Is Ageia's PhysX Failing?

So, Cell Factor Doesn't Need a PhysX Card? Continued

In May, a post went up over at Hardforum. Lowrider007 posted screenshots and benchmark results from his system. This activity in the community is what we have been expressing to ourselves, as our US and German test labs have been looking for proper ways of testing the Ageia PhysX processor.

Our previous Ageia’s PhysX processor article looked at Ghost Recon Advanced Warfighter, which utilizes special software paths that allow the game to process additional physics simulations. We concluded that the additional effects physics in the game were not worth the cost. Obviously, the community agrees that forking over $300 for a piece of hardware that does not positively affect their gameplay or give them an experience above that of the regular game effects is not worth it. Many gamers can spare enough money for a good graphics card, but feel that physics is secondary - no matter what the cost.

Last year, we got a glimpse at graphics processor powered physics at ATI’s tech day in Berkley, California. We only saw a water simulator at that time, but the fact that ATI was working on physics excited us. The initial demonstration was on a CrossFire platform with one card rendering graphics and the second simulating physics. Almost jokingly I asked the presenter when ATI was planning to come out with a PCI, PCI Express x1 or x4 R520-based graphics card. The presenter said "Why would anyone want that ?" I stated that hardcore gamers would want to use both cards in the x16 slots for rendering and another, a third card, for physics calculations. Interestingly enough, we saw that ATI showed off its interpretation of physics on a third graphics card during Computex (hey, where are my royalties ?). This is a cheaper choice for physics, as the entry-level model, which is a Radeon X1600, can be purchased for under $150.

Here is a slide from ATI demonstrating that during particular picoseconds there are shader unit cycles that are not being utilized.

Here is another example showing the vertex and pixel shader unit utilization for a single frame. Obviously, there is some room for effects physics calculations and a unified shader core with additional geometry shader units.

Soon after our visit to Berkley, Nvidia announced SLI-based physics, and the community agreed that this would be a good thing. SLI was becoming even more accepted, as each new series of graphics cards made SLI an easier purchasing decision. With each card launch, older cards came down in price and today gamers can purchase two graphics cards for less than $400. The best part about physics that is executed by graphics chips is that even if a game does not utilize physics acceleration, the gamer can still enjoy having both cards rendering frames in SLI mode.

Test Setup

We wanted to know if there is a difference in 3D graphics experience with and without a physics accelerator card. To find out, we took graphics cards from across the spectrum to see whether or not there is a possible advantage to newer cards over older models. The most recent generation of high-end cards is represented by the XFX GeForce 7950GX2, a PNY GeForce 7900GTX and a Connect3D Radeon X1900XTX. We also included an EVGA GeForce 6800GS and a Shader Model 2.0 ATI Radeon X850XT. We ran each card on the same platform, which consists of an AMD Athlon 64 FX-60 dual core processor running at 2.6 GHz and 2 GB of DDR400 system memory.