Tommy Minyard, TACC | SC23
summary
TACC announces Stampede3 as an Intel Sapphire Rapids system with at least 560 liquid‑cooled Dell nodes (60 kW per rack), funded by a $10M NSF award and running five years, plus 80 Intel Ponte Vecchio GPUs for experimentation; a separate Vista Arm deployment uses NVIDIA Grace and 300 Grace Hopper nodes to test unified memory AI workflows. Emphasis lands on high‑bandwidth interconnects, surrogate/proxy models to speed optimization, open standards for portability, and practical issues like data hygiene, MFA security, and workload partitioning.
AI-written summary · about
