The San Diego Supercomputer Center (SDSC) at UC San Diego has announced a partnership with Core Scientific, a provider of infrastructure and software solutions for artificial intelligence and blockchain, to offer industrial users HPC functions. SDSC and Core will integrate the Core Scientific Plexus AI software stack into the Expanse Petascale supercomputer, launched by SDSC with support from the National Science Foundation (NSF) late last year.
The key to Expanse lies in the ability to support composable systems that integrate computing elements such as the central processing unit (CPU), graphics processing unit (GPU), and other resources into scientific workflows that can include data acquisition and processing, machine learning, and traditional simulation. While the main expansion system is funded by NSF, SDSC recently installed additional capacity to support industrial workloads.
“By deploying SDSC’s Expanse supercomputing platform through Core Scientific’s Plexus software stack, users receive a consumption-based high-performance computing (HPC) model that does more than just dissolve on-premise infrastructure,” said SDSC in its announcement , “But also the ability to run HPC workloads in supercomputer centers as well as in one of the four large public cloud providers – all from a single source. “
“We are excited to bring the power of the Expanse supercomputer to industry researchers and other customers in need of cutting-edge supercomputer capabilities,” said Ron Hawkins, SDSC director of industry relations. “Core Scientific’s Plexus software stack gives end customers access to our infrastructure in an intuitive, cloud-like point-and-click experience that is a potential game changer for supercomputer centers.”
Expanse, which went live last December, consists of nearly 800 AMD EPYC (Rome) -based compute nodes with a parallel 12 petabyte file system and a Mellanox InfiniBand HDR connection with more than 100,000 available cores.
“SDSC’s Expanse is a unique and innovative supercomputer that supports a wide range of scientific workloads,” said Ian Ferreira, chief product officer, artificial intelligence, Core Scientific. “By democratizing this infrastructure through our Plexus software stack, researchers and scientists can work seamlessly between CPU- and GPU-bound problem classes through a single, user-friendly interface.”