Skip to main content
Running Cog on Windows is possible thanks to WSL2. This guide shows you how to enable WSL2 and GPU passthrough on Windows 11.
Windows 10 is not officially supported, as you need to be on an insider build to use GPU passthrough.

​
Prerequisites

Before beginning installation, ensure you have:
  • Windows 11 (Windows 10 is not officially supported)
  • NVIDIA GPU from one of these series:
    • RTX 2000/3000 series
    • Kepler/Tesla/Volta/Ampere series
Other GPU configurations are not guaranteed to work.

​
Installation

1

Install the GPU driver

Download and install the latest Game Ready drivers for your NVIDIA GPU from NVIDIA’s website.Select your GPU model from the dropdown menus, click “Search”, and follow the installation wizard.
Restart your computer once the driver installation is complete.
2

Enable WSL2

Open Windows Terminal as an administrator:
  • Search for “Terminal” in the Start menu
  • Right click and select “Run as administrator”
Run the following PowerShell command:
dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart
If you see a permissions error, ensure you’re running as administrator and have administrator-level privileges.
3

Enable virtualization

In the same administrator terminal, run:
dism.exe /online /enable-feature /featurename:VirtualMachinePlatform /all /norestart
If this command fails, enable virtualization capabilities in your computer’s BIOS/UEFI.A successful output will print The operation completed successfully.
4

Reboot your computer

Restart Windows to enable WSL2 and virtualization features.
5

Update the Linux kernel

Download and run the WSL2 Linux kernel update package for x64 machines.When prompted for elevated permissions, click ‘yes’ to approve.Verify the kernel version by opening Windows Terminal as administrator and running:
wsl cat /proc/version
You should see output like:
Linux version 5.10.102.1-microsoft-standard-WSL2 (oe-user@oe-host) (x86_64-msft-linux-gcc (GCC) 9.3.0, GNU ld (GNU Binutils) 2.34.0.20200220)
The kernel version should be at least 5.10.43.3. If it’s not, open Settings → Windows Update → Advanced options and enable “Receive updates for other Microsoft products”, then check for updates.
6

Configure WSL 2 as default

Set WSL 2 as the default version in Windows Terminal with administrator privileges:
wsl --set-default-version 2
7

Install Ubuntu

Go to the Microsoft Store and download Ubuntu 18.04.Launch the “Ubuntu” app from your Start Menu and create a Linux user account when prompted.
8

Configure CUDA Toolkit

Open PowerShell or Windows Command Prompt in administrator mode and enter:
wsl.exe
This drops you into your Linux VM. Run the following commands to install the CUDA Toolkit for WSL-Ubuntu:
sudo apt-key del 7fa2af80 # if this fails, you may remove it
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/11.7.0/local_installers/cuda-repo-wsl-ubuntu-11-7-local_11.7.0-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-11-7-local_11.7.0-1_amd64.deb
sudo apt-get update
sudo apt-get -y install cuda-toolkit-11-7
Always use the WSL-Ubuntu CUDA Toolkit package, not generic Linux installation instructions. The CUDA version above may need to be adjusted based on your GPU’s capabilities.
9

Install Docker Desktop

Download and install Docker Desktop for Windows.After installation:
  1. Run Docker Desktop
  2. Go to Settings → General
  3. Ensure Use the WSL 2 based engine is checked
  4. Click Apply & Restart
Reboot your computer one more time after configuring Docker Desktop.
10

Install Cog

Open Windows Terminal and enter your WSL 2 VM:
wsl.exe
Download and install Cog inside the VM:
sudo curl -o /usr/local/bin/cog -L https://github.com/replicate/cog/releases/latest/download/cog_`uname -s`_`uname -m`
sudo chmod +x /usr/local/bin/cog
Verify the installation:
which cog        # should output /usr/local/bin/cog
cog --version    # should output the cog version number

​
Running Your First Model

Test your setup by running a model: 
cog predict 'r8.im/afiaka87/glid-3-xl' -i prompt="a fresh avocado floating in the water" -o prediction.json
While your prediction runs, use Task Manager to monitor GPU memory consumption. The model output will be saved as JSON for complex return types.

​
Processing the Output

If your model returns base64-encoded images, use jq to convert them: 
# Install jq
sudo apt install jq

# Convert base64 to PNG
jq -cs '.[0][0][0]' prediction.json | cut --delimiter "," --field 2 | base64 --ignore-garbage --decode > prediction.png

​
Viewing Results

WSL 2 allows you to access Windows binaries with the .exe extension: 
explorer.exe prediction.png

​
Troubleshooting

​
GPU not detected

If your GPU isn’t detected:
  1. Verify NVIDIA drivers are installed on Windows (not in WSL)
  2. Check that virtualization is enabled in BIOS
  3. Ensure WSL 2 is the default version: wsl --set-default-version 2

​
Docker errors

If Docker commands fail:
  1. Verify Docker Desktop is running
  2. Check that WSL 2 backend is enabled in Docker settings
  3. Restart Docker Desktop

​
CUDA errors

Never install CUDA using generic Linux instructions in WSL 2. Always use the WSL-Ubuntu CUDA Toolkit package.

​
References

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us