How do we report driver cheats to Futuremark?

[digitalwanderer]

Just curious, is there any official channels to tell FM about the new beta 53.03 set that "fixes" the 3.40 patch? Or is just posting up here about it enough to get Worm's attention and hopefully reaction to this?

 

[Reverend]

There's no need to "report". FM checks out drivers themselves.

BTW, FM isn't concerned about "beta" drivers or those released by board vendors when it comes to their business model. What's important is the stance taken by the IHV, which means to say the "official" drivers, which should mean drivers announced by the IHV.

Of course, feel free to post specific "fixes" in drivers wrt FM software in the forums... I'm sure FM will read all of them. Who knows, you guys may discover something FM missed.

 

[digitalwanderer]

Ok, thanks Rev.

So they're probably already aware of the 53.03 betas (files dated yesterday in em!) that restore the missing 3dm2k3 points that nVidia lost with the 3.40 patch...which nVidia probably won't ever officially release because then they'd be violating FM's rules.

Then again, no one will be allowed to post scores of un-official drivers to the ORB...right?

 

[zsouthboy]

How much do you want to bet they knew way ahead of time that the first beta set of drivers(and most likely, the next WHQLs) to come out after the 340 patch would have the performance "fixed"?

They knew. They're probably already on top of it.

EDIT: man, am I slow today.. is the internet down?

 

[{Sniping}Waste]

http://discuss.futuremark.com/forum...9&page=0&view=collapsed&sb=5&o=0&fpart=5&vc=1

Here are some test thhat show the problem With the VS 2.0 on the 5900U.
I have a Idea that the NV35 has 3 VS pipelines and each will do FP16 and it needs to cobine 2 pipelines to make FP32 and it kills the proformans like the 5800U and the PS 2.0 were it had to cobine 2 FP16 to make 1 FP32.

 

[Xmas]

Here are some test thhat show the problem With the VS 2.0 on the 5900U.
I have a Idea that the NV35 has 3 VS pipelines and each will do FP16 and it needs to cobine 2 pipelines to make FP32 and it kills the proformans like the 5800U and the PS 2.0 were it had to cobine 2 FP16 to make 1 FP32.

This doesn't make sense as FP16 is definitely unusable for vertex transformation.

Also, the 5800 doesn't combine two FP16 units to form one FP32 unit. All FP ALUs are purely FP32 (except rsq, which is faster in FP16 mode). However, it can store two FP16 vectors in one FP32 register, and the architecture is very sensitive to register usage. That's the reason why FP16 is faster than FP32.

 

[{Sniping}Waste]

Here are some test thhat show the problem With the VS 2.0 on the 5900U.
I have a Idea that the NV35 has 3 VS pipelines and each will do FP16 and it needs to cobine 2 pipelines to make FP32 and it kills the proformans like the 5800U and the PS 2.0 were it had to cobine 2 FP16 to make 1 FP32.

This doesn't make sense as FP16 is definitely unusable for vertex transformation.

Also, the 5800 doesn't combine two FP16 units to form one FP32 unit. All FP ALUs are purely FP32 (except rsq, which is faster in FP16 mode). However, it can store two FP16 vectors in one FP32 register, and the architecture is very sensitive to register usage. That's the reason why FP16 is faster than FP32.

Thanks for the info. So how is nvivda boosting the FPS in game test 2,3, and 4? By going to FP16 for the VS reduse the register usage?

 

[FUDie]

Here are some test thhat show the problem With the VS 2.0 on the 5900U.
I have a Idea that the NV35 has 3 VS pipelines and each will do FP16 and it needs to cobine 2 pipelines to make FP32 and it kills the proformans like the 5800U and the PS 2.0 were it had to cobine 2 FP16 to make 1 FP32.

This doesn't make sense as FP16 is definitely unusable for vertex transformation.

Also, the 5800 doesn't combine two FP16 units to form one FP32 unit. All FP ALUs are purely FP32 (except rsq, which is faster in FP16 mode). However, it can store two FP16 vectors in one FP32 register, and the architecture is very sensitive to register usage. That's the reason why FP16 is faster than FP32.

Thanks for the info. So how is nvivda boosting the FPS in game test 2,3, and 4? By going to FP16 for the VS reduse the register usage?

The vertex shaders doesn't contain FP16 at all, and has no need to. It's the pixel shader on the GeForce FX that can use FP16 to reduce register usage.

-FUDie

 

[bloodbob]

A) unless there is a hotfix driver then it non WHQL drivers won't get certified by futuremark.
B) if they ignored A) futuremark might not care and release 341 or just let them have their drivers as they currently allow cheating drivers in as without the latest patch the drivers cheat and with AF they cheat.

Vertex shaders for just about everything need 32 bit registers you might be able to get away with 16 registers for some operation in VS 3.0 but I don't know if the spec allows for them there.

 

[digitalwanderer]

Hold on a sec, I just was informed that people are submitting the 53.03 set for comparison and the ORB is accepting them and giving out compare URL's for them and are just not allowing them to be considered for top scoring and what not...

...WHAT?!?!?!

What's the bloody point then?!?! I couldn't tell from the compare link that it was an un-official score, they should have "CHEATER DRIVERS!!!!" flashing in big red letters or something if they're going to allow 'em to submit scores!

I really didn't think un-allowed drivers were allowed to post scores, I don't see any incentive NOT to cheat now.

 

[AzBat]

Hold on a sec, I just was informed that people are submitting the 53.03 set for comparison and the ORB is accepting them and giving out compare URL's for them and are just not allowing them to be considered for top scoring and what not...

...WHAT?!?!?!

What's the bloody point then?!?! I couldn't tell from the compare link that it was an un-official score, they should have "CHEATER DRIVERS!!!!" flashing in big red letters or something if they're going to allow 'em to submit scores!

I really didn't think un-allowed drivers were allowed to post scores, I don't see any incentive NOT to cheat now.

Dig,

Actually that's been known since at least when the patch was released. Only results using the approved drivers will be used for the Hall of Fames, Performance Analyzer(Game Advisor) and any other data services. Yes, you can post your results to the ORB, but the only way people can see them is if they change the default options in the Search & Compare Published Projects page. By default it's set to "Default Projects - WHQL Drivers". If you want to see the non-approved drivers results, then you have to select "Default Projects - All Drivers". Personally, I can understand your response and I partly agree, but I also don't have much of a problem with segregating the results the current way. The only way NVIDIA would benefit from this is from end-users who know they are showing results from non-approved drivers(measuring their ...). What's still left in the air is how Futuremark handles organizations(publications and web sites) that post results using non-approved drivers. If they don't do something against these parties, then I believe Futuremark might want to hang up the hat.

Tommy McClain

 

[digitalwanderer]

Thanks for the explanation Tom, I wasn't aware of it being that way and it did sort of catch me out-o-the-blue like a nasty slap in the face.

I really was under the impression that "cheater" drivers weren't even allowed to publish scores and really do not like the way the system is currently, I'd much rather only have allowed drivers be able to even publish scores.

 

[AzBat]

Thanks for the explanation Tom, I wasn't aware of it being that way and it did sort of catch me out-o-the-blue like a nasty slap in the face.

I really was under the impression that "cheater" drivers weren't even allowed to publish scores and really do not like the way the system is currently, I'd much rather only have allowed drivers be able to even publish scores.

No problem Dig. I would as well, but I understand why they don't and I'm not going to knock them for it. Provided they keep their promise to not use the cheated results for official reasons and always keep the cheated results segregated from the ones that are not cheating. Seems like a pretty good compromise to me. It will be interesting to see what happens once NVIDIA officially releases their next set of WHQL drivers.

Tommy McClain

 

[digitalwanderer]

Pagh, they should make the background for un-official results blazing red.

I'm curious if FM is going to take any actions with this 53.03 set which I'm betting nVidia is going to push as their "un-official official" set...at least for review purposes.

 

[AzBat]

Pagh, they should make the background for un-official results blazing red.

I'm curious if FM is going to take any actions with this 53.03 set which I'm betting nVidia is going to push as their "un-official official" set...at least for review purposes.

LOL. From what little response I've seen on Futuremark's forums, I believe Futuremark are just saying they're not official yet. So they only suggest using the ones on the approved list. Until they're official I doubt will hear anything new.

BTW, other than Hanners' article on EliteBastards, have any other site tried posting results using non-approved drivers in their articles and reviews? If there are, then maybe we need to notify Futuremark of them.

BTW, I'm still surprised we haven't heard anything out of Extremetech. I wonder why they're so quiet?

Tommy McClain

 

[digitalwanderer]

BTW, other than Hanners' article on EliteBastards, have any other site tried posting results using non-approved drivers in their articles and reviews? If there are, then maybe we need to notify Futuremark of them.

OMG, EB is in violation of FM's EULA!

I really never thought of that until just now...