0

There are loads of questions on here about this & I have read those that I can find. Still pytorch ignores my GPU.

I have a Win10 64x machine with a NVIDIA Maxwell GeForce GTX 960 with a compute ability of 5.2 so well within CUDA 11.x territory. I updated my drivers prior to pytorch installation.

I went over to the Pytorch - get started page selected the appropriate combination & copied the conda command for that set up. In my case this was,

conda install pytorch torchvision torchaudio cudatoolkit=11.6 -c pytorch -c conda-forge

I go to my conda command terminal & open up my environment that I use for things like OpenCV and enter the command. Everything installs, no errors.

I then check the installation by opening Python & entering the following,

>>>import torch
>>>print(torch.__version__)

1.12.1

>>>print(torch.cuda.is_available())

False

So what am I missing here?

DrBwts
  • 3,470
  • 6
  • 38
  • 62
  • What version of Python do you have? Apparently `pytorch` only supports Python v3.7-3.9 on Windows. Perhaps your Python version is outside that range? – tdpu Oct 01 '22 at 20:18
  • @tdpu python 3.8.11 – DrBwts Oct 01 '22 at 22:18
  • That's strange. Looking at the [CUDA compatibility chart](https://en.wikipedia.org/wiki/CUDA#GPUs_supported), it looks like CUDA Toolkit v11.6 covers CUDA compute support for Maxwell versions 3.5 - 8.7. I'm wondering if somehow the Nvidia/CUDA drivers are busted - you could try reinstalling or rolling back. See also: https://docs.nvidia.com/cuda/archive/11.6.0/maxwell-compatibility-guide/index.html. There's a lot of suggestions (with better explanation) in this [detailed answer](https://stackoverflow.com/a/61034368/13752965), you might find some help there. – tdpu Oct 03 '22 at 03:57

0 Answers0