Cloud Provider

Create a managed Kubernetes cluster in the cloud to be used as a backend for job execution. This guide provides links to setup instructions for various cloud providers.

Prerequisites

You will need access to a cloud provider account (AWS, Azure, or GCP) with permissions to create and manage Kubernetes clusters.

Important

• Kubernetes Version: v1.30.0 or later

• Networking: Ensure the nodes in the cluster have outbound internet access to container registries and access to the OSMO service

Setup Guide

Cloud Provider	Documentation	Recommended Instance Types
Amazon Web Services (EKS)	Getting started with EKS Creating an EKS cluster EKS nodes	Backend-Operator: m5.xlarge (4 vCPUs, 8 GB RAM, auto-scaling: 1-3 nodes) Compute (CPU): m5.2xlarge Compute (GPU): p3.2xlarge, p4d.24xlarge (A100), p5.48xlarge (H100) - GPU instances
Microsoft Azure (AKS)	Quickstart: Deploy AKS Create an AKS cluster Use multiple node pools	Backend-Operator: Standard_D4s_v3 (4 vCPUs, 8 GB RAM, auto-scaling: 1-3 nodes) Compute (CPU): Standard_D8s_v3 Compute (GPU): Standard_NC6s_v3, Standard_ND96asr_v4 (A100), Standard_ND96isr_H100_v5 (H100) - GPU VMs
Google Cloud Platform (GKE)	Quickstart: Deploy GKE Creating a GKE cluster Node pools	Backend-Operator: n1-standard-4 (4 vCPUs, 8 GB RAM, auto-scaling: 1-3 nodes) Compute (CPU): n1-standard-8 Compute (GPU): n1-standard-4 with T4/A100, a2-highgpu-1g (A100), a3-highgpu-8g (H100) - GPUs on Compute Engine

Note

Configure auto-scaling for compute nodes based on your expected workload patterns.