Installing CUDA (Ubuntu) - norlab-ulaval/Norlab_wiki GitHub Wiki

This post is here in case anyone ever has to deal with setting up CUDA, which can easily become a massive pain. Following these steps, you should be able to set up a working environment with CUDA in only a few minutes. It has been tested only with Ubuntu 22.04.

sudo apt update && sudo apt upgrade

sudo apt autoremove nvidia* --purge

ubuntu-drivers devices

You will install the NVIDIA driver whose version is tagged with recommended

sudo ubuntu-drivers autoinstall

My recommended version is 525, adapt to yours

sudo apt install nvidia-driver-525

reboot

after restart verify that the following command works

nvidia-smi

sudo apt update && sudo apt upgrade

sudo apt install nvidia-cuda-toolkit

nvcc --version

You can download cuDNN file here. You will need an Nvidia account. Select the cuDNN version for the appropriate CUDA version, which is the version that appears when you run:

nvcc --version

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:

sudo apt update
sudo apt install libcudnn8
sudo apt install libcudnn8-dev
sudo apt install libcudnn8-samples

sudo apt-get install python3-pip
sudo pip3 install virtualenv 
virtualenv -p py3.10 venv
source venv/bin/activate

pip3 install torch torchvision torchaudio

import torch
print(torch.cuda.is_available()) # should be True

t = torch.rand(10, 10).cuda()
print(t.device) # should be CUDA

Install the nvidia-container-runtime for compatibility with Docker containers. Follow this installation guide, don't forget to do the configuration steps.