Does the new GTX 200 GPU have DMA capability?

Does anybody know if the new Geforce GTX 200 core have DMA master capability?

I remember a while back in the forums there was talk of it. I am interested in the capability of the GPU being able to independently fetch CPU SDRAM memory.



All CUDA-capable GPU’s have this capability as far as I know. At least the compute capability 1.1 devices can do cudaMemCpyAsync while running a kernel.
And from what I have read here, it is always the GPU who initiates DMA transfers.

Doesn’t the cudaMemCpyAsync call have to be done from a CPU thread? What I would like is a cuda thread operating and by itself launch a DMA transfer from CPU SDRAM to GPU SDRAM.


That is for sure not possible. In the case you mention, you call the memcopy from CPU, but it is actually the GPU that issues a DMA to copy the data from System RAM to GPU Mem.