I have installed pytorch cuda by running this command:
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
My cuda version is 11.2 . I am using windows 10 .
Pytorch cuda 11.2 is not available right now . (pytorch.org)
So I have install 11.1 version .
(using nvidia-smi)
Cuda version
But it show false.
torch.cuda.is_available() >>> False
I have tried both 10.2 and 11.1 version.
As far as I know, I do not need to install cuda toolkit for pytorch
You should not install package to your base environment. Create a separate environment with necessary tools.
Example: create env called dlearn with Python v3.7 and torch packages
conda create -n dlearn python=3.7 pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
Activate and use your dlearn environment
conda activate dlearn
python -c "import torch;print(torch.cuda.is_available())"
# this should echo True if all is well
At the moment the supported cudatoolkit is 11.1 which works fine with 11.2 driver. They will update it sooner or later. You can build from GitHub PyTorch to get the latest if you want. The steps are more complex than above.