Setting Up WSL for GPU Compute
Published:
This post explains how to prepare and set up Windows Subsystem Linux (WSL) for GPU compute to run CUDA, OpenCL and oneAPI. In addition, it explains how to set up TornadoVM, but after installing the required SDKs and drivers, one can install any tool supported by WSL, such as PyTorch.
Configure WSL
This tutorial assumes WSL is already installed. If this is not the case, on Windows 11 is as easy as running the following command in PowerShell in admin mode:
wsl --install
This command installs, by default, the latest LTS version of Ubuntu. At the time of writing this post, the latest version is Ubuntu 24.04 LTS.
For more details about WSL configuration and installation, follow the official documentation: https://learn.microsoft.com/en-us/windows/wsl/install.
Set-up CUDA in WSL
If you have an NVIDIA GPU, chances are that you have already installed the NVIDIA driver for Windows PC. If not, the first step is to install (or update) the NVIDIA driver for Windows 11. The NVIDIA driver for Windows also includes support for WSL. Thus there is no need to install the NVIDIA driver within our WSL terminal. We only need the CUDA SDK.
## Update the system
sudo apt-get update
sudo apt-get dist-upgrade
Download and install CUDA for WSL
Use the NVIDIA documentation to obtain the latest instructions: https://docs.nvidia.com/cuda/wsl-user-guide/index.html
The following commands install CUDA Toolkit 12.6 for WSL:
sudo apt-key del 7fa2af80
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/12.6.3/local_installers/cuda-repo-wsl-ubuntu-12-6-local_12.6.3-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-12-6-local_12.6.3-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-12-6-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-6
Update ~/.bashrc file to include the following content:
export C_INCLUDE_PATH=/usr/local/cuda/include
export CPLUS_INCLUDE_PATH=/usr/local/cuda/include
export LD_LIBRARY_PATH=/usr/local/cuda/lib64
export PATH=/usr/local/cuda/bin/:$PATH
And open a new terminal. So now, CUDA is ready to run on our NVIDIA GPUs within WSL!
Compiling and Running CUDA Samples
$ git clone https://github.com/NVIDIA/cuda-samples.git
$ cd cuda-samples/Samples/1_Utilities/deviceQuery
$ make
$ ./deviceQuery
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 1 CUDA Capable device(s)
Device 0: "NVIDIA GeForce RTX 3070"
CUDA Driver Version / Runtime Version 12.7 / 12.6
CUDA Capability Major/Minor version number: 8.6
Total amount of global memory: 8192 MBytes (8589410304 bytes)
(046) Multiprocessors, (128) CUDA Cores/MP: 5888 CUDA Cores
GPU Max Clock rate: 1725 MHz (1.73 GHz)
Memory Clock rate: 7001 Mhz
Memory Bus Width: 256-bit
L2 Cache Size: 4194304 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total shared memory per multiprocessor: 102400 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 1536
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 1 copy engine(s)
Run time limit on kernels: Yes
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Device supports Managed Memory: Yes
Device supports Compute Preemption: Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: No
Device PCI Domain ID / Bus ID / location ID: 0 / 1 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 12.7, CUDA Runtime Version = 12.6, NumDevs = 1
Result = PASS
Note that, after this point, we can install PyTorch and unsloth and create fine tune LLM models in our system (if you have a powerful enough GPU, and plenty of VRAM!).
Intel Compute Runtime: Enabling OpenCL and Level Zero
Additionally, you can install the Intel Compute Runtime to get access to OpenCL and Level Zero for iGPUs, ARC GPUs and CPUs.
Go to https://github.com/intel/compute-runtime/releases/ and download the latest release. In my case, at the time of writing this post, the latest version is 24.48.31907.7.
mkdir -p ~/bin/neo
cd ~/bin/neo
wget https://github.com/intel/intel-graphics-compiler/releases/download/v2.2.3/intel-igc-core-2_2.2.3+18220_amd64.deb
wget https://github.com/intel/intel-graphics-compiler/releases/download/v2.2.3/intel-igc-opencl-2_2.2.3+18220_amd64.deb
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/intel-level-zero-gpu-dbgsym_1.6.31907.7_amd64.ddeb
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/intel-level-zero-gpu_1.6.31907.7_amd64.deb
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/intel-opencl-icd-dbgsym_24.48.31907.7_amd64.ddeb
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/intel-opencl-icd_24.48.31907.7_amd64.deb
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/libigdgmm12_22.5.4_amd64.deb
Verify CheckSums:
wget https://github.com/intel/compute-runtime/releases/download/24.48.31907.7/ww48.sum
sha256sum -c ww48.sum
Install Intel Compute Runtime packages:
sudo dpkg -i *.deb
Update link for OpenCL
To run OpenCL, we need to do one more step for Ubuntu-based distributions:
sudo ln -s /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 /usr/lib/x86_64-linux-gnu/libOpenCL.so
Ready! Let’s set up TornadoVM
Configuring TornadoVM for WSL
Install Python venv module
sudo apt install python3-venv
And set up a new Python environment as follows:
python3 -m venv ~/bin/venv
source ~/bin/venv/bin/activate
Finally, let’s configure TornadoVM:
cd ~/
git clone git clone https://github.com/beehive-lab/TornadoVM.git tornado
cd tornado
## Install OpenCL only
./bin/tornadovm-installer --jdk jdk21 --backend=opencl
## Install OpenCL and PTX
./bin/tornadovm-installer --jdk jdk21 --backend=opencl,ptx
## Install All backends:
./bin/tornadovm-installer --jdk jdk21 --backend=opencl,ptx,spirv
source setvars.sh
Run tests
make tests
Print devices:
tornado --devices
Conclusions
By following the outlined steps, readers can successfully configure WSL, install CUDA for NVIDIA GPUs, and set up the Intel Compute Runtime for OpenCL and Level Zero support on compatible hardware. Furthermore, the guide has demonstrated how to install TornadoVM. Importantly, the guide has emphasized that once the necessary SDKs and drivers are in place, users can readily install other GPU-accelerated tools like PyTorch, opening up a vast landscape of possibilities for AI, deep learning, and scientific computing within the convenient and flexible WSL framework. Happy coding!