In June 2020, the NSF awarded the National Center for Supercomputing Applications (NCSA) $10 million for its post-Blue Waters “Delta” supercomputer. Now, that funding has come to fruition: NCSA has announced that Delta has entered full production.

The HPE Cray-based Delta system (pictured in the header) includes 338 nodes. The 124 CPU nodes are powered by dual AMD Epyc “Milan” processors and 256GB of memory; 100 nodes have quadruple Nvidia A100 (40GB) GPUs, individual Milan CPUs, and 256GB of memory; 100 similarly built nodes swap out the A100s for quadruple Nvidia A40 GPUs; 5 nodes boast octuple A100s, dual Milan CPUs and 2TB of memory; and a single, lonely node has octuple AMD MI100 GPUs, dual Milan CPUs and 2TB of memory. (The eight remaining nodes serve utility functions.) The system is networked with HPE Slingshot-10, which will be upgraded to Slingshot-11 pending the next software release.

Following the funding in June 2020, NCSA completed installation of Delta in December 2021. Applications for time (via XSEDE) opened in January of this year. Delta was approved for full production this summer following an NSF panel assessment of system readiness. Given the hundreds of GPUs in the system, NCSA has been encouraging GPU-heavy applications but said that the process “wouldn’t preclude a CPU-only allocation.”

“The Delta project team is excited to move beyond the initial deployment to see the amazing discoveries to come from the teams making use of Delta,” said Brett Bode, an assistant director at NCSA and the deputy project director for Delta. “We are looking forward to working with teams to improve the use of GPUs for existing GPU codes, broaden the use of GPUs to new domains and improve the accessibility of Delta and HPC resources in general.”

While NCSA has often eschewed the popular Linpack benchmark, NCSA Director Bill Gropp had — at least as of a couple years ago — suggested that the benchmark may be run on Delta at some point alongside “some other, more relevant benchmarks as well.” However, Delta has not yet appeared on the Top500 list, nor has NCSA provided a flops estimate for the system, preferring to emphasize its utility to real-world research.

“Delta is a tremendous resource for AI and machine learning, as well as simulation,” Gropp said. “Combined with its high-performance file system and features for broader accessibility to communities that have not historically used HPC systems, Delta will help accelerate the adoption and use of these techniques into all areas of research. NCSA thanks the panel for their time, comments and recommendations, which will help us make Delta even more useful to the research community.”

To the point of being “even more useful to the research community,” Delta will be allocable through the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program that is succeeding XSEDE. Further, NCSA says it has partnered with the San Diego Supercomputer Center’s Science Gateways Community Institute to reach even more “traditional and emerging HPC user communities.”