Originally Posted by jlippo
PhysX doesn't use DirectX for anything, it is calculated using Cuda.
You can be sure that 4xx series is best in terms of performance due to the better arcitechture for computing.
I'm sure the 480 is the fastest card for processing Physx, but when you want to move the Physx load to another card, you want something that won't be a total waste, so obviously you aren't going to run a second 480 for dedicated Physx. That's why I mentioned the GTS 250.
Hopefully nvidia will release a "perfect" dedicated Physx card using the Fermi architecture before too long. Maybe 128SPs, and high clock speeds in all the right areas, low(ish) power requirements and extremely low idle temps and power consumption.