## Install cuDNN ### Download cuDNN .deb file You can download cuDNN file [here](https://developer.nvidia.com/rdp/cudnn-download). You will need an Nvidia account. Select the cuDNN version for the appropriate CUDA version, which is the version that appears when you run: ```shell nvcc --version ``` ### Install cuDNN ```shell sudo apt install ./<filename.deb> sudo cp /var/cudnn-<something>.gpg /usr/share/keyrings/ ``` My cuDNN version is 8, adapt the following to your version: ```shell sudo apt update sudo apt install libcudnn8 sudo apt install libcudnn8-dev sudo apt install libcudnn8-samples ``` ## Test CUDA on Pytorch ### Create a virtualenv and activate it ```shell sudo apt-get install python3-pip sudo pip3 install virtualenv virtualenv -p py3.10 venv source venv/bin/activate ``` ### Install pytorch ```shell pip3 install torch torchvision torchaudio ``` ### Open Python and execute a test ```python import torch print(torch.cuda.is_available()) # should be True t = torch.rand(10, 10).cuda() print(t.device) # should be CUDA ```