View a markdown version of this page

EKS-optimized accelerated AMIs for AI/ML workloads - Amazon EKS

Help improve this page

To contribute to this user guide, choose the Edit this page on GitHub link that is located in the right pane of every page.

EKS-optimized accelerated AMIs for AI/ML workloads

Tip

Register for upcoming Amazon EKS AI/ML workshops.

Running AI/ML workloads on Amazon EKS requires an operating system image that includes the GPU drivers, kernel modules, and container runtime configuration that accelerated instances depend on. Amazon EKS publishes EKS-optimized accelerated AMIs that bundle these components with the standard EKS node components, so your nodes can run GPU-accelerated containers as soon as they join the cluster.

This section covers the accelerated AMI options for both node operating systems:

  • Use EKS-optimized accelerated Linux AMIs for GPU instances — The EKS-optimized Amazon Linux and Bottlerocket accelerated AMIs, which include the NVIDIA and AWS Neuron drivers for G, P, Inferentia, and Trainium instances. This topic also covers which instance types each AMI variant supports and how to find the latest AMI IDs.

  • Build EKS-optimized Windows AMIs for GPU instances — The options for GPU-accelerated Windows containers, including building a custom EKS Windows optimized AMI with the GPU drivers pre-installed, or installing the drivers on worker nodes after launch.