Will R 5xx GPUs be usable for physics?

[Anon Lamer]

Since AMDATI now has a GPGPU product that uses a R580 core, can that be interpreted as the R 5xx generation being usable for accelerating game physics? From what I understand such acceleration will only be possible on DX10 since the GPU is a shared resource...?

 

[rwolf]

By the time it is used by developers people will be using DX10 cards.

 

[AlexV]

Umm, no, you could do it on R5XX. The question is, will it be done in a timely manner?I still see it as an interesting alternative to getting a new DX10 card-I`d like to keep one of my X1900s and use it as a physics accelerator, alongside a SLi 8800GTX rig>

 

[pax]

Damn never thought about that... yeah my x1900xt could very well outperf the ageia... Short of a new mobo Id have to forgo crossfire tho but maybe the r600 will be goo enough for the 1-2 years the next rig will last. Havent really seen the need for sli\crossfire yet even on this 24" lcd...

 

[Anon Lamer]

By the time it is used by developers people will be using DX10 cards.

Well yes but does dx10 cards bring any new functionality in shader model 4.0 thats absolutely necessary for physics? Otherwise it seems that dx10 is necessary if you want to use your (single) GPU for both physics and GFX rendering at the same time. But if you have two cards physics acceleration should be doable on dx9.... ?

 

[swaaye]

I think the CPU will be vastly more popular for physics. You guys are looking at really, really niche setups. Might be optional but I imagine little, if any, gain.

 

[Anon Lamer]

I think the CPU will be vastly more popular for physics. You guys are looking at really, really niche setups. Might be optional but I imagine little, if any, gain.

From what I've read I've gathered that physics is very suitable to parallellization. So GPU>> CPU.

 

[AlexV]

Some aspects are highly parallelizable, others aren`t, so it`s not as clear-cut as that. I`m not sure in which boat I am on this, even though I`d lean towards a CPU implementation, given the fact that it should be a tad bit more flexible, IMO, and targeted towards a larger audience. And seeing the quite sorry state that programming is nowadays, I`d stick with traditional approaches, like the ones done on CPU