Quick Install of Run:AI on a Single Node¶

Below are instructions on how to install the Run:AI cluster on a single node. This process is good for learning Run:AI or using a Run:AI cluster on a single node. Multiple nodes are not supported with this installation. To install a cluster with multiple nodes or for running a formal pilot with Run:AI, use Cluster Installation.

The installation process below is comprised of a single script and includes the installation of a built-in Kubernetes using minikube.

Prerequisites¶

The installation below assumes:

• A single node, with at least one GPU.
• Running Ubuntu 20.04.
• sudo access to the node (you may be prompted for sudo password during the installation).
• An email and a password provided by Run:AI customer support.
• Outbound internet connectivity

If NVIDIA drivers are not installed, the script will install the latest NVIDIA Drivers. However, the script is wired with the Ubuntu 20.04 script for NVIDIA drivers and will fail on other Ubuntu versions.

Installation steps¶

Log in to app.run.ai for the first time to change your password. Do not create a new Cluster, or the script will fail.

Get the script:

wget https://raw.githubusercontent.com/run-ai/docs/master/install/single-node-install.sh && chmod +x single-node-install.sh

Run the script:

./single-node-install.sh <email> '<password>'

Node Shutdown and Restart¶

To shut down your node, you must first perform an orderly shutdown of Kubernetes by running:

minikube stop

When you restart your node, Kubernetes must be restarted as well, using the following command:

minikube start --driver=none --apiserver-ips 127.0.0.1 --apiserver-name localhost

The Run:AI cluster will automatically start following the Kubernetes restart.

Deleting Run:AI¶

Run:

minikube delete --all

Next Steps¶

• Set up at least one Project Working with Projects.
• Review Quickstart Guides to run workloads.