Amazon ECS now auto-repairs failing GPUs and instances

Amazon added automatic detection and recovery of impaired instances and GPUs to ECS: integration with NVIDIA DCGM catches hardware faults, and the system drains and replaces the instance on its own. It works on ECS Managed Instances and Fargate.

• ECS integrates with NVIDIA DCGM to watch for GPU error classes
• Impaired instances are automatically drained and recycled
• The feature works on ECS Managed Instances and AWS Fargate
• ECS patches OS, kernel and GPU drivers with automatic rollback