The National Center for Supercomputing Applications (NCSA) has been given a grant to upgrade its Delta supercomputer with an AI partition.
Hosted at the University of Illinois, the NCSA has been awarded $10 million by the National Science Foundation to create DeltaAI with Nvidia H100 GPUs.
According to the NSF award announcement, the upgrade will see “more than 300 next-generation Nvidia graphics processors delivering over 600 petaflops of half-precision floating point computing, distributed across an advanced network interconnect for application communications and access to an innovative, flash-based storage subsystem.”
Launched in April 2022 and ready for use in October of that year, Delta has 124 Apollo 2500 nodes each with a pair of 63-core AMD Milan Epyc 7763 CPUs, 100 Apollo 6500 nodes with one Milan CPU and four 40GB Nvidia A100 accelerators, and another 100 Apollo 6500 nodes with four Nvidia Ampere A40 accelerators.
This combines for a total of six petaflops of computing ability and was described at the time of entering operations as being a “tremendous resource for AI and Machine learning, as well as simulation” by NCSA director Bill Gropp.
Of the new grant, Gropp said: “We’re very excited about this expansion in our ability to support AI research. DeltaAI provides about double the performance of Delta and will be integrated into the same complex while also remaining separate projects.”
While the announcement has not detailed how the upgrade will be done, The NextPlatform has theorized that it could be 38 servers with eight H100 SXM5 GPU accelerators which would bring the GPU count to 304 and 601.6 petaflops peak at FP16 on the H100 Tensor Cores with sparsity on while meeting the budgetary requirements.
If this were the case, Delta and DeltaAI would reach a peak performance of 16.6 petaflops (Linpack) and 732.8 petaflops for AI work. DCD has reached out to the NCSA for more information.
The NCSA is home to several high-performance computing (HPC) clusters, including Nightingale, Radiant, vForge, Hydro, and the HAL Cluster. Delta is the most powerful HPC at the site.