[Machine Learning Basics] nvidia-driver, CUDA Toolkit

I recently received a server which GPU were available, and found out how to go through GPU settings.

 

Content
1. Connecting to (SSH) server with VSCode
2. Installing GPU driver (nvidia-driver)
3. Installing CUDA

 

1. Connecting to SSH server with VSCode (with password, not the SECRET_KEY)

 VSCode seems to be the predominant code editor. VSCode has numerous extensions for such things like Git, Remote-SSH, Go-Live, and so on. You can download VSCode at https://code.visualstudio.com.

 

First of all, install the "Remote - SSH" extension in Visual Studio Code. You can do this by navigating to the Extensions view by clicking on the square icon on the left sidebar or by pressing Ctrl+Shift+X. Search for "Remote - SSH" and install the extension.

 

Second, open the command palette by pressing Cmd+Shift+P. This will open a text box at the top of the window. In the command palette, type "Remote-SSH: Connect to Host" and select this option from the dropdown that appears. A list of options will show up.

 

 

Choose "Add New SSH Host..." if you're connecting to a new SSH host for the first time. This will open a new text box in which you can enter the SSH connection information. Go to ".../.ssh/config" and add the following information:

Host XXX.XXX.XXX.XXX
    HostName XXX.XXX.XXX.XXX
    User /*Probably 'root'*/

 

Finally, you can enter the server by clicking your server name, and then entering the password.

 

2. nvidia-driver

You should start with:

$ sudo apt-get update
$ sudo apt-get upgrade
$ sudo apt-get install python3-pip

 

After you successfully installed pip3, you manually install gpu driver by typing:

$ sudo apt-get install ubuntu-drivers-common
$ ubuntu-drivers devices
$ sudo ubuntu-drivers autoinstall

 

Check if you successfully got the driver:

$ sudo reboot
$ nvidia-smi

This image is what you should get.

 

3. CUDA

Go to https://developer.nvidia.com/cuda-toolkit-archive in order to install CUDA Toolkit. I downloaded Toolkit 11.8:

$wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
$ sudo sh cuda_11.8.0_520.61.05_linux.run

(Note that in most of the case, Operating System: linux, Architecture: x86_64, Distribution: Ubuntu, Version: 20.04, Intaller Type: runfile (local))

 

Then, go to .bashrc and add the following rows to do the CUDA_PATH settings:

export PATH=/usr/local/cuda-11.8/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

 

Lastly, check if the CUDA has been successfully installed:

$ source ~/.bashrc
$ nvcc -V

CUDA installation complete

 

You are done. Try:

$ python3
>> import torch
>> t = torch.rand(10).to("cuda")
>> print(t)
tensor([0.3170, 0.9243, 0.4765, 0.5617, 0.3952, 0.8225, 0.4612, 0.7638, 0.3209, 0.5149], device='cuda:0')

 

Reference

https://velog.io/@hun_dev/개발-환경-GCP에서-GPU를-이용해-개발해보자-03

[개발 환경] GCP에서 GPU를 이용해 개발해보자! - 03

https://code.visualstudio.com

Visual Studio Code - Code Editing. Redefined

https://developer.nvidia.com/cuda-toolkit-archive

CUDA Toolkit Archive