Multiple GPUs and copying data.

[_OskaR]

I want to use two GPUs - 1st for rendering in OpenGL, 2nd for CUDA (8800M GTX SLI|). The bottleneck is copying data between GPUs - I must copy it from one card to main memory and later copy it from main memory to next card (to VBOs). Is possible to do it faster? Maybe is it possible in new cards like GTX4x0?

[cbuchner1]

I want to use two GPUs - 1st for rendering in OpenGL, 2nd for CUDA (8800M GTX SLI|). The bottleneck is copying data between GPUs - I must copy it from one card to main memory and later copy it from main memory to next card (to VBOs). Is possible to do it faster? Maybe is it possible in new cards like GTX4x0?

Try overlapping copying and computation using the streams API at the expense of extra memory getting used. This works best if computation roughly consumes the same time as your copy operation.

[cbuchner1]

I want to use two GPUs - 1st for rendering in OpenGL, 2nd for CUDA (8800M GTX SLI|). The bottleneck is copying data between GPUs - I must copy it from one card to main memory and later copy it from main memory to next card (to VBOs). Is possible to do it faster? Maybe is it possible in new cards like GTX4x0?

Try overlapping copying and computation using the streams API at the expense of extra memory getting used. This works best if computation roughly consumes the same time as your copy operation.