Introducing GPU Health Monitoring and Auto Repair for Amazon ECS Managed Instances

Amazon Elastic Container Service (Amazon ECS) now offers NVIDIA GPU health monitoring and auto repair functionality for Amazon ECS Managed Instances. The new capability automatically detects critical NVIDIA GPU hardware failures and replaces impaired instances, helping customers improve the availability and reliability of their GPU-accelerated containerized workloads.

Running GPU-accelerated workloads, such as GenAI inference, requires specialized hardware management to mitigate failures and minimize disruption. Amazon ECS Managed Instances now continuously monitor GPU health using NVIDIA Data Center GPU Manager (DCGM) and proactively replace impaired capacity when critical failures occur. You can monitor GPU health through the DescribeContainerInstances API and receive notifications through Amazon EventBridge when instances become impaired. For workloads where you prefer to manage instance lifecycle manually, you can opt out of auto repair at the capacity provider level and handle GPU error events with your own remediation logic.

GPU health auto repair is enabled by default on all Amazon ECS Managed Instances running on supported NVIDIA GPU instance types at no additional cost. The capability is available in all AWS Commercial Regions. To learn more, visit the Amazon ECS Developer Guide.