Original post by smitty1276 I think that the dedicated PPUs will probably end up finding a home in military simulation, etc., where their costs aren't as noticable. I think, in the end, we are likely to see something like DirectPhysics added to DirectX, and that it will transparently use the GPU or, if present, a PPU... making a proprietary PPU API sort of a non-issue.
I was led to believe that the PhysX API does this already.
I was led to believe that the PhysX API does this already.
I think it does, but if I understand correctly it will only use the Ageia PPU, I think. I'm not real familiar with it, but I don't think that the PhysX API will utilize extra horsepower on your GeForce8800, for example.
I can imagine a scenario, though, where Microsoft releases a DirectPhysics API, which will use your Ageia PPU, or another manufacturers PPU, or uses the GPU to pick up the slack in a manner that is transparent to the developer. I think the future holds "DirectX 11.0 compatible" physics cards... or something like that at least.
Isn't the problem with GPU-based physics that GPUs are optimized for one-way communication, and thus are best for physics calculations that don't affect gameplay, but merely visuals? If GPUs in the next generation or two address this, then I could see GPUs being used for physics quite heavily. Barring that, I'm guessing that multi-core CPUs have the best future for physics. I agree with SimonForsman; the advantage the PhysX cards have is significantly improving gameplay mechanics, but it's hard to design a game around gameplay mechanics that need hardware acceleration for physics, and expect the game to make any money in the current market. And it's hard to design a game around gameplay mechanics that can make excellent use of hardware acceleration for physics, but still plays well without it. If it really makes good use of the acceleration, then the gameplay with acceleration is going to be noticeably different from gameplay without acceleration. If it was different enough, then they'd just drop the requirement for acceleration, and few people would care to buy the card.
"We should have a great fewer disputes in the world if words were taken for what they are, the signs of our ideas only, and not for things themselves." - John Locke
I believe that the GPUs don't have enough branch ability to be useful for a number of intense physics (and AI) operations that a chip like the PhysX can do. The actual bigger competitor for Ageia is the quad-core CPUs, and/or the SPUs in the Cell, rather than the GPUs, IMO.
Of course, the mind share is all about GPU physics, even though that's not really the right way to do the end-to-end solution, again IMO.
