I am running the command given on page Start Locally | PyTorch to install pytorch locally. I have a GPU (GeForce GTX 1070), the latest version of NVIDIA driver (455.32.00) and have previously installed CUDA (11.1). However, conda automatically fetches a cpu-only pytorch package to install:
pkgs/main::pytorch-1.3.1-cpu_py27h62f834f_0
and then I couldn’t use pytorch with my GPU at all. Does any one know how to solve this issue?
I guess conda somehow decides that I can’t run cuda-11.1, so it gave me a cpu-only version.
But I already had cuda-11.1 installed. Maybe it is because I am recently upgraded my OS to Ubuntu 20.04?)
I’m having a similar problem where I can only install the cpu version of pytorch and I’m trying to install the newest (1.7.1) pytorch. I’m running conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch
And it will only install the cpu version. I’ve updated conda and I’ve deleted previous versions of all of these packages. I’m using a Windows computer.