Workshop: Thirteenth SC Workshop on Best Practices for HPC Training and Education
Authors: Uthpala Herath and Katie Kilroy (Duke University)
Abstract: Most introductory HPC training assumes a local cluster, local staff, and a local allocation process-resources many institutions lack. NCShare, an NSF-funded coalition of North Carolina institutions, operates shared statewide cyberinfrastructure, including NVIDIA H200 GPUs, open to all member campuses. We describe the design and operation of the NCShare HPC Workshops, a free series taking participants from no prior HPC experience to a completed scientific job. Co-taught by NCShare staff and instructors across member institutions, the curriculum covers access, storage and data movement, containers, CPU-parallel and GPU-accelerated applications, and scientific visualization. Two design decisions follow from the shared infrastructure. With no environment-module system, containers are the primary route to scientific software making a participant's first environment portable and reproducible. Because allocation latency is the main barrier to participation, workshops run on a dedicated scheduler partition. We report workshop assessment, post-workshop usage, and release all materials for reuse.
Back to Thirteenth SC Workshop on Best Practices for HPC Training and Education Archive Listing
Back to Full Workshop Archive Listing