NVIDIA, PhysX, and the "C" Word
Calling Collect to Canada
I thought I would get the opinion of the folks up north to see if they thought NVIDIA was cheating. I was able to set up a conference call with several of the technical marketing guys, and I was able to chat with them for a little while about the situation.
Their primary concern was that a 3rd party benchmark should be “agnostic” when considering how to run it. It took me a while to entirely figure out what “agnostic” meant in this situation, but I’m pretty sure I can sum it up here. Basically what they are saying is that by NVIDIA enabling GPU Physics, it is giving the NVIDIA setup an advantage that cannot be overcome by an AMD card in the same platform. So for instance, if the NVIDIA card was tested on a platform with an Ageia PPU installed, then the results will be more directly comparable to an AMD card running with a PPU. But because NVIDIA now owns Ageia and PhysX, they do have an advantage in benchmarks which can access the PPU functionality.
I see where they are coming from, and in many ways I agree. Even though a tested AMD card may in fact compete at the same level as the NVIDIA card at the same price point, the NVIDIA card will have a higher 3D Mark result because it can act as the PPU in a manner which is not reflective of a real world situation. CPU Test 2 does limited rendering which does not push the GPU and leans heavily on the CPU, and if there is a PPU present then it will use that. So when comparing apples to apples, the NVIDIA card will always look better in that test when you look at overall score as well as CPU Test 2.
CPU Test 2 is really a theoretical test, and it will never be used in any way, shape, or form in a real application. So in many ways it is an unrealistic representation of what we would see in any game. From their point of view it is one thing to show off the GPU enabled PhysX using Unreal Tournament 3, but it is quite another to see GPU accelerated physics in a benchmark which is an unrealistic indicator of performance in an actual game.
Will AMD get on board with PhysX perhaps?
So far AMD has not participated with NVIDIA to develop GPU physics using the PhysX middleware. AMD has repeatedly said that they have heard nothing from NVIDIA, and had not been invited to participate. What AMD is doing is working with Havok, which is now owned by Intel. So far AMD is very happy with what they have done with Havok, and they are actively working with Havok to help enable their “Cloth and Destruction” technologies to potentially work on the AMD GPU. Nothing official has been announced yet, but this does appear to be the direction they are heading.
AMD does feel that there needs to be an open relationship, but I am not entirely sure how they are going about making this happen. As a funny side note, I had brought this subject up with Roy Taylor at NVIDIA. He said that so far nobody from AMD has tried to contact him about joining in with support for GPU accelerated PhysX from the AMD side. And furthermore all they had heard from AMD was that they had complained, “Nobody from NVIDIA has contacted us about joining in.” At which point another person on the NVIDIA call commented, “Well I saw this really attractive girl at the dance, but she didn’t come over to talk to me…obviously she hates me!”
So I am here to put, in writing, a message from NVIDIA to AMD. Richard Huddy, please call Roy Taylor at NVIDIA ASAP. He is awaiting your call.
PhysX in Actual Games
To get a developer’s point of view, I was able to contact Mark Rein from Epic Games to see what his opinion on enabling GPU accelerated PhysX on 3D Mark Vantage. Mark was exceptionally nice with this interview, as I called around 6pm EST and I’m thinking he was in the middle of something non-work related. So my sincerest apologies to Mark, and I certainly do appreciate the time you gave me for this interview.
I explained the situation as best I could without trying to throw in any questions that could have lead Mark one direction or the other. Once I finished explaining best I could, Mark essentially replied, “What the hell is wrong with that? Is there really a problem here?”
As we spoke his feelings essentially were made known with a refreshing stream of 4 lettered words. This is not a complaint mind, as it was nice to get away from the buttoned-down collar world that I had been immersed in throughout the afternoon. His candor was greatly appreciated.
PhysX in Ghost Recon
His basic opinion was that NVIDIA was doing a favor to both customers and to developers. NVIDIA is giving customers the ability to experience content which will utilize these advanced physics. Developers will have the option of including these advanced effects without having to worry about the 100,000 people (or less) who actually bought the Ageia PPU. With one driver revision, NVIDIA will create a user base of millions which can enjoy GPU accelerated Physics (I am assuming GeForce 8800’s and above will get this enabled sometime in the near future).
He also brought me back to Earth in that 3D Mark Vantage is a benchmark. Who exactly plays with their benchmark? Do users sit for hours on end just watching the benchmark run over and over again? Of course not. So for Mark, seeing a slightly higher value on a benchmark is not a big deal. Where the real kicker comes is seeing these graphics cards play games with actual GPU accelerated physics. That is where we will see the real difference. If an application comes out that really utilizes GPU physics acceleration, then the differences between gameplay with and without physics will make the small increase in a benchmark score seem meaningless.
So he obviously does not feel that NVIDIA is cheating, as they are simply uncovering a feature that was present in their product.