Quote:Original post by NightCabbage
I'll re-phrase...

How much of a performance hit is it to get a texture form the video card?

It depends...

Moving the actual texture from GPU memory to system memory will depend on how large the texture is, and how much bandwidth is available on the AGP/PCI-e bus. However the problem you really need to watch out for is locking a resource that the GPU needs for rendering can cause a stall. Since a GPU works in parallel with the GPU and can be up to 2 frames ahead of where the CPU is, locking a resource that GPU needs for rendering to means that that it will have to wait until that resource is relinquished.

The common solution to this is to double-buffer your resources, so you can be rendering to one texture while reading the texture from last frame at the same time. It introduces a single frame of lag, but that's usually not a big deal.