AMD Director confirms
CPU and GPU acceleration of in-game physics is something that AMD’s graphics part, ATI has in mind. Godfrey Cheng, AMD’s Director of technical marketing, graphics product group, has confirmed that this is the path that the company wants to take.
AMD just demonstrated its Cloth and Destruction demos based on the Havoc engine, and AMD was able to balance the performance between CPU and GPU. Godfrey confirmed that AMD believes in open standards such as OpenCL, and therefore AMD was happy to demonstrate graphics accelerated physics with the Havok engine.
The funny part is that Intel owns Havok and this cooperation between Intel and AMD works really well, at least on a technical level. OpenCL lets AMD choose if the physics should run on GPU or CPU and demos even had toggle scrolls between CPU and GPU acceleration.
AMD believes that for a lot of physics effects the CPU should be just fine, but for massive parallel operation that you get with cloth simulation or destruction you need a GPU. As you probably know by know, a GPU can run many things simultaneously and we use to call the parts of the GPU pipelines, but nowadays the term shader is more appropriate, and for example Radeon HD 4870 or 4890 has 800 shaders that theoretically can run 800 different operation simultaneously, at least in a perfect world.
So for ATI currently Nvidia’s PhysX and its proprietary Cuda will stay Nvidia’s, while ATI will concentrate completely on GPU acceleration using Havok engine. Many believed that Intel will never let ATI to accelerate Physics on a GPU but hey, as you can see it has happened and game developers are the ones that should drive this further.