By enabling custom Generic Resource (GRES) declarations and hardware topology configuration, AWS PCS improves job placement for HPC and machine learning workloads. Slurm 26.05 integration treats NUMA domains as sockets and auto-detects GPU affinities, helping tightly coupled MPI and GPU tasks optimize resource locality.
Support for NVIDIA Multi-Instance GPU (MIG) allows GPUs partitioned into isolated instances to be scheduled independently by Slurm, increasing resource utilization efficiency. AWS PCS manages cluster updates and observability, offering scalable HPC environments that integrate these advances across all AWS Regions where the service is available.
Source & further reading
AWS PCS now supports custom GRES, hardware topology, and MIG ↗Source published 2026-09-11. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.