Amazon SageMaker HyperPod added support for on-demand deep health checks targeting GPU accelerators in both Amazon EKS and Slurm-orchestrated clusters. This allows users to execute comprehensive hardware stress and connectivity tests on running instances before workload scheduling, helping prevent delays caused by unhealthy nodes. Slurm clusters also gained the ability to perform these checks during node provisioning at cluster creation time.
The feature provides visibility at both instance group and individual instance levels through the SageMaker console and APIs, identifying issues with GPU health, network connectivity, and multi-node communications. Failed instances are automatically isolated and rebooted or replaced via HyperPod’s automatic recovery mechanisms, improving cluster reliability. This capability is available in all AWS regions supporting SageMaker HyperPod.
Source & further reading
Amazon SageMaker HyperPod now supports on-demand deep health checks ↗Source published 2026-04-17. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.