Amazon SageMaker HyperPod now provides comprehensive observability for Restricted Instance Groups

This article announces comprehensive observability capabilities for Amazon SageMaker HyperPod Restricted Instance Groups, enabling teams to monitor foundation model training with unified visibility across infrastructure and workloads.

• Monitor GPU utilization, NVLink bandwidth, CPU pressure, and FSx for Lustre usage from single dashboard
• Pre-configured Amazon Managed Grafana dashboard backed by Amazon Managed Service for Prometheus
• Metrics collected across four exporters covering GPU, system health, network, and Kubernetes state
• Curated logs automatically available for epoch progress, training logs, errors, and tracebacks
• Automatically enabled for new clusters; can be enabled for existing clusters with few clicks
• Available in all AWS Regions supporting SageMaker HyperPod RIG

SageMaker HyperPod now eliminates manual metric collection, providing unified observability for foundation model training infrastructure and workloads.