I recently purchased a K80, I was planning on using it with PyTorch and Deep Learning. I origonally purchased it for how much memory it had 24GB. However, then I realized it has 2 gpus on a single board and the memory was split in half amongst them.
My question is: to train a neural network with the K80, would I have to load a copy of the Network into each of the 2 gpus on a single board for the K80? Or will just one copy do, and the other gpu on the chip can access it?
It also looks like I need to downgrade to pytorch 1.3. Is that correct?
Thank you for your help.