Geforce series dual DMA


I read about Geforce series dual DMA before, but still cannot understand what’s correct. (Some said only Quadro and Tesla support it.)

I’m using Geforce GTX 750 and OpenCL.

I want to run host to device copy and device to host copy synchronously for overlapping.

However, it didn’t work.

I also tested on Geforce GTX 1050 Ti and it didn’t work, neither.

I wonder whether Geforce GTX 750 and 1050 Ti is possible for dual DMA.

Please help!

Can’t help you with OpenCL, this is a CUDA forum. CUDA ships with an example program called deviceQuery. Run it, and look for the following line in the output:

If it says 2 copy engines, the GPU supports full-duplex DMA transfers across PCIe. The above output is from a Quadro K2200 (basically a glorified GTX 750 Ti), and it has only one copy engine available. Based on that it would be extremely unlikely that the GTX 750 supports more than 1 copy engine.