I am facing a very strange issue. I am working with cuda 9.0 version of PyTorch and yesterday it was working properly in my system ( ubuntu 14.04). Today when I tried to run my model, I noticed that it is using CPU. When I looked into it, I found that torch.cuda.is_available() is returning false even though the cuda driver is available. How is it possible? It was working properly till yesterday. How can it change syddenly?
Did you update and NVIDIA drivers etc.?
Could you try to restart your machine and check, if it’s working again?
I had similar issues after Ubuntu updated some drivers.
I was working on a remote machine. It will take some time to reboot and check. I will inform after checking. NVIDIA drivers are up to date