Physics on a chip or future GPU function?

[jpr27]

Just a quick question:

With the news of AGEIA and other focuses on the future of physics, which option in form of performance would be a better solution and why?

A. A seperate chip with its own distinction like a GPU / CPU?

B. Intergrated into current GPU in hardware form similar to T&L etc?

C. Leave in software form and let CPUs take care of it. Dual core CPU's are coming and they will be able to handle physics.

D. None of the above...(Your thoughts on direction of physics handling)


I know cost is always in the equation but from a performance standpoint, so im looking at performance aspect.

Personally I see B being the best option for performance but of course "Cost" would probably be prohibitive and not feasable in the near future.

I also think that upcoming Dual Core CPU's can greatly improve physics (If coded to take advantage) and probably will be for the short term. But then again with A.I. evolving (hopefully a MAJOR evolution soon), we will need that processing power again.

Your Thoughts?

[rwolf]

A card with one graphics chip and one physics chip and shared memory.

The drawback with AGEIA is that you have to move all the data from main memory to the physics card and them move it to the graphics card.

[Karma Police]

Yeah, but if it gives a 100 times the performance gain, who cares? (except programming difficulty)

[jvd]

http://www.beyond3d.com/forum/viewtopic.php?t=21450

read this

[nutball]

Ask yourself who is in the position of having more transistors than they can sensibly spend on useful new features. Intel/AMD, or ATI/NVIDIA?

[rwolf]

Yeah, but if it gives a 100 times the performance gain, who cares? (except programming difficulty)

Try the exploding building demo and you will see.

[MuFu]

http://www.beyond3d.com/forum/viewtopic.php?t=21450

read this

[Karma Police]

Try the exploding building demo and you will see.

Yeah, that's what I meant. I'm definately in favor of a dedicated PPU in addition to the CPU/GPU.

[jpr27]

So no one really thinks that Dual Core (or more) CPU's will fair well maybe if its even in the short term?

I agree to some extent but with physics. I think integration to the GPU whether PPU or other form would be better. In the the long term the future of CPU should be reserved for A.I. I'm hoping that in the near future or even the next generation of games and consoles, that developers really push A.I. to its limits. With dual core cpu's at some time becoming mainstream it would seem feasable. I would actually like to see large leaps in A.I. over physics at this point. Then again its only an opinion :)

Thoughts?

[Frank]

So no one really thinks that Dual Core (or more) CPU's will fair well maybe if its even in the short term?

I agree to some extent but with physics. I think integration to the GPU whether PPU or other form would be better. In the the long term the future of CPU should be reserved for A.I. I'm hoping that in the near future or even the next generation of games and consoles, that developers really push A.I. to its limits. With dual core cpu's at some time becoming mainstream it would seem feasable. I would actually like to see large leaps in A.I. over physics at this point. Then again its only an opinion :)

Thoughts?

Agreed. Although I agree with nutball as well, that the CPU has much more transistors to spend.

[hoom]

I'm hoping that in the near future or even the next generation of games and consoles, that developers really push A.I. to its limits.
Given that AI is notoriously hard -thus expensive- to do well and that most of the budget goes on the visuals/art, that seems unlikely.

[jpr27]

Well Arrse if that is the case then im very worried about the future of gaming. I see graphics ever evolving and physics to a degree (though slower) also evolving. A.I. in retrospect I feel is falling behind in a big way. I know its not feasable to have A.I. to the lvl of a human etc but Im speaking more to the interaction between say NPC and environments and situational behaviors etc. Again I know the cost to bring A.I. to the forefront takes alot of power but I would rather see some serious progress in the A.I. versus improving graphics/physics. (Graphics improvement is already a given).

Nothing worse then having graphics getting closer to realism (Far from it now) but having the A.I. as smart as an elegant painting in comparison. :?

As for developers pushing it, I hope MS was serious when they spoke of taking not only physics but A.I. to the next level with its next console (I believe XNA to some degree is helping with this?). I think the time is right (with next generation consoles / Dual core Cpu) to really push the A.I. envelope.

[hoom]

Oh, I'd love it to happen, just I'll believe it when I see it :?