DNA Insider
繁中
Back
Official Source· AWS Artificial Intelligence· AI· Published Oct 8, 2026

AWS outlines shared SageMaker HyperPod EKS cluster architecture for multiple teams

AWS has published a reference architecture for multiple teams sharing one Amazon SageMaker HyperPod EKS cluster. It combines identity and access controls, team namespaces, task governance, and cost allocation, while cautioning that Kubernetes namespaces are not hard security boundaries.

AWS outlines shared SageMaker HyperPod EKS cluster architecture for multiple teams

Full Report

Dated October 8, 2026, AWS’s reference architecture describes how multiple teams can share a single Amazon SageMaker HyperPod EKS cluster. Its example has two teams, each with a Kubernetes namespace, and offers access through CLI workflows or SageMaker Studio. In the CLI path, users run `aws sso login` to obtain temporary credentials and use `kubectl` to submit tasks.

The identity example pairs Microsoft Entra ID with AWS IAM Identity Center, with SCIM synchronization for provisioning and de-provisioning users and groups. EKS access entries map IAM roles to Kubernetes permissions scoped to each team’s namespace. HyperPod Task Governance is used for fairness, while namespace-level cost allocation supports chargeback.

For storage, the article describes Amazon FSx for Lustre or Amazon FSx for OpenZFS, with shared team directories and individual home directories; object storage can use team-specific or shared Amazon S3 buckets. AWS stresses that namespaces provide an isolation boundary, not a hard security boundary, and positions the design for teams within one organization—not mutually untrusted customers. The architecture adds implementation detail to AWS’s related HyperPod governance guidance by showing how identity, workload fairness, and costs can be handled in a shared-cluster setup.

Why This Matters

Organizations evaluating shared GPU capacity get a concrete pattern for tying team access, workload fairness, and cost allocation to one cluster. AWS’s qualification about namespace isolation is also important when deciding whether the architecture fits the trust boundaries of a deployment.

Evidence

Start with the primary evidence, then review supporting sources and context.

1 items

Primary Evidence

1
  • OfficialAWS Artificial IntelligenceDocumentOriginal

    Share GPU clusters across teams with isolation and fairness using Amazon SageMaker HyperPod

    Giuseppe Angelo Porcelli

    “A reference architecture for securely sharing one Amazon SageMaker HyperPod EKS cluster across multiple teams, using AWS IAM Identity Center for authentication, per-team SageMaker Domains and Kubernetes namespaces for isolation, HyperPod Task Governance for fairness, and namespace-level cost allocation for chargeback.”

Recommended signals

Recommended by brand, topic, and recent related coverage

No more recommended signals right now

Official statement

Open original source
Back to Live FeedExplore more signals