It’s my first time trying to use PyTorch with a GPU and I’m struggling.
I’m using an Azure Virtual Machine (NCasT4_v3-series) which requires manual configuration as detailed here: https://learn.microsoft.com/en-us/azure/virtual-machines/windows/n-series-driver-setup. This was apparently successful.
Then I had to download and install CUDA 12.1 from here: https://developer.nvidia.com/cuda-downloads. This also seemed to work.
Finally I installed miniconda and installed an existing environment (I’m transferring an existing project to this VM). I then duly ran the code I found on https://pytorch.org/get-started/locally/ i.e. conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia - which also ran successfully.
But after all that, torch.cuda.is_available() returns False.
I’m at a loss for what to do next or any possible troubleshooting steps. Any help would be much appreciated.
I've managed to get this working, although I still don't have any insight into what the problem was. On logging back into the machine I found a problem with the NVIDIA device driver.
I uninstalled everything I had installed manually, and instead added the NVIDIA drivers via the Azure Portal as detailed here.