HPC · High-Performance Computing
The scientific supercomputing tradition AI clusters descend from — gang scheduling, low-latency fabrics, parallel filesystems — now repurposed for training at scale.
Current numbers
~70% / ~20% / ~10%AI/HPC scheduler share: Slurm / Kubernetes / in-house (rule of thumb)
500–2,000 AVery-high-current delivery range for modern AI/HPC processors at sub-1V core voltages; peak requirements reach approximately 2,000 A
1.07 µsHDR InfiniBand: 8-byte host-memory MPI ping-pong mean, EPYC Rome / ConnectX-6 (HPC Advisory Council, July 2020)