ISCL awarded supercomputer access at Lawrence Berkeley National Laboratory

UNIVERSITY PARK, Pa. — The National Energy Research Scientific Computing Center (NERSC) awarded a significant computing allocation to the Interdisciplinary Scientific Computing Laboratory (ISCL). The lab received 32,000 GPU compute node hours and 20,000 CPU hours on the supercomputer, Perlmutter, at the Lawrence Berkeley National Laboratory.

ISCL is housed in the Penn State College of Information Sciences and Technology (IST) and the Mathematics and Computer Science Division at Argonne National Laboratory. Romit Maulik, assistant professor in the College of IST, is the lab’s principal investigator. Maulik is also a co-hire of Penn State’s Institute of Computational and Data Sciences and holds a faculty joint appointment at Argonne.

NERSC, the mission science computing facility for the U.S. Department Energy (DOE) Office of Science, provides high-performance computing and data resources and services. It enables computational science of scale in which large interdisciplinary teams of researchers attack fundamental problems in science and engineering that require massive calculations and that have broad scientific and economic impacts, according to the NERSC website.

ISCL responded to NERSC’s call for AI for Science (AI4 Science), inviting proposals for projects that would leverage Perlmutter to advance the state of the art AI for science and produce novel science outcomes. Projects were evaluated on scientific significance and innovation, technical feasibility and readiness for effective leveraging of NERSC’s computational resources, among other criteria.

“We are pleased with this significant computing allocation,” Maulik said. “Xuyang Li, a postdoctoral scholar in the College of IST, was the lead writer for our proposal, which we titled, ‘Scientific Machine Learning for Fluid Dynamics Surrogate Modeling and Multiscale Data Recovery.”

The lab will be developing new AI models to accelerate computer simulations of various physical processes, such as fluid dynamical systems — for example, the weather or simulations of nuclear fusion reactors. According to Maulik, using these models will dramatically reduce the time it takes to make decisions, such as controlling the reactor for preventing a disruptive event or making logistics decisions for extreme weather.

“The allocation lets us use the supercomputer’s GPUs and CPUs for our numerical experiments,” Maulik said. “We will use them to prototype our AI models and then deploy them for large problems typically infeasible at university-scale resources.”