I am curious about the fastest way to get data from the GPU back to system memory. I am currently using an FBO to render to a 32 bit float texture. I am wondering if I can use a PBO to readback the texture data into system memory. Does anyone know of an example floating around that shows how to do that? Is there a faster method?
Dual Operton/64 Bit/SuSE Linux 10/NVIDIA Quadro FX 4500
PBO readback offers the benefit of asynchronous behavior. If you stall after a ReadPixels via PBO to read the data, then you are no better off than performing a standard ReadPixels. The one case where PBO readback could be faster is if the driver allows the PBO to reside in VRAM and either the ReadPixels or MapBuffer call do not force the PBO back into system memory. In that case, the ReadPixels itself will be fast. However, depending upon the setup, the CPU read from VRAM could be slower/faster.