NVIDIA's PhysX: Performance and Status Report

[Rob Williams]

PhysX is getting a lot of attention right now, but the reasons vary wildly. Since we haven't taken a look at the technology in a while, this article's goal is to see where things stand. We'll also be taking an in-depth look at GPU PhysX performance, using both 3DMark Vantage and UT III.

You can read the article here and discuss it here.

HOPEFULLY ATI HAS SOMETHING UP THERE SLEEVE SPECIALLY WITH 3XCROSSFIRE AND CROSSFIRE X,THINKING MABEY 2 4870 X2S WHEN THEY COME OUT AND A 3850 FOR A PHYSIC PEOSSESING UNIT....OPPS THATS 5 GPUS HEHEHEHE SKYS THE LIMIT

Very interesting article, has anyone given it a shot on SLI to see if two (or three) cards can handle it better than one?

Some guys already got PhysX running on ati cards, (at NGOHQ I believe). And teh rumors are they're gonna license it from nvidia, too. So it seems this *should* work out for all

The argument about the score inflation is that it's different from a dedicated card since in the real world the GPU processing is reduced for the physics calculations (I don't know if it is or not). Plus the point of a CPU test is to test the CPU and remove GPU influence. They're saying it's not pure.

[Rob Williams]

Originally Posted by Unregistered ANTHONY

HOPEFULLY ATI HAS SOMETHING UP THERE SLEEVE SPECIALLY WITH 3XCROSSFIRE AND CROSSFIRE X,THINKING MABEY 2 4870 X2S WHEN THEY COME OUT AND A 3850 FOR A PHYSIC PEOSSESING UNIT....OPPS THATS 5 GPUS HEHEHEHE SKYS THE LIMIT

I questioned NVIDIA about their plans for something like that, but didn't get a response before publishing time. I'm pretty sure both companies have talked about it though. Having SLI and a smaller dedicated GPU (probably one you already had that you don't want to rid) being put to physics use would be great.

Originally Posted by Plague

Very interesting article, has anyone given it a shot on SLI to see if two (or three) cards can handle it better than one?

As ridiculous as it may seem, I didn't have doubles of any card here that could support PhysX, which is why it was not tested.

Originally Posted by Unregistered

Some guys already got PhysX running on ati cards, (at NGOHQ I believe). And teh rumors are they're gonna license it from nvidia, too. So it seems this *should* work out for all

The argument about the score inflation is that it's different from a dedicated card since in the real world the GPU processing is reduced for the physics calculations (I don't know if it is or not). Plus the point of a CPU test is to test the CPU and remove GPU influence. They're saying it's not pure.

As I mentioned in the article, nothing would stop ATI from being able to use PhysX on the GPU. It's just a matter of getting permission from NVIDIA, and that patch from NGOHQ will not be pushed by ATI in any way. If ATI -does- license it from NVIDIA, it would be a huge win for PhysX.

Regarding the inflation... it's a very good point. Still, I'd blame Futuremark before I'd consider shifting the blame to NVIDIA. NVIDIA stuck within the rules. Futuremark weighed the physics aspect way too heavily, which is why we are seeing CPU scores of 30,000+. Really, for the test to have been accurate, it should have been actual gameplay, not a severely throttled back game environment, where it doesn't push the GPU at all.

I have little doubt that the next version of Futuremark will be a little different, especially with these new methods.

What I think the reason is why GPU physX is overinflated is , that in the cpu2 test it can use 100% of the GPU processing capability. This is a situation that in no way represent the situation as present in a game. I assume that when you play a 3d game e.e UT3 with physX enabled, only about 10 - 40% of processing capability will remain for running physX tasks.. Which means that the processing power as shown in the CPU2 test might be from 3 to 10 x as high as in a real game situation.


Difference with the PPU approach is while it might have a lower performance in the CPU2 test, ALL of its power is available in a game. Which I think shows in the graphs where the PPU version results in higher frame rates.

I have nothing against the running of physX on a GPU and I hope it takes well soon. But this move of nvidia to inflate its benchmarketing score does definitely not feel right.

One thing I am wondering about, if AMD does not accept /join physX, would it be possible to run physX on a cheapo nvidia card in future, while the ATI card(s) take care of the graphics and Havok physics??? What do you guys think...

Originally Posted by Unregistered

One thing I am wondering about, if AMD does not accept /join physX, would it be possible to run physX on a cheapo nvidia card in future, while the ATI card(s) take care of the graphics and Havok physics??? What do you guys think...

If you're going to do that why not just get a PPU? They are pretty "Cheap".

Vista only allows one display driver, XP will allow two.

[Kougar]

The original PhysX special effects and physics was to much for a CPU to handle, especially with the graphics driver overhead it also had to compute for the game.

About running ATI+NVIDIA GPUs at once... you can do this for Folding@home's GPU clients believe it or not. Users can fold on both a 3870 and a 9800 and the CPU all at the same time, assuming they have a quadcore. So it might actually be possible to run PhysX on a secondary NVIDIA GPU while gaming on the ATI GPU...

[davidm71]

Just wondering if anyone has any idea what your fps would be like if you have two video cards in SLI? I have two 8800GTX video cards and waiting for nvidia to release new drivers they promised for physx. It seems like sometimes sli advantage is less than expected so wondering what it would be the difference comparing SLI config to Physx on the second card?

Dave.

[Rob Williams]

Originally Posted by davidm71

Just wondering if anyone has any idea what your fps would be like if you have two video cards in SLI? I have two 8800GTX video cards and waiting for nvidia to release new drivers they promised for physx. It seems like sometimes sli advantage is less than expected so wondering what it would be the difference comparing SLI config to Physx on the second card?

Dave.

Your performance would likely be better than what I saw in my testing. Sadly, I was unable to test dual-GPU due to not having double GPUs of any supported card. Once the new driver is released, I'll take SLI 8800GTS 512 for a spin and release a follow-up article.

The impression i'm left with when reading this article is that GPU-enabled physx in gaming is awesome, and yet the best solution for physx-enabled gaming with your test rig is a dedicated PPU.

Surely that is the conclusion to be made from these results. Or put another way, "Nvidia GPU-assisted physx is not as good as a dedicated PPU".

Sure, the new drivers make physx processing 'free' to those with decent Nvidia GPUs, but a dedicated gamer who is considering what to invest his or her money in, may well find buying a PPU to be the best bang for the buck.

Well, how about buying latest NV GPU and keeping the old one just for physX ? Will different models be compatible (one for graphics and the other for physX) ?

Keep in mind that these results are posted while using the GPU to both render graphics and calculate physx.

performance will undoubtedly go up in theory if theres a second card to calculate physx.

unfortunately, there is no substantial evidence yet if a second card will strongly boost physics to the same level as a dedicated ppu or the fact that the gpu will be just as fast as the ppu currently in production

[Rob Williams]

Hi guys,

Just a note that this article in no way concludes the testing for PhysX I have planned. When the new drivers come out with support for older GPUs, things are going to be a little interesting. At that time, I'll be able to test the benefits of adding in a second card for SLI (8800 GTS 512). I'll also elaborate with more testing using higher-end cards, such as the GTX 280.

I've also heard back from NVIDIA regarding the "spare GPU" issue, and that is apparently their goal. Eventually, drivers should support the ability to designate a certain GPU for PPU purposes, which would effectively allow you to use one of your "hand-me-down" cards for the physics. I'm not sure how long this will take to become an option, though.

I have a PhysX PCI card & a GTX 260 so I set the PhysX properties from GeForce PhysX to AGEIA PhysX and hit Apply but when I start UT3 the PhysX card's blue light doesn't come on & I don't hear it's fan spin up & my CPU pegs @ 100%.

Am I forgetting to do something to activate the PPU? When I go into the PhysX properties the blue light comes on and turns off when I close the window.

When I 1st put it in my machine it seemed to install itself w/o the hardware wizard (the nVidia 178.24 drivers were already installed).