San Diego Supercomputer Center Receives NSF Grant to Enable AI Research on NVIDIA DGX Cloud Resources

Published December 5, 2024

By Kimberly Mann Bruch, SDSC Communications

San Diego Supercomputer Center (SDSC), part of the School of Computing, Information and Data Sciences at UC San Diego, has been awarded an Early-concept Grants for Exploratory Research (EAGER) award from the U.S. National Science Foundation (NSF) to provide IT support to research groups who will use  the NVIDIA DGX Cloud platform. SDSC will focus on optimization of system setups, performance monitoring and determining the best ways to run National Artificial Intelligence Research Resource (NAIRR) Pilot projects on the resources.

“Our work will focus on supporting dedicated systems for each research group with NAIRR Pilot awards,” said Principal Investigator (PI) Mahidhar Tatineni, who is the user services director for SDSC’s high-performance computing (HPC) systems. "Unlike systems that are shared by many users, the dedicated systems will allow for researchers to have uninterrupted access to perform modeling that require weeks or months to complete. For example, some tools need special configurations that aren’t possible in a shared system environment, and NVIDIA DGX Cloud allows us to work with NAIRR Pilot researchers to create custom environments for their work.”

Tatineni explained that once the researchers have created the right setup, they can have the best of both worlds by using cloud servers and physical servers on premises at supercomputing centers to increase the potential for new science discovery.

“In short, our EAGER project will provide guidelines, training and tools to help NAIRR Pilot researchers use cloud platforms like NVIDIA DGX Cloud more effectively—making it easier to optimize their work and speed up research processes,” Tatineni said.

NVIDIA DGX Cloud is a high-performance, fully managed AI platform for generative AI development that provides developers access to the latest NVIDIA accelerated computing architecture. The platform offers dedicated access to capacity that supports the lifecycle of AI, from building and customizing leading foundation models to serverless inference.

“These attributes make the system ideal for NAIRR Pilot projects that need large-scale AI resources for focused research campaigns,” Tatineni said.

NVIDIA has contributed significant DGX Cloud resources to the NAIRR Pilot program. Usage of DGX Cloud is aimed at enabling proposals that can benefit from dedicated access to a 32-node cluster for sustained AI computing campaigns that are anticipated to take weeks or months to complete.

NAIRR connects U.S. research and education communities to responsible and trustworthy AI resources, as well as computational, data, software, training and educational resources to advance research, discovery and innovation.

Experts at SDSC have been involved with NAIRR activities since the start. Former SDSC Director and UC San Diego Distinguished Professor of Astrophysics Michael Norman was one of the 12 NAIRR Task Force members who, at the request of the Biden Administration in 2021, worked to develop a blueprint for a national approach to AI. In January 2023, the NAIRR Task Force submitted its final report/implementation plan, which guided the creation of the NAIRR Pilot program.

Major SDSC contributions include:

• the NAIRR Pilot portal, developed by SDSC’s SGX3 Center of Excellence for Science Gateways and led by Michael Zentner, retired director of SDSC’s former Sustainable Scientific Software Division;
• computational resources from the Expanse supercomputer (CPU, GPU allocations) and the Voyager resource featuring Habana processors customized for deep learning; 
• the National Research Platform (NRP), led by SDSC Director Frank Würthwein, offering an array of NAIRR classroom resources and
• the National Data Platform (NDP) NAIRR Pilot Demonstration Project, led by SDSC Chief Data Officer and NDP Principal Investigator Ilkay Altintas.

The EAGER grant will be instrumental in enabling researchers to make effective use of their NAIRR Pilot awards on the NVIDIA DGX Cloud platform. The project is funded by the NSF (award no. 2438294).