I put all my tensor to(device) device = torch.device(“cuda:0”)
I add .to(device) for every operation I make with tensors
torch.cuda.get_device_name(0) return Tesla P100-PCIE-16GB
But in !nvidia-smi : no runnung processes found
As I see gpu doesn’t work properly.
nvidia-smi might now show all processes depending on the “environment”. E.g. I see this behavior often in docker containers and don’t know where this issue comes from.
ps should show the processes however.
The GPU seems to be filled with ~1GB of data and its utilization is at 15%, so how did you determine that the GPU is not working properly?