Installing NVIDIA GPU driver

This is the procedure for installing the GPU driver in the following environment.

  • GPU: NVIDIA A100
  • OS: Ubuntu20.04
  • GPU Driver: 550.x (LTS)

1. install NVIDIA GPU driver only

1.1. Download the package

Please visit the following download site

NVIDIA Official Driver Download | NVIDIA

Please enter A100 in the manual driver search and select the following items.
Data Center / Tesla | A-Series | NVIDIA A100

Please select Linux 64-bit Ubuntu 20.04 for the OS.

Please download the required version after agreeing to NVIDIA’s terms and conditions on your own.

1.2. Installation

You can download a file with a .deb extension, and copy it to the target server.

After deployment, execute the following commands on the server.

sudo dpkg -i <downloaded file>.deb
sudo cp /var/nvidia-driver-local-repo-ubuntu2004-550.xx.xx/nvidia-driver-local-xxxxxx-keyring.gpg /usr/share/keyrings/
## Copy and execute the part of the cp command displayed when executing the dpkg command

sudo apt update
sudo apt install nvidia-driver-550
sudo reboot

After rebooting, run the nvidia-smi command, and if you get a list of GPUs, the installation was successful.

To prevent unintended suspensions after installation that may cause the machine or GPU to become inaccessible, please also execute the following commands.

sudo systemctl enable multi-user.target --force
sudo systemctl set-default multi-user.target
sudo systemctl mask sleep.target suspend.target hibernate.target hybrid-sleep.target
sudo reboot

2. Install CUDA Toolkit

If you need CUDA, select the version you need at the following site and follow the instructions to install.

CUDA Toolkit Archive