step0: Preface
This article can help you take the first step of CUDA in the simplest way, from now on, you will be as deep as the sea, and from now on, your hair will be a passerby.
Prerequisite: You need to be on Windows 11:
- There are Nvidia graphics cards and drivers in the computer. Since WSL Cuda does not support the Maxwell GPU architecture, it requires graphics cards of 10 series or above.
- Install wsl2, configure ubuntu mirroring and make sure it can be entered.
Used here:
- The operating system image is: ubuntu 22.04.3 LTS
- Install cuda version: 12.8.1
Note that configuring cuda on WSL is different from configuring programs on ordinary Linux, be sure to pay attention!
step1: Preparation
Due to the ubuntu 22. version used here, we need to do some early configuration
You can query the corresponding relationship between cuda 12.8.1 and operating system, kernel, gcc, and glibc versions from this link:/cuda/cuda-installation-guide-linux/
Theoretically, other versions may not be supported, but for the need to cause less trouble for yourself, the version will be updated according to the official tutorial.
Confirm the operating system and toolchain version:
lsb_release -a # Confirm the operating system version
gcc -v # Confirm gcc version
cmake --version # Confirm the cmake version
ldd --version # Confirm the glibc versionInstall gcc12.3.0:
sudo apt install gcc-12 g++-12 -y
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-12 100
gcc --version # Confirm the gcc versionInstall cmake (as of this article, cmake latest version is 3.22.1):
sudo apt install cmake
cmake --version
At this point, the advance tool chain has been configured.
step2: Install CUDA toolkit
CUDA toolkit is an essential toolchain in the development process. We need to install and configure this stage.
Refer to the official tutorial:
- /cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local
- /cuda/wsl-user-guide/#getting-started-with-cuda-on-wsl
The existing CUDA for running and the CUDA toolkit for compilation and development are completely different things, don't mess it up.
Execute strictly in the following code, and do not change the order:
sudo apt-key del 7fa2af80
wget /compute/cuda/repos/wsl-ubuntu/x86_64/
sudo mv /etc/apt//cuda-repository-pin-600
wget /compute/cuda/12.8.1/local_installers/cuda-repo-wsl-ubuntu-12-8-local_12.8.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-12-8-local_12.8.1-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-12-8-local/cuda-*- /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-8
sudo apt-get install cuda
step3: Try to compile and run the sample project
nvidia officially provides a sample project for beginners to quickly get started with CUDA:
Execute the following code, clone it from github and build it
git clone /NVIDIA/
cd cuda-samples/
mkdir build && cd build
export PATH=/usr/local/cuda-12.8/bin/:$PATH
cmake ..
You can also enter a specific project path to compile and run. Here, take vectorAdd as an example:
cd Samples/0_Introduction/vectorAdd
cmake .
make
./vectorAdd
If everything goes well, you will see the following output:
~/cuda-samples/Samples/0_Introduction/vectorAdd$ ./vectorAdd
[Vector addition of 50000 elements]
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Copy output data from the CUDA device to the host memory
Test PASSED
Done
At this point, all configurations have been completed, and you should be able to freely compile and run CUDA programs in your own wsl.
In the next issue, we will officially enter the world of CUDA. Starting from the CUDA kernel, we will squeeze every inch of the performance of the GPU. In this process, you will deeply feel the power of computing power.
Thank you for reading the above. If you think this article is helpful, follow the Snow Leopard Meow, follow the Snow Leopard Thank you Meow.