Displacement mapping with shaders

[Sage]

i think that radar is thinking of doing physics on the GPU... which is entirely possible, as we have seen water simulations on GPU's in which a character can swim through the water and leave a wake. granted, that's an effect done for the graphics engine but it's still doing physics calculations.

 

[Ostsol]

That's true, but in that case you still have to find out where the collision is occuring. In the pixel shader variants of this effect, the surface is perfectly flat, making it the point of collision a 2d matter. When the surface is oscillating, you need a collision hull of some sort. I agree that in such a dynamic case you could try an extract the tesselated mesh for the collision hull, but the matter of the cost of that operation must be considered. Would it really be less costly to pull that data across the bus? Or would it be cheaper to mirror the operation in software?

 

[AndrewM]

You can do a lot of this stuff on the GPU, which looks all flash and kewl, but if you need the results to interact with your "game" in any way, you're SOL.

PCI-Express could change this a few years down the track.

 

[flick556]

I don't know about pci express but when I even get the idea of pulling something useful back across the bus speed dies.

As for physics most good systems I have seen use some sort of voxel or other volume based approach and generate the actual geometry to be rendered after the physics pass.

 

[Sage]

I don't know about pci express but when I even get the idea of pulling something useful back across the bus speed dies.

As for physics most good systems I have seen use some sort of voxel or other volume based approach and generate the actual geometry to be rendered after the physics pass.

8Gb/s both ways at the same time... I'm assuming PCI-Express when I talk doing these things on the GPU. no way they are possible on AGP.

 

[Basic]

No 4 GB/s full duplex.