Amazon SageMaker HyperPod Slurm clusters now support specifying minimum capacity requirements with continuous provisioning

This article announces that Amazon SageMaker HyperPod now supports minimum capacity requirements (MinCount) for Slurm clusters with continuous provisioning.

• MinCount specifies minimum instances required before cluster becomes available for job scheduling
• Particularly useful for distributed training frameworks like PyTorch FSDP and Megatron-LM
• Guarantees baseline GPU count for SLA and cost-efficiency targets
• Set MinInstanceCount via CreateCluster or UpdateCluster API requests
• Instance group remains in Creating status until threshold met, then transitions to InService
• System automatically rolls back if MinCount cannot be satisfied within 3 hours
• Available in all AWS Regions supporting Amazon SageMaker HyperPod

MinCount provides greater control over cluster availability for distributed training workloads requiring guaranteed minimum node capacity.