AWS PCS Gets Smarter GPU Scheduling with Custom GRES and MIG Support

AWS Parallel Computing Service now lets you fine-tune how Slurm schedules your HPC and ML workloads with custom GRES settings, hardware topology awareness, and NVIDIA MIG support—meaning you can partition single GPUs into isolated instances and optimize task placement across all AWS regions. Slurm 26.05 brings automatic NUMA domain detection and GPU core affinity, so your tightly-coupled MPI jobs land exactly where they should for maximum performance.