During testing, I became pretty curious about PhysX performance with dedicated versus shared PhysX processing. I was testing two games which utilize the technology, Mafia II and Metro 2033. I’ve read a lot of questions about PhysX, and many center around the debate concerning dedicated PhysX. So I decided to do a little extra bit of testing on this.
Dedicated vs Shared PhysX Processing:
PhysX comparison - minimum FPS
PhysX comparison - average FPS
As you can see here, with three GPUs in use, the two games behave completely differently. Mafia II seems to have very heavy usage on the dedicated PhysX GPU, where Metro uses a dedicated GPU very little. Since Metro is so heavily GPU bound, the loss of a card dedicated for PhysX really hinders performance. Mafia II on the other hand gets a very nice performance boost when one of the three GPUs is dedicated for PhysX processing.
When dropping the system to only two GPUs, the Mafia II runs just as well with the dedicated card as it does with the PhysX shared. But again, since Metro uses so little PhysX, the loss of an entire GPU for the job really hurts performance.
Our conclusion on this subject is that not all games with PhysX are equal. I would submit to Nvidia that they need to implement the PhysX configuration into the application profile area. If I were running a triple card setup, I would want PhysX shared for most stuff, but dedicated for games like Mafia II which need the extra PhysX power. Manually changing it each time is tedious at best.
Bookmarks