Amazon EKS
User Guide

Amazon EKS-Optimized AMI with GPU Support

The Amazon EKS-optimized AMI with GPU support is built on top of the standard Amazon EKS-optimized AMI, and is configured to serve as an optional image for Amazon EKS worker nodes to support GPU workloads.

In addition to the standard Amazon EKS-optimized AMI configuration, the GPU AMI includes the following:

  • NVIDIA drivers

  • The nvidia-docker2 package

  • The nvidia-container-runtime (as the default runtime)

The AMI IDs for the latest Amazon EKS-optimized AMI with GPU support are shown in the following table.


The Amazon EKS-optimized AMI with GPU support only supports P2 and P3 instance types. Be sure to specify these instance types in your worker node AWS CloudFormation template. Because this AMI includes third-party software that requires an end user license agreement (EULA), you must subscribe to the AMI in the AWS Marketplace and accept the EULA before you can use the AMI in your worker node groups. To subscribe to the AMI, visit the AWS Marketplace.

Kubernetes version 1.11

Region Amazon EKS-optimized AMI with GPU support
US West (Oregon) (us-west-2) ami-0c9e5e2d8caa9fb5e
US East (N. Virginia) (us-east-1) ami-0ff0241c02b279f50
US East (Ohio) (us-east-2) ami-006a12f54eaafc2b1
EU (Frankfurt) (eu-central-1) ami-0d6f0554fd4743a9d
EU (Stockholm) (eu-north-1) ami-0b159b75
EU (Ireland) (eu-west-1) ami-097978e7acde1fd7c
Asia Pacific (Tokyo) (ap-northeast-1) ami-036b3969c5eb8d3cf
Asia Pacific (Seoul) (ap-northeast-2) ami-0b7f163f7194396f7
Asia Pacific (Singapore) (ap-southeast-1) ami-093f742654a955ee6
Asia Pacific (Sydney) (ap-southeast-2) ami-05e09575123ff498b

Kubernetes version 1.10

Region Amazon EKS-optimized AMI with GPU support
US West (Oregon) (us-west-2) ami-0ebf0561e61a2be02
US East (N. Virginia) (us-east-1) ami-0131c0ca222183def
US East (Ohio) (us-east-2) ami-0abfb3be33c196cbf
EU (Frankfurt) (eu-central-1) ami-000622b1016d2a5bf
EU (Stockholm) (eu-north-1) ami-cc149ab2
EU (Ireland) (eu-west-1) ami-0dafd3a1dc43781f7
Asia Pacific (Tokyo) (ap-northeast-1) ami-0afc9d14b2fe11ad9
Asia Pacific (Seoul) (ap-northeast-2) ami-0d75b9ab57bfc8c9a
Asia Pacific (Singapore) (ap-southeast-1) ami-0ecce0670cb66d17b
Asia Pacific (Sydney) (ap-southeast-2) ami-03b048bd9d3861ce9


These AMIs require the latest AWS CloudFormation worker node template. You cannot use these AMIs with a previous version of the worker node template; they will fail to join your cluster. Be sure to upgrade any existing AWS CloudFormation worker stacks with the latest template (URL shown below) before you attempt to use these AMIs.

The AWS CloudFormation worker node template launches your worker nodes with Amazon EC2 user data that triggers a specialized bootstrap script that allows them to discover and connect to your cluster's control plane automatically. For more information, see Launching Amazon EKS Worker Nodes.

After your GPU worker nodes join your cluster, you must apply the NVIDIA device plugin for Kubernetes as a daemon set on your cluster with the following command.

kubectl apply -f

You can verify that your nodes have allocatable GPUs with the following command:

kubectl get nodes ",GPU:.status.allocatable.nvidia\.com/gpu"

Example GPU Manifest

This section provides an example pod manifest for you to test that your GPU workers are configured properly.

Example Get nvidia-smi output

This example pod manifest launches a Cuda container that runs nvidia-smi on a worker node. Create a file called nvidia-smi.yaml, copy and paste the following manifest into it, and save the file.

apiVersion: v1 kind: Pod metadata: name: nvidia-smi spec: restartPolicy: OnFailure containers: - name: nvidia-smi image: nvidia/cuda:9.2-devel args: - "nvidia-smi" resources: limits: 1

Apply the manifest with the following command:

kubectl apply -f nvidia-smi.yaml

After the pod has finished running, view its logs with the following command:

kubectl logs nvidia-smi


Mon Aug 6 20:23:31 2018 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 396.26 Driver Version: 396.26 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 Tesla V100-SXM2... On | 00000000:00:1C.0 Off | 0 | | N/A 46C P0 47W / 300W | 0MiB / 16160MiB | 0% Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | No running processes found | +-----------------------------------------------------------------------------+

On this page: