RTX3080ti RuntimeError: Torch is not able to use GPU

Mark_Lazarus · February 11, 2024, 2:35pm

i reinstall drivers, install cuda 11.8 devkit but nothing, torch cant recognize RTX
What will be do ? Thanks

ptrblck · February 11, 2024, 4:23pm

What does python -c "import torch; print(torch.randn(1).cuda())" return?

Mark_Lazarus · February 11, 2024, 4:58pm

nothing ( i had phyton and git on board, and made cmd in SD folder

ptrblck · February 11, 2024, 5:17pm

In that case you might not have installed a valid Python environment as a random tensor should be returned:

python -c "import torch; print(torch.randn(1).cuda())"
tensor([0.2099], device='cuda:0')

or an error message in case your installation does not ship with CUDA support.

Mark_Lazarus · February 11, 2024, 5:20pm

i will check and report ,tanks
update: reinstall python

ptrblck · February 11, 2024, 5:54pm

You didn’t install PyTorch in this environment, so stick to these install instructions and check again if you can import torch and if you can create a simple CUDATensor.