[fb III IX ca IV]

Nvidia intentionally uses outdated x87 (modern processors are not optimized for them) instructions in the CPU version of PhysX rather than SSE instructions, which run very fast on CPUs from within the last 7-8 years. They do this to artificially slow down the CPU implementation to make it look like their GPUs are much better than CPUs at physics simulation, when in reality, you can get comparable results from using a competing physics library on a spare CPU core.