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
up to ~2,000 A (typical peak, per Vicor)Peak core current Vicor calls typical for AI/HPC processors at sub-1 V; the newest parts exceed it at full power
~70% / ~20% / ~10%AI/HPC scheduler share: Slurm / Kubernetes / in-house (rule of thumb)