Wsl2 using GPU for deep learning

Environment building

First, check your system version , When needed win10 Preview or win11 To succeed ..
Reference link
https://docs.nvidia.com/cuda/wsl-user-guide/index.html#installing-wip

1.wsl2 install

wsl2 The installation is actually very simple , Generally speaking, it can be divided into two steps

• Install activation wsl2 function
  In fact, you can directly refer to Official reference documents . But in fact, I didn't use the command directly to succeed here , Here we refer to another Official documents , Successfully installed wsl2.
• install linux Distribution version
  linux The installation of the distribution is also mentioned in the official tutorial above , In fact, you can download and install the corresponding version directly from the Microsoft application store .（ What I have installed here is ubuntu18.04.）

2.GPU Relevant environment installation

If you are good at English, you can see The official tutorial , It's very detailed .

2.1cuda Drive installation

This part is in windows Installed in the system , from link Download for wsl Developed cuda drive . Install... After downloading , The first option I choose by default （ Install the graphics driver and experience Program ）, After installation, you need to restart the computer to take effect . What needs to be reminded here is that the official prompt computer only needs this one cuda drive , It can be either windows System services can also be wsl2 Inside ubuntu system service .

After installation, it can be installed in windows Runs from the command line nvidia-smi To get GPU The current running state of , Otherwise, the installation may fail .

2.2.cuda toolkit install

Here we will directly follow the official order in wsl2 Of ubuntu Execution in terminal .

Here, I personally suggest that before executing the command ubuntu Bring it with you apt-get updated （ Use update and upgrade command ）.
And here in the third line of command dep The bag is very large , Yes 2 More than a G, If possible, you can download it in a scientific way .

$ 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/11.4.0/local_installers/cuda-repo-wsl-ubuntu-11-4-local_11.4.0-1_amd64.deb
$ sudo dpkg -i cuda-repo-wsl-ubuntu-11-4-local_11.4.0-1_amd64.deb
$ sudo apt-key add /var/cuda-repo-wsl-ubuntu-11-4-local/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get -y install cuda

2.3. Example verification

$ cd /usr/local/cuda-11.4/samples/4_Finance/BlackScholes
$ make BlackScholes
$ ./BlackScholes

Initializing data...
...allocating CPU memory for options.
...allocating GPU memory for options.
...generating input data in CPU mem.
...copying input data to GPU mem.
Data init done.

Executing Black-Scholes GPU kernel (131072 iterations)...
Options count             : 8000000
BlackScholesGPU() time    : 0.207633 msec
Effective memory bandwidth: 385.295561 GB/s
Gigaoptions per second    : 38.529556

Error message appears , It can be seen that the permission is insufficient , Use sudo Can be generated .

A new error message appears , The reason is that before cuda Driver installation failed .

terms of settlement ： open windows Update advanced options in settings , Check the first one , The system will automatically wsl Perform driver related updates .

Finally, it was found that it seemed that the operating system kernel caused , There are two solutions , One is to join win10 Preview experience plans to upgrade the operating system kernel , The second is to upgrade win11. My version is 19043.1348, It is said that it is necessary to 20145 Or higher .
I decided to upgrade win11 了 .,,, upgrade win11 After that, everything was normal , Even in wsl Can be used directly in the console of nvidia-smi Command to view the graphics card , I'm convinced .