NVIDIA Responds To GPU PhysX Cheating Allegation

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
The crew at HotHardware is reporting on the recent GPU PhysX cheating allegations leveled against NVIDIA. They talk to several folks at NVIDIA, Epic Games and Futuremark in an attempt to shed a little more light on the situation. What do you guys think of this whole deal?

3DMark Vantage tested in "Performance" preset mode, with and without GPU PhysX enabled, on Core 2 Extreme QX6850 powered test system running Windows Vista Ultimate with the Forceware v177.39 graphics drivers and v8.06.12 PhysX drivers on a pair of GeForce 9800 GTX+ cards running in SLI mode. With GPU PhysX disabled, the overall 3DMark Vantage score is P11720 with a CPU Test 2 score of 14.95 FPS. With GPU PhysX enabled, the overall score jumps to P14298 with a CPU Test 2 score of 127.77 FPS.
 
Meh. Hopefully Intel & AMD will pressure Futuremark to add Havok support in the same test later on. No biggie.
 
Giving gamers free PhysX by using the GPU - A good idea, shame the implementation is only for this and that in a properitary language, but what I have not seen and consider important - what kind of performance does it take away from the GPU in say, one of those PhysX levels? I'd like to see benchmarks. The hype makes me think as high FPS as before but this time with PhysX super-physics.

Having it in 3D Mark Vantage - Really weird. Why not Havok? Why not both physics solutions? Why at all? As a way to measure an entire system it's shit because it tells you to buy Nvidias PhysX driver capable cards or forget about it.


It is not cheating, but it is complete and utter bullshit aimed at fooling the simple minded to buy Nvidia.
 
damned nvidia, first sell us rehashed last-gen hardware for next-gen prices, then make 3-way sli that scales terribly and who's only purpose is to enhance our e-penises AND THEN implement this fucking "UMAP" which makes it a pain in the ass to buy cards from you, and now analrape anyone's 3dmark scores who doesnt have a nvidia card with phyX enabled. i hope that 4870x2 comes out soon, i need an upgrade and i WILL NOT buy from nvidia for some time now.
 
Is it nvidia's fault that Futuremark implemented PhysX, then also allowed for hardware PhysX? Well, no not really. All Futuremark had to do was disable hardware PhysX and there would have been no issue. Do people running the PhysX card get the same advantage? If so, this is pretty much a big to-do about nothing.
 
Highlights:
Oliver Baltuch of Futuremark said:
"<blah, blah about nvidia and ATI are following the driver submission rules>... Outside of this matter, we have been introduced to this technology from NVIDIA and it is truly innovative for future games and game development. As you know, we have an actual game coming as well and it could also make use of PhysX on the GPU.”
Futuremark doesn't seem outraged. Gosh, Charlie was going to bring the rope for FM to hang nvidia with.
Mark Rein said:
"... So Nvidia has done a cool optimization that allows their customers to get more value and performance out of their graphics cards. How anyone could possibly confuse this for a bad thing is beyond me."
You and me both, Mark. But then again, this is Charlie the newsmaker.
hothardware said:
and finally, Futuremark doesn't have a problem with the driver (and actually called the technology innovative) ... So what's all the fuss about? Exactly, move along...there's nothing to see here; except for maybe just better physics effects.
 
My best guess is that if that test is meant to stress physics and is light on the graphics the card just does more physics which would increase the score (obviously). If the score is based on an assumed separate physics board it could be an unintentionally large boost to overall score.
 
damned nvidia, first sell us rehashed last-gen hardware for next-gen prices, then make 3-way sli that scales terribly and who's only purpose is to enhance our e-penises AND THEN implement this fucking "UMAP" which makes it a pain in the ass to buy cards from you, and now analrape anyone's 3dmark scores who doesnt have a nvidia card with phyX enabled. i hope that 4870x2 comes out soon, i need an upgrade and i WILL NOT buy from nvidia for some time now.

Yes! How DARE nvidia give people who already have their hardware new features? I mean the nerve of them to roll out a feature upgrade for people who already have the cards! They should totally limit this to only brand new cards, and they should stop innovating! God dammit I want only simple graphics and no physics at all! I am so pissed at nvidia for making improvements. :rolleyes:

Next time please read the article. The benchmark didn't change, and it isn't coded for it. In fact, its just a runtime update for PhysX - any game that uses PhysX will benefit just as much as 3D mark - there isn't any foul play going on here at all. To those who say "why not havok?", my response would be WHY havok? Havok doesn't support GPU physics as far as I know, nor does it have any hardware support. It would be a complete side step for no reason, other than to remove an advantage that nvidia and PPU owners rightly have.

From the interview:
Mark Rein, Epic Games chimes in: And finally, to complete the picture a bit more, Mark Rein of Epic Games had this to say..."We couldn't care less about synthetic benchmarks. Real players play real games and in this case there is a real performance gain for a large number of users playing our game. The updated drivers that people are talking about allow properly-equipped Unreal Tournament 3 players users to use some amazing UT3 content that was specifically designed to take advantage of hardware-accelerated physics. So now that content works on high-end Nvidia graphics cards, as opposed to Ageia PPUs. This is a great thing as high-end Nvidia cards number in the millions, or tens of millions, compared to Ageia PPUs which obviously numbered far, far less. So Nvidia has done a cool optimization that allows their customers to get more value and performance out of their graphics cards. How anyone could possibly confuse this for a bad thing is beyond me."

+1
 
Meh. Hopefully Intel & AMD will pressure Futuremark to add Havok support in the same test later on. No biggie.
What difference will that make? HavokFX is dead and other Havok products run only on the CPU. I had thought Havok would be interested in porting to GPUs, but it makes little sense for Intel now since Larrabee is just a big array of vector unit enhanced x86 cores.

The level of Havok's commitment to GPU physics seems to have taken a huge step backwards:
... Havok and AMD intend to "investigate the use of AMD’s massively parallel ATI Radeon GPUs to manage appropriate aspects of physical world simulation in the future." According to David O'Meara, managing director of Havok, "the capabilities of massively parallel products offer technical possibilities for computing certain types of simulation. We look forward to working with AMD to explore these possibilities.
They already *had* HavokFX, now they "intend to investigate" and "explore the possibilities?" No wonder AMD's agreement specifically mentioned CPU physics. Havok doesn't seem to be giving GPU physics a priority anymore.
 
I wouldn't call this cheating at all. Using every ounce of your hardware to perform better? Absolutely. Cheating? No way.

Optimizing isn't cheating if the gameplay is preserved. Using a new beta drive to achieve better PhysX support? Fine. This is waaaay different than Quack.exe.
 
The problem with this driver is the fact, that it alters score. There are tons of people who judge cards based on these scores (yes, they are stupid, but they exist). Now with this driver it looks like NV cards with PhysX are superiour to ATI cards - which is actually not true.
 
with no physics = crappy fps
with physics = high fps

whats the problem?
don't get it

cheating would be if they renamed the exe and with physics is was till crappy fps.
 
The problem with this driver is the fact, that it alters score. There are tons of people who judge cards based on these scores (yes, they are stupid, but they exist). Now with this driver it looks like NV cards with PhysX are superiour to ATI cards - which is actually not true.
Read the article, and especially note Futuremark's reaction to Charlie's hissy fit.

The point you're missing is that Vantage can run with or without hardware PhysX acceleration. There is even an option to disable hardware PhysX, and the option still works whether it's PPU or GPU accelerated. While you narrowly redefine 3DMark as a GPU benchmark (despite there always having been CPU tests), it's allegedly some kind of system benchmark. Someone who wants to "catch up" in some pointless 3DMark race can easily add a faster CPU and a PPU. Mark Rein, after putting down synthetic benchmarks, said it perfectly about GPU physics: "How anyone could possibly confuse this for a bad thing is beyond me."

nvidia broadcast 4 months ago that it was porting PhysX to CUDA. Programs written to support the PPU will be accelerated by the GPU. Considering the limited hardware accelerated support and really no killer apps/games, I don't understand what all the CUDA/PhysX whining is about.
 
I am at a loss as to why anyone even cares whether or not someone cheats in 3dmark whatever anymore. :confused:

Due to Intel and Ati saying "no thanks" to PhysX, I have my doubts as to whether we will see more than just token implementations of gpu accelerated PhysX in future titles. However, at least we now know for certain that it can be done, and be done fairly well.
 
What do you guys think of this whole deal?

How come there aren't any allegations about Futuremark making useless benchmarks? Its like those standardized high school exit exams, even if they are cheating does it really matter?
 
Well, the thing is see cpu score, meaning, a amd system with X2 4400+ can beat a QX9770 then ? :S
"
he overall 3DMark Vantage score is P11720 with a CPU Test 2 score of 14.95 FPS. With GPU PhysX enabled, the overall score jumps to P14298 with a CPU Test 2 score of 127.77 FPS."

LOOL

I dont see this tech be standard anytime with single player, if it so true that intel will join in this ship.

Sooo, if I'm get this right, a cpu A64 X2 4400+ toledo(my 2nd comp, ofc with a dfi venus :D ) can beat a QX maybe with nv card + drivers, will some idiot buy a 939 system then ? :D
 
continued validation of real world real gameplya gaming reviews

Yeah, seems like something futuremark might get upset about since it undermines their benchmark numbers.

It's shady on Nvidia's part if PhysX only amounts to a "FutureMark driver", but we'll see.
 
It's shady on Nvidia's part if PhysX only amounts to a "FutureMark driver", but we'll see.
It's already working in UT3 too. The special PhysX levels (I know, yawn) get almost 3x speedups. It should also work in the regular game, but those gains are modest even with a PPU installed, and might disappear with GPU PhysX.
 
I'm a tad confused how its cheating, aren't they just enabled hardware support for a particular feature? Isn't this a GOOD thing??
 
Couting it on the CPU score doesn't make a whole lot of sense. If it were a seperate section of the benchmark specifically labeled to test a systems physics processing then it would seem a little more legit. But it's not.

Also, it looks like a pretty fundamentally flawed way of testing how good a gpu is at accelerating PhysX. Most enthusiasts crank up the eye candy until the framerates drop below what's tolerable, putting the card under it's maximum load. If this were run when Vantage was stressing the GPU to it's limits and THEN throwing the extra calculations at it it would be a better performance metric.

Looks like Vantage is being played The Way It's Meant To Be.
 
It is a good thing. Especially if performance improvements in UT3 are obvious.

This is a relatively new driver, give [H] some time to test this for us and investigate the benefit.

The jump in CPU score more than anyting is likely just a goof up in the 3dmark code and how it compiles the scores. When they wrote the bench there may not have been good examples of how to score it, and scale that score. If anything they just need to seperate the physics calculations form the cpu scoring and produce 2 scores instead of the one (apparently) combined CPU score. It's still relatively new stuff (physics acceleration) and only now hitting a widespread use with this driver and ut3 powered games. Not too suprising the benchmark is not quite ready to test it for us.

And although the synthetic benchmarks don't really give an idea of actual gameplay, they do allow comparisons among say different ati cards, or different nvidia cards and give you an idea of the performace differences. Since you are comparing one nvidia card to another and the system and drivers are the same, and all you are changing is the card, it can still produce a quantified comparison that approximates the differences in the hardware, or the differences in non-oc'd vs oc'd, etc. Obviously it wouldn't replace the real world gameplay testing that [H] does.

I only wish that the 3dmark tests could be put on a looping test that didnt stop every 30 seconds and say "scanning system" blah blah which interrupts the load on the GPU. but im getting off topic.....
 
Couting it on the CPU score doesn't make a whole lot of sense. If it were a seperate section of the benchmark specifically labeled to test a systems physics processing then it would seem a little more legit. But it's not.

Also, it looks like a pretty fundamentally flawed way of testing how good a gpu is at accelerating PhysX. Most enthusiasts crank up the eye candy until the framerates drop below what's tolerable, putting the card under it's maximum load. If this were run when Vantage was stressing the GPU to it's limits and THEN throwing the extra calculations at it it would be a better performance metric.

Looks like Vantage is being played The Way It's Meant To Be.

But that is a fundamental flaw with 3dmark, not with nvidia. nvidia is simply enabling GPU accelerated PhysX - they aren't doing anything special for 3dmark, so any resulting score increase is a result of 3dmark using hardware accelerated PhysX and has nothing to do with nvidia (other than the GPU now counts as a hardware accelerator for PhysX instead of just the PPU)
 
anyone with half a brain will realise that if the GPU is busy doing physics, it has LESS time to render
 
anyone with half a brain will realise that if the GPU is busy doing physics, it has LESS time to render
Exactly. Unless the video card has a dedicated on-board PPU, then processing physics should take performance down. The thing is, having those thousands of particles or objects with physics processed on the CPU alone would make it a slideshow, regardless of the game/test's graphics. That's why you can't do a proper comparisson. When PhysX is on, you get MORE particles, so you can't compare that to using no PhysX and LESS particles.

Also, this will not speed up UT3. Mark Rein clearly said it speeds up the special content (aka Tornado map and other PhysX tech demos). UT3 uses almost no physics. You've got particles, few of which bounce off other objects and then you have vehicles and characters when they die or fake death. PhysX on the GPU would really only help is really particle-heavy scenes so I don't expect any miracles in current levels.

Anyway, the Futuremark test, as usual, is just stupid, which is what confuses people. The test adds more 3D models and particles when you use a faster processor (be it a CPU, GPU or PPU) so it's inherently flawed. The problem is, the only fair way to compare CPU and GPU physics processing speed would be to make a test that's a slideshow on a quad-core CPU and would then run well on a GPU...or a test which runs at 30 fps on a CPU and 300 on the GPU. I don't know why they didn't just do that.
 
But that is a fundamental flaw with 3dmark, not with nvidia. nvidia is simply enabling GPU accelerated PhysX - they aren't doing anything special for 3dmark, so any resulting score increase is a result of 3dmark using hardware accelerated PhysX and has nothing to do with nvidia (other than the GPU now counts as a hardware accelerator for PhysX instead of just the PPU)

For me it's a little sketchy that it's an optimization enabled by nVidia's drivers and not Futuremark. I haven't heard of a single game where you get a 20% boost in performance from a PPU.
 
For me it's a little sketchy that it's an optimization enabled by nVidia's drivers and not Futuremark. I haven't heard of a single game where you get a 20% boost in performance from a PPU.

auto assault, city of heroes, warmonger, cellfactor, ut3, etc.
 
While you narrowly redefine 3DMark as a GPU benchmark (despite there always having been CPU tests), it's allegedly some kind of system benchmark.

Did you read what i wrote ? I'm talking about noobs, who use 3DMark as system comparison tool. And with this PhysX enabled driver, they will think that their 8800GT on E6400 is better than HD4870 with Q9450. Which is simply not true, who will higher framerates in all games ? The second one. Who will have higher ratings ? The first one.

I don't really care about these marks for myself, but even if it's Futuremarks fault for not disabling HW PhysX, it is still NV's responsibility to add exception on drivers side. They do zilions of exceptions and optimizations in driver for specific games, why not to add exception for this case too ? Because it lowers their score ?
 
Did you read what i wrote ?
Ditto.

I don't see the reason for any outrage over this. CPU test 2 was made by Futuremark to include PhysX acceleration. nvidia bought Ageia back in February. PPU (now) and GPU (soon) accelerated PhysX are available. * This * is * a * good * thing *. It has limited support right now, but it doesn't cost current owners of supported cards an extra cent for the feature.

-----
Earlier today nvidia announced another 15-16 PhysX titles are coming out by the end of the year. Hopefully some are good and make use of hardware accelerated PhysX. I'm not holding my breath.
 
Once again :
1) It is a fundamental flaw from Futuremark not to disable PhysX GPU acceleration in ther CPU test.
2) It is a fundamental flaw from NVIDIA not to disable PhysX GPU acceleration for Futuremarks CPU test.

It is a CPU test, not a PhysX test. Yes, it uses PhysX API and it can be accelerated by GPU, but it shouldn't be. That is the point, that is the problem. If they would call that test PhysX test, i wouldn't complain. But it is called CPU test, thus it is meant to be run on CPU, even if your GPU supports PhysX acceleration.

Yes, we can stop using 3DMark due this (i actually never used it for comparison), but unfortunately there are zilions of people who use it as a ultimate GPU comparison tool. And no, they don't look at SM2/SM3/HDR and these tests - they look only at final score, which in this case doesn't reflect the real power of their system.
 
Please just read the article. You're twisting yourself into a pretzel trying to ignore how Vantage works. Your problem with it seems to be due to just one reason (GPU).
 
Please just read the article. You're twisting yourself into a pretzel trying to ignore how Vantage works. Your problem with it seems to be due to just one reason (GPU).

Once again, the problem is not how i use Vantage (i don't use it), but how people interpret it. They can call it general benchmark, but that is not how people use it. People use it (unfortunately) only as graphics card benchmark. Many people don't care about gameplay, for them higher score in 3DMark or Vantage = better card (is 8800GT better than 4870 ? Don't think so).
Just look what info we got about 4850 before release date. 3DMark, then 3DMark, then Vantage, then 3DMark, then Crysis, then 3DMark.... No one did a Bioshock, UT3, Source engine, ... comparison. All they did was run 3DMark/Vantage to see how good is the new graphics card and posted total score everywhere.
And now we have a point, where that score in Vantage means absolutely nothing. If they will have this also in ORB certified driver, then we will have 2 separate categories for Vantage scores. One for NVidia G80/G9x/G200 cards (with PhysX), and another one for everything else. Total scores from each category won't be comparable.
 
cheating nope
making the vantage useless for comparisions hell yes

(or rather making i even more useless :D )
 
They already *had* HavokFX, now they "intend to investigate" and "explore the possibilities?" No wonder AMD's agreement specifically mentioned CPU physics. Havok doesn't seem to be giving GPU physics a priority anymore.

HavokFX wasn't a complete physics solution. It just used some DX9-level shaders to accelerate non-interactive 'eyecandy' physics such as particles.
What nVidia has now is a complete physics implementation, comparable to the full Havok library, running on their GPGPU interface.
Havok has never supported GPGPU at all, and I somehow doubt that Intel wants to add it so they can make AMD look good, and let people buy Radeons instead of Core2s for faster physics.

At any rate, the whole thing is rather silly.
The moment FutureMark decided to use the PhysX library, they deliberately chose to support both PPUs and CPUs for their physics test. nVidia didn't do anything wrong in making their GPU work as a PPU, and be compatible with the PhysX library.
The problem is that the benchmark uses the PPU test to get the CPU score. Apparently people get hung up over a benchmark score being called 'CPU', because it can be assisted by a PPU or GPU.
The only issue here is that Futuremark didn't name their tests right, and/or didn't design the right tests, depending on how you look at it.
 
As for the physics + graphics argument...
I suppose people don't realize that there actually ARE graphics in the physics test in 3dmark Vantage?
Sure, they're relatively simple, but already the GPU has to do both graphics and physics, and apparently it still gets over 120 fps, so it can balance well between physics and graphics workloads.

This is a LOT faster than what any CPU can get (apparently nearly 10 times as fast), because the physics in their test are way heavier than in any game (it seems that a lot of people think that physics are physics, but CPUs are incredibly poor at physics).
So we can say that if the GPU would spend roughly 10% of its resources on physics, it could offload the CPU completely, and still have 90% of its resources left for graphics.
Additionally, when doing graphics, the efficiency is not 100%, so there will be some unused resources. It could well be that there are already 10% unused resources in most games... So that would mean that there's a whole CPU's worth of physics power going idle currently.

What this comes down to is that offloading the CPU physics to the GPU may have little or no impact on the framerate... And when you were previously CPU-limited, you would actually get a higher framerate by offloading the physics.

Ofcourse an alternative is to use much more complex physics at a slightly reduced framerate or simpler graphics, for games that were previously not possible on a CPU.
So this could be a big step in gaming. It will use your GPU more efficiently, it will free up resources on your CPU, and it allows for new levels of realism in physics.
 
In the physics tests the gpu is being used as a cpu, that's why it's in the cpu section. What do you think CUDA does? If a game was to come out with physics requirements like the futuremark cpu test then it would go that much faster with gpu accelerated physics, no problem there.

What I do have a problem with is the fact that the graphics card can't do both rendering and physics at the same time, hence future mark should really bring out a combined test with both intensive rendering and intensive physics. This is not nvidia's fault however, and I suspect by the time they do you'll be able to run a second graphics card solely for physics - i.e. I buy a next gen card to do graphics, and then stick my old 8800GT in my second * 16 slot and use it as an uber physics accelerator.
 
Back
Top