Prerequisites for Installation

This topic describes the prerequisites required for installing Elastic GPU Service (EGS). It is essential to have a well-configured environment that fulfills the installation requirements for EGS.

Kubernetes Cluster Requirements

• A Kubernetes cluster with GPU-enabled nodes.
• To support fractional GPU usage, the GPU must support MIG (Multi-Instance GPU), such as NVIDIA A100.
• GPU Operator must be installed on the cluster.
  • If MIG capability and shared GPU features are required, ensure that the GPU Operator is properly configured. See GPU Operator with MIG for more details.

Monitoring and Observability

• NVIDIA GPU Operator is running and the nvidia-dcgm-exporter pod is active.
• Prometheus is deployed with Persistent Volume Claims (PVCs) for data retention.
• Prometheus is configured to scrape metrics from the DCGM exporter.
• Grafana is installed and configured with the NVIDIA DCGM dashboard.

Kubernetes Cluster Access and Permissions

• An admin kubeconfig file to access the Kubernetes cluster.
• The cluster must have outbound internet access to pull container images.
• Permissions to create the kubeslice-controller, kubeslice-system, and kubeslice-<project_name> (replace with your project name) namespaces.
• Permissions to create Kubernetes LoadBalancer services.

PostgreSQL Database and Ingress

• A PostgreSQL database with persistence storage using PVCs.
• An Ingress controller (for example, NGINX) installed on the cluster.

Required Tools

The following command-line tools must be installed on the system or workstation where EGS will be deployed. These tools enable interaction with the Kubernetes cluster and help manage the installation process.

Tool	Description	Installation Guide	Minimum Version
Helm	Kubernetes package manager	Install Helm	3.15.0
kubectl	Kubernetes CLI	Install kubectl	1.23.6
kubectx and kubens	CLI tools for switching clusters and namespaces	Install kubectx & kubens	—
jq	Command-line JSON processor	Refer to jq installation	1.6.0
yq	Command-line YAML processor	Refer to yq installation	4.44.2

EGS License

A valid license is required to install EGS. To get a trial license, you must register with Avesha. To register, visit the EGS registration page.

To know more about how to install a license, see License Management.