OpenCL is very interesting to run physics engine on. It is very scalable and runs on CPUs, GPUs and even DSPs when OpenCL-capable.

Aerodynamics simulation is possible, to have proper CFD you need dual precision floating point operations and many of them simultaneously.
An AMD HD 5800 series GPU should be fairly capable of doing CFD, not to mentioned nVidia's Fermi.

The problem is that you will need brand new hardware to run this. Which doesn't really fit Live for Speed since that is a very light optimized software package.
Maybe it is best to add OpenCL support now, but to add more demanding physics in the future.
Multithreading should be integrated in the game as soon as possible.

PhysX isn't interesting anyway. CUDA is a proprietary language and will be superseded by standard API's like DirectX compute shaders and OpenCL which will work on all GPU's.

@GobLox: You don't need continuous synchronizing between clients and servers or huge data transfers.
There is a lot possible with partial prediction and low-level synchronizing.


Multithreading is a must though, Valve Software has proven it doesn't have to strain singlethreaded systems and can increase performance in huge factors.

Last edited by SgtH3nry3, Thu 25 Dec 2008, 23:55.