Originally Posted by dantoddd
Thanks for the response. But I'm still a little confused. I was under impression that Physx effects like cloth simulation & smoke, not physics, can only be enabled on Nvidia systems. So how can you even compare nvidia performance to Amd performance when there is Physx involved
You can compare them with GPU PhysX effects turned off, at which point it should be like comparing the video cards in any other game. With GPU PhysX enabled framerate should theoretically drop on the Nvida system, as the GPU that is being used for rendering now has to render effects created by the other GPU that is calculating the physics for those effects.
Generally I would say that games using GPU-enabled physics effects run well on Nvidia cards because Nvidia does a lot of work with developers to make the games run well on their hardware. At my last job Nvidia said, unofficially, that if we supported and touted their new 3D stereoscopic glasses and monitors with our game they would help make sure we had support with our graphics setup, fortunately we declined and took care of it ourselves, I was glad we didn't advertise for one company or the other. They would've given the same offer for PhysX however we were too far along with development and already using Havok, which is quite a bit more efficient anyway, it just lacks GPU acceleration. Havok is efficient enough to do plenty of physics effects on the GPU, we were simulating hundreds of objects per frame with physics updates under 1ms, leaving 15.6 ms for everything else to still get 60 fps, and of those 15ms, about 10 were used for rendering tasks (on the CPU), Havok also supports things like cloth without a GPU, Nvidia has incentive to make some things GPU-only so they can sell more GPUs and hurt AMD by getting proprietary techniques into games.
IMO physics on the GPU is pretty gimmicky until GPUs can be accessed for queries and work just as fast as a CPU can, but a the moment asking the GPU for data takes about 30-45ms, whereas querying for information on memory in the stack (cache), or even system memory, is hundreds of thousands of times faster in most cases. Until the time when there is almost no late cy between CPU and GPU it will just be a little extra visuals at the cost of another entire GPU.