If no other users are sharing a K520, can you grab both GPUs for doing CUDA compute jobs? Our application uses the GPU for both display and compute. When we run on AWS K520 instances, CUDA can only see one of the GPUs on the K520. Our application makes use of multi-GPU if available like on the multi-GPU AWS p2 instances’ K80s.
It is not critical, but it would be nice if some of our operations were 2x faster on K520 when no one is using the other GPU.
Thanks for any replies.