CuraDevOps

tag: Ml-Infrastructure · 3 items

Learn AWS Neuron 2.33 adds TorchNeuron beta and Neuron Observer beta
2026-10-22 · AWS What's New · source ↗ #aws-neuron#ml-infrastructure#trainium
  • Platform/SRE — Learn: TorchNeuron and Neuron Observer are both public beta, so no production action is warranted yet; worth tracking for teams running Trainium workloads on Trn2/Trn3 instances.
  • CI/CD — Skip
  • Leader — Skip
Plan archived EC2 Capacity Reservation Resource Groups now support Capacity Blocks and interruptible reservations
2026-08-26 · AWS What's New · source ↗ #aws-ec2#capacity-management#ml-infrastructure
  • Platform/SRE — Plan: If your platform manages ML workloads or uses mixed reservation types, this GA change lets you consolidate Capacity Blocks and interruptible ODCRs into unified resource groups with prioritization and On-Demand fallback — worth incorporating into capacity planning and Auto Scaling group configs this quarter.
  • CI/CD — Skip
  • Leader — Skip
Learn archived SageMaker HyperPod adds partition-level topology for Slurm clusters
2026-07-18 · AWS What's New · source ↗ #ml-infrastructure#aws-sagemaker#hpc-scheduling
  • Platform/SRE — Learn: If you operate HyperPod Slurm clusters for distributed ML training, partition-level topology is now automatic and enabled by default with Slurm 25.11+; no action required, but worth knowing that block vs. tree topology is now applied per-partition based on instance type.
  • CI/CD — Skip
  • Leader — Skip