follow-up appointment for 3 months. I just have to keep exercising, eating right, taking me meds (4 pills a day, nothing major), and NOT SMOKING and I'll be good to go.

I'm actually a lot better off than before my heart attack. I had a 95% blockage on one of my valves and my circulation sucked, now that it's opened up I have a LOT more energy. The only weird thing is my pulse is kind of high from the meds I'm on, I'm averaging about 80bpm compared to a normal 40-50 for me. It's normal and to be expected the Doc says, but it's just a bit weird to me. (Yeah, I'm a bit more conscientious about my pulse now. )

Back on topic...

Hans-Wolfram Tismer, a Managing Director for Gainward Europe GmbH said today: “According to my information patch 340 disables the GPU compiler. The compiler has to run on the CPU instead resulting in code harder to digest and taking away 20% of the performance. This may not reflect gaming performance and may point in the wrong direction. To me 3DMark03 may look less and less suitable to be used for benchmarking. By the end of the day the end users are running games and applications, not 3DMark03.”

The only problem with that is it's utter and total bullcrap of course. The GPU compiler would work with 3dm2k3 on a non-application specific basis or it would compromise nVidia's own internal policies on optimizations. (As well as FutureMark's. )

[nelg]

Originally posted by DMA
Well, the ATI baised sites are gonna use it again in their next reviews for sure

The perfect benchmark suite for the ATI baised reviewer:

Mafia
Tomb Raider
BF 1942
3DMark-03 Build 340
Max Payne 2
UT2003

Of course you run all the tests at high res with AA/AF kicked up as high as possible. (IMPORTANT: Not one single test, except 3DM-03 of course, can be run without at least 4xAA/8xAF! NV might do better then.)

And throw in a good old Q3 bench just to give NV one win, cause we don't wanna seem baised right?

How many of those are in the TWIMTBP program ?

[Razor04]

I didn't really comment on that one paragraph in my previous post but...

Shouldn't any compiler run on a CPU? My understanding is it takes the HLSL code that is given to it...transforms it into code that will run on the GPU and sends it off to the GPU to be processed. Either my understanding is wrong or somehow a compiler can now be done in hardware on a NV3X GPU.

Now lets have another little hypothetical situation where the compiler runs on the GPU not the CPU. Shouldn't they be outputting identical code? I would hope so...but then again is it even possible? (I am a Mechanical Engineer not a Computer Engineer or Comp Sci. major so please excuse me if I am wrong)

The whole line about 3DMark not being a game just reeks of PR. I mean come on...NV PR has been slinging that bunch of BS our way for months...yet there are still idiots out there that take it at face value.

[Hanners]

Originally posted by nelg
How many of those are in the TWIMTBP program ?

I think all of them except for 3DMark03 and Mafia off the top of my head. Even Mafia might be one, I don't remember.