Torch.cuda.is_available() returns False with CUDA 7.0

Hi all,

I installed PyTorch successfully using the command:

conda install pytorch torchvision -c soumith

After installation, torch.cuda.is_available() returns False, this may due to the older CUDA version 7.0.

Is it possible to set the CUDA path manually for PyTorch to use GPU?

I have tried two ways to deal with the problem:

  1. Update CUDA (failed, no root privilege)
  2. Install from source (failed, no root to update cmake)

But unfortunately both of them are failed due to the above reasons.

What happens if you use conda install cuda80 pytorch torchvision -c soumith? PyTorch ships an updated copy of CUDA that you can use without root.

I used this command but torch.cuda.is_available() still returns False :frowning: .

upgrade your nvidia driver on your machine and it will be fixed.

Thanks for your reply, but I do not have the root privilege to update the nvidia driver.

we dont support drivers in the cuda 7.0 generation unfortunately