WSL_2_WITH_CUDA.md

WSL 2 with CUDA

Set up Windows Subsystem for Linux (version 2) with CUDA support + TensorFlow.

Before you start

• You must be on Windows 10 21H2 or above.

  Note: On my system 21H2 could not be updated to. I ended up with a bricked install as Windows tried to install Windows 11 on my unsupported CPU. I recommend just installing Windows 10 21H2 directly from an ISO.

• You don't have to be on OS build 20140 or above as some sources state. My build: 19044.1469
• Install the latest "game ready driver" for your card.

  Note: You don't need a special driver despite the Nvidia documentation, I got mine here: https://www.nvidia.com/Download/index.aspx

• The apt-cache search <PACKAGE> command is your friend. Use it to see what CUDA versions are available.

Steps to install

1. Open CMD as administrator and run:

   wsl --install -d ubuntu
   

2. In the Ubuntu WSL, update the system:

   sudo apt update
   sudo apt upgrade
   

3. Install Git:

   sudo apt install git
   

4. Figure out what version of CUDA & cuDNN you need, I recommend looking at this page: https://www.tensorflow.org/install/source#gpu

5. Find the desired version of CUDA on this page: https://developer.nvidia.com/cuda-toolkit-archive

6. Enter the following information and follow the instructions, but do not run the apt-get install cuda command!

   

7. Install the specific version of CUDA, for example:

   sudo apt install cuda-toolkit-11-8 cuda-11-8
   

8. Install cuDNN for CUDA 11.x and Ubuntu 22.04:

   1. Download the TAR file from: https://developer.nvidia.com/rdp/cudnn-archive
   2. Decompress the file, for example:

      tar xf cudnn-linux-x86_64-8.9.5.30_cuda11-archive.tar.xz
      

   3. Copy the library files into the right place:

      cd cudnn-linux-x86_64-8.9.5.30_cuda11-archive/
      
      sudo cp include/cudnn.h /usr/local/cuda-11/include/
      sudo cp lib/libcudnn* /usr/local/cuda-11/lib64/
      sudo chmod a+r /usr/local/cuda-11/include/cudnn.h /usr/local/cuda-11/lib64/libcudnn*
      

9. Edit ~/.bashrc and add the following:

   # CUDA environment variables
   export PATH=/usr/local/cuda-11/bin${PATH:+:${PATH}}
   export LD_LIBRARY_PATH=/usr/local/cuda-11/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
   

10. Restart your terminal or run source ~/.bashrc

11. Verify CUDA installation (optional)

    git clone https://github.com/NVIDIA/cuda-samples.git
    cd cuda-samples/Samples/5_Domain_Specific/BlackScholes/
    make
    ./BlackScholes
    

12. Install Python and Tensorflow (optional)

    sudo apt install python3-dev python3-pip
    pip3 install tensorflow
    

13. Try not to think about https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguish (optional)