Dr. Lisa Perez, Texas A&M High Performance Research Computing, and David Schmidt, Dell Technologies
summary
ACES is a test‑bed for hardware composability that composes disaggregated resources (Dell R760s plus Nvidia and Intel GPUs) onto bare‑metal compute nodes so PyTorch workflows see accelerators as local rather than remote. The team runs an open on‑demand portal and short courses to get K–12 through grad users running jobs, removes provisioning friction, and demonstrates that pooling heterogeneous accelerators can simplify software models and speed time to experiment.
AI-written summary · about
