Aegia physics and rendering

[K.I.L.E.R]

I'm assuming the state of physics objects will be stored and calculated within the card itself.
Now if I wanted to do a particle system that is all calculated on the PPU and I want to render it to my video card, I assume that this is possbile?

Accessing states from the PPU by parsing a buffer to it and returning one I could get the positions of every particle?
I would love to get my hands on one of these cards.

 

[Blazkowicz]

you can get the ageia physx SDK and try doing stuff.
(physx is both the name for the PPU and the API it seems. without the card work is done by the CPU)
I assume that either way you do your things via the API anyway !?


wait : I don't get if your question makes sense . off course you can get the position of particles, else it would be very worthless isn't it?
PPU simulates particles but keeps the results for itself?
I guess this will go to system RAM first, then your app asks nicely the GPU to render the particles.
well, stupid trivial stuff I say. Or do you think of some DMA transfer between PPU and GPU? that would be quite interesting

 

[K.I.L.E.R]

So far it was claimed that the PCI-X bridge will be used.
I would have thought that you could send a command to the PPU to send the positions of the particles directly to your video card without copying to RAM first.

It would make sense to do especially so if people without PCI-X would use AEGIA.

 

[Blazkowicz]

by PCI-X you mean PCIe bus?
I don't think the GPU would be able to receive random data from the card.
the PPU wouldn't know how to talk to each specific GPU if even that's possible.
really, you're wanting the PPU to ask the GPU to render.

IMO it needs to get to the driver. and to get to the driver it has to come from the 3D API (D3D, OGL). and for that data comes from your program, so from system RAM.

or you'd have the PPU to run the opengl/D3D APIs and GPU driver, and run your renderer.. so the PPU would have to be a full blown CPU?

really I think you've done some brainfart and you make me brainfarting much as well
(I need to go sleep ..)

 

[Frank]

Waiting for results from the card will kill your framerate any way. Nice for eye candy, otherwise that card needs to do the scene management by itself. And you ask it to perform a change.

Quite a difference compared to the current way. Developers are allowed anything they can code, now. They would have to see about that: is it supported?

But then again: that would only be one more layer/limitation. It might simplify and steamline things.