wsl

Setting Up CUDA in WSL

Running CUDA on native Linux is possible, but NVIDIA driver support can be inconsistent across distributions. Using CUDA in Windows Subsystem for Linux (WSL) offers a reliable alternative, providing direct driver support and a smoother experience for AI development.

Prerequisites

  • Windows 10/11 with WSL enabled
  • An NVIDIA GPU with driver support for CUDA

1. Install or Import a WSL Distro

If you don’t have WSL installed, run the following command in PowerShell as administrator:

1
wsl --install

If you want to use a fresh Ubuntu WSL image, download it from Ubuntu WSL Images and import it to a new folder:

1
wsl --import <NewDistroName> <InstallLocation> <PathToTarFile> [--version <Version>]

After importing, launch your new distro:

1
wsl -d <DistroName>

By default, you’ll log in as root. You can create a regular user later if needed.

2. Update and Prepare the System

No need to install the NVIDIA driver inside WSL; just install the CUDA toolkit. First, update your package sources:

1
2
sudo apt-get update
sudo apt-get upgrade

Remove any old NVIDIA GPG keys:

1
sudo apt-key del 7fa2af80

3. Install CUDA Toolkit for WSL-Ubuntu

Download and install the CUDA toolkit using the official NVIDIA repository. Here’s how to use the deb(network) installer:

1
2
3
4
wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt-get -y install cuda-toolkit-13-0

To make the CUDA toolkit available every time you open a new terminal, add the following line to your ~/.bashrc file:

1
echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc

After editing .bashrc, reload it with:

1
source ~/.bashrc

Now, you can use CUDA commands without manually updating your PATH each session.

4. Verify Installation

Check if the CUDA compiler is installed:

1
nvcc --version

To verify your CUDA and driver versions:

1
nvidia-smi

Conclusion

With these steps, you should have a working CUDA environment in WSL, ready for AI and GPU-accelerated development. If you encounter issues, refer to the NVIDIA CUDA on WSL documentation for troubleshooting tips.