Amazon SageMaker HyperPod now supports flexible instance groups

This article announces that Amazon SageMaker HyperPod now supports flexible instance groups, simplifying multi-instance type and multi-subnet cluster management.

• Define multiple instance types and subnets in a single instance group
• Automatic fallback to lower-priority instance types when capacity unavailable
• Eliminates need to create separate instance groups per type/availability zone
• Reduces operational overhead for cluster configuration, scaling, and monitoring
• Training workloads benefit from multi-subnet distribution to avoid subnet exhaustion
• Inference workloads get automatic priority-based fallback without manual retries
• Karpenter autoscaling automatically detects and provisions optimal instance types
• Available via CreateCluster, UpdateCluster APIs, AWS CLI, and Management Console
• Supported on SageMaker HyperPod with EKS orchestrator across all AWS regions

Flexible instance groups streamline HyperPod cluster management by consolidating multiple instance types and subnets, reducing operational complexity and improving resource utilization.