Nvidia is trying to convince the world that PhysX is the best way of doing in-game physics, but it still has to make this physics approach much more attractive than today.

PhysX needs much wider game support, and once Nvidia convinces the world that there is much sense to accelerate physics on a GPU, the world will follow.

Havoc is currently the dominant way of doing physics. It has support for 300 games, and growing, but it can only be accelerated on a CPU. ATI supports Havoc, but for the time being ATI doesn’t run Havoc on its GPUs, but rather optimizes its code to run Havoc faster on Intel and AMD CPUs.

Common sense is telling us that running Havoc on a GPU should be much faster, as Shaders are independent pipelines or cores and you have quite a lot of them on a modern GPU.

ATI says that it plans to accelerate Havoc on a CPU, while for the time being ATI says that it will also investigate if it makes sense to run Physics on a GPU, and that it will do it when it makes sense. This probably has a lot to do with a fact that Intel owns Havoc, but doesn't really have a GPU to run physics on.

To spice up this already complicated story, sources close to Nvidia believe that Havoc will only be accelerated on Larrabee graphics parts, and that ATI will never get a license to do Havoc physics on its GPUs. We cannot confirm or deny this claim, but it sounds like a good conspiracy theory.