NVIDIA CUDA installation guide on Windows

Created: February 20, 2024

This guide provides detailed steps to install NVIDIA CUDA on a Windows environment using Windows Subsystem for Linux 2 (WSL2) and Miniconda.

Requirements:

• NVIDIA Graphics Card
• Ensure your system is updated to the latest version through Windows Update

Steps:

1. Install WSL2
2. Install Miniconda in WSL
3. Install Graphics Card Drivers
4. Install RAPIDS
5. Install VS Code Extensions in WSL
6. Verify CUDA in VS Code
7. References

1. Install WSL2

To perform the correct installation of WSL, follow these detailed steps. Alternatively, you can refer to the Manual installation steps for older versions of WSL in the official Microsoft documentation, which serves as the basis for this point in the guide.

Enable WSL

Open PowerShell as an administrator and run the following command to enable the WSL features:

dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart

Enable Virtual Machine feature

Open PowerShell as an administrator and run the following command to enable virtualization features:

dism.exe /online /enable-feature /featurename:VirtualMachinePlatform /all /norestart

Proceed to restart the computer.

Download the Linux kernel update package

In the official Microsoft documentation Manual installation steps for older versions of WSL navigate to the section to download the latest version of the Linux kernel and subsequently execute it:

Set WSL 2 as default version

Open PowerShell and execute the following command to set WSL2 as default version:

wsl --set-default-version 2

If you have already installed WSL with an earlier version (WSL1), you must update it to version 2

Install Linux distribution

Open Microsoft Store and install the Ubuntu Linux distribution, which generally has the most updated version.

Open the Ubuntu terminal, create a user account and set a password.

Update the Linux distribution package

In the Ubuntu terminal, execute the following command to update the WSL packages:

sudo apt update && upgrade

or

sudo apt update && sudo apt upgrade

Enter the password established earlier.

Install Neofetch (Optional)

In the Ubuntu terminal, install Neofetch to view the distribution features with the following command:

sudo apt install neofetch

To view the features, type neofetch in the terminal.

2. Install Miniconda in WSL

Download Miniconda

Visit the Latest Miniconda installer links by Python version section on the official Conda website.

To determine the Python version used by your OS, open the Ubuntu terminal and excute the following command:

python3 --version

In this guide, we assume an OS with Python 3.10 and an X64 architecture. Download the corresponding file for your system; in our example, we downloaded the Miniconda3 Linux 64-bit file.

Installation

Navigate to the folder where you downloaded the file using the Ubuntu terminal and execute the following command, considering the file name:

chmod +x ./Miniconda3-py310_23.11.0-2-Linux-x86_64.sh

Next, run the file with the following command:

./Miniconda3-py310_23.11.0-2-Linux-x86_64.sh

Press Enter to begin. You can read the end-user license agreement or skip it by pressing q. Then, accept the license by typing yes.

Installation directory

Set the default directory using the following command:

{HOME}/.miniconda3

After installing the base environment, press Enter to complete the installation.

Return to the base directory with the command:

cd

Then run:

nano -wc .bashrc

Scroll to the end by pressing Ctrl + down, press Enter, and type:

export PATH=${PATH}:${HOME}/.miniconda3/bin

Exit the editor by pressing Ctrl + x and type yes.

Verify Miniconda installation

Run the following commands:

source .bashrc

Then, execute:

conda update conda

If no errors are returned, the installation has been successful. At this point, Miniconda should check for updates; if there are any, type y to proceed with the update.

Conda

Run the following command:

conda config --set auto_activate_base false

Exit the terminal with the command:

exit

Open a new Ubuntu terminal and execute the command:

conda init

Miniconda has now been installed and configured.

3. Install Graphics Card Drivers

Visit the official NVIDIA website in the NVIDIA Driver Downloads and fill in the fields with the corresponding grapichs card and OS information. In this guide, we used an NVIDIA GeForce GTX 1650 Ti graphics card.

Download the driver and run the file to install it on the Windows OS.

Note: the user needs to have a compatible NVIDIA GPU to perform these steps.

Install NVIDIA Drivers in WSL

Open the Ubuntu terminal and run the following command to update the packages:

sudo apt update && upgrade

Then, install the drivers according to the version. In this example, we installed version 545:

sudo apt install nvidia-driver-545

To verify that the installation was successful, run the following command:

nvidia-smi

You should see the information corresponding to the installed drivers.

4. Install RAPIDS

Visit the RAPIDS Installation Guide in the Install RAPIDS section and copy the command. In this example, we've selected specific packages like cuDF and cuML, and the virtual environment name was set as cuda

Use the following command as an example:

conda create --solver=libmamba -n cuda -c rapidsai -c conda-forge -c nvidia  \
    cudf=24.02 cuml=24.02 python=3.10 cuda-version=12.0

Virtual Environment

Activate the virtual environment cuda (or whatever you name it) and run the following command to verify that CUDA libraries are installed:

conda list

Additional Libraries

Install Pytorch:

conda install -c pytorch pytorch

Then, install ipykernel or any additional libraries you may need:

conda install -c ipykernel

5. Install VS Code Extensions in WSL

In the Ubuntu terminal, open VS Code with the following command:

code .

It'll be installed within WSL, and once the IDE is open, make sure to have the following extensions installed:

6. Verify CUDA in VS Code

Finally, ensure that you can use CUDA. Open a Jupyter Notebook in VS Code and execute the following code:

import torch

torch.cuda.is_available()

If the result is True, it means that CUDA is available and ready to be used. Congratulations!

7. References

1. Craigloewen-Msft. Manual installation steps for older versions of WSL. Microsoft Learn. https://learn.microsoft.com/en-us/windows/wsl/install-manual. Published 20 de noviembre de 2023.
2. TrujilloSoft. WSL 2: Instalar Linux en Windows 11. YouTube. marzo 2023. https://www.youtube.com/watch?v=6U2caEujxZ4.
3. The Strawberry Data Scientist. Miniconda installation guide (Ubuntu 22.04 LTS). YouTube. agosto 2023. https://www.youtube.com/watch?v=KkEoaPgeuCc.
4. Download the latest official NVIDIA drivers. https://www.nvidia.es/Download/index.aspx?lang=en.
5. Installation Guide - RAPIDS Docs. RAPIDS Docs. https://docs.rapids.ai/install?_gl=1*1l55d5n*_ga*MjAzMzQ4OTI2Ni4xNzA4MjA5MzI0*_ga_RKXFW6CM42*MTcwODIwOTMyNC4xLjEuMTcwODIxMDEzMy42MC4wLjA.#selector.
6. RomanAcademy. How to Install NVIDIA Drivers on Ubuntu 22.04 LTS - RomanAcademy - Medium. Medium. https://roman-academy.medium.com/how-to-install-nvidia-drivers-on-ubuntu-22-04-lts-6186e2f66749. Published noviembre 7, 2022.
7. Python Simplified. CUDA Simply Explained - GPU vs CPU Parallel Computing for Beginners. YouTube. diciembre 2021. https://www.youtube.com/watch?v=r9IqwpMR9TE.