Installing CUDA + NVHPC on WSL #

This page describes the steps to setup CUDA and NVHPC within the WSL2 container (Windows 10) - avoiding the need for dual-boot or a separate Linux PC. Note that WSL2 must not have been installed when beginning these steps.

1. Install the latest Windows CUDA graphics driver
2. Install WSL2
   • open PowerShell as administrator
   • Make sure to update WSL kernel to latest version wsl –update
     • if accidentally rolled back, follow the instructions here
     • then wsl –update again
   • check which Linux flavours are available with wsl –list –online
   • install the desired flavour by wsl –install -d
   • start WSL with wsl, or opening the WSL application from the Windows search bar
3. sudo apt update && sudo apt upgrade -y
4. Close and restart WSL
5. sudo apt update && sudo apt upgrade -y && sudo apt autoremove -y
6. Install CUDA for WSL
   • check which CUDA version is compatible with the desired version NVHPC kit here
   • select the correct CUDA version here
   • Select the right setup: Linux -> x86_64 -> WSL-Ubuntu -> 2.0 -> dev (local)
   • before beginning install, delete old GPG key sudo apt-key del 7fa2af80
   • Perform the install with code below:

wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-wsl-ubuntu.pin
sudo mv cuda-wsl-ubuntu.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-11-7-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda

7. Install NVHPC
   • Download the desired NVHPC kit
   • tar xzvf nvhpc*.tar.gz
   • sudo nvhpc*/install

Uninstalling NVHPC and CUDA #

sudo rm /opt/nvidia/hpc_sdk
sudo apt-get purge -y cuda && sudo apt-get autoremove -y
sudo rm /usr/share/keyrings/cuda-*-keyring.gpg
sudo dpdkg -P cuda-repo-wsl-ubuntu-*-*-local_*amd64.deb

You can then perform the installation steps again for the desired NVHPC-CUDA combination