Installing Ubuntu with nvidia-docker support // A random tech blog

This is a quick guide to configure Ubuntu to run docker services on GPU, like deep learning programs for instance.

Before starting

You need a clean and up-to-date install of Ubuntu 18.04 on a computer with an NIVIDA GPU and internet connection.

If your computer doesn’t have an NVIDIA GPU then you can follow this guide to install docker and docker-compose. Just don’t follow the parts regarding NVIDIA drivers or docker-nvidia.

First update you computer:

sudo apt update
sudo apt upgrade

Basic install

NVIDIA Driver

It is highly recommended to start by installing NVIDIA drivers, as it the trickiest part. The commands bellow :

Add a ppa (Personal Package Archive). This one is maintained by X.org.
Update local package list.
List the possible drivers and the recommended ones.

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
ubuntu-drivers devices

At the date of this article, the last command says that the recommended version for my GPU is nvidia-driver-440, so we can run:

sudo apt-get install nvidia-driver-440

Then restart the computer to load the new driver. Run nvidia-smi to check if the driver is running.

Docker CE

To install Docker CE follow docker wiki or run

sudo apt-get install \
    apt-transport-https ca-certificates curl gnupg-agent software-properties-common
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu \
    $(lsb_release -cs) \
    stable"
sudo apt-get update
sudo apt-get install docker-ce docker-ce-cli containerd.io

and add the user to the docker group

sudo usermod -aG docker $USER

To update the user permissions, logout and log back in. To check if it works, try to run the command docker images

nvidia-docker

To install nvidia-docker follow nvidia-docker wiki or run

curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | \
    sudo apt-key add -
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
    sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update
sudo apt-get install nvidia-docker2
sudo pkill -SIGHUP dockerd

docker-compose

Follow the doc or run

sudo curl -L "https://github.com/docker/compose/releases/download/1.23.2/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
sudo chmod +x /usr/local/bin/docker-compose

Post install tips and tricks

From this point, you should have a working computer with docker, docker-compose and nvidia-docker. The following commands are extra tools to increase my productivity. Feel free to install them or not.

zsh installation and configuration

zsh is the command shell that I prefer. oh-my-zsh brings a set of useful tools to speed productivity. Run

sudo apt install zsh
sh -c "$(curl -fsSL https://raw.githubusercontent.com/robbyrussell/oh-my-zsh/master/tools/install.sh)"
cd .oh-my-zsh/themes
wget https://raw.githubusercontent.com/caiogondim/bullet-train-oh-my-zsh-theme/master/bullet-train.zsh-theme
sudo apt-get install fonts-powerline

then change the theme in .zshrc to setup bullet-train

Other packages

htop is a clearer top. openssh-server allow you to turn your computer into an ssh server, net-tools gives you tools like ifconfig, and nfs-common is useful when working with remote storage system like a NAS

apt install openssh-server net-tools nfs-common htop