Initial Goal of NAIRR Pilot Project, Also Supported by Allocations Software Developed by PSC and ACCESS Partners, Will Be to Explore Trustworthy AI

The Pittsburgh Supercomputing Center’s Neocortex AI system is among six national AI supercomputers chosen to participate in a pilot program to support novel and transformative AI research and education at a national scale. The National AI Research Resource (NAIRR) will also utilize allocations software and procedures developed by PSC and its partners in the National Science Foundation-funded ACCESS project.

The launch of the program in a proof-of-concept phase, announced by the NSF today, will initially focus on “trustworthy AI.” The NSF leads the NAIRR pilot effort in collaboration with 10 other federal agency partners, including the Department of Energy and the National Institutes of Health, which are leading development of a NAIRR Secure element of the pilot.

Neocortex is available for the NAIRR pilot project in part because of its innovative, AI-specialized design. Meant to advance the state of AI beyond what was possible with the graphics processing units (GPUs) pioneered in the mid 2010s, the system makes use of two Cerebras “wafer scale engines,” a new computing architecture meant to ease AI machine learning with massive data-handling ability. Neocortex featured an innovative pairing of two WSEs with a massive HPE Superdome Flex server that coordinates communications and computation between them.

“The National AI Research Resource program will play a key role in defining AI’s impact on scientific and engineering research, influencing daily life in the U.S. and across the globe,” said Paola Buitrago, PSC’s Director of Artificial Intelligence and Big Data and Principal Investigator for Neocortex. “We are thrilled to actively contribute to the program’s initial pilot phase through PSC’s Neocortex AI-specialized supercomputer. Our team is eager and prepared to provide cutting-edge computing technologies and specialized expertise, supporting researchers as they explore the frontiers of both fundamental and applied AI research. We believe in the program’s potential to shape the AI landscape and look forward to playing a key role in its success.”

NAIRR will eventually aim to spur innovation in the development of AI tools, increase the diversity of the AI workforce, and improve the nation’s capacity for AI research and development. The NSF plans to achieve this by providing scientists across the U.S. wide access to secure high-performance, privacy-preserving computing; high-quality datasets; catalogs of testbeds and educational materials; and training tools and user support mechanisms.

“We’re excited to participate in such an important, multiagency effort,” said Stephen Deems, PSC’s Assistant Director for Project Management and a member of the NAIRR Pilot’s Allocations Working Group that will coordinate scientists’ access to the program’s resources. “PSC is contributing access to its expert staff and Neocortex, as well as the software infrastructure for researchers and educators to request use of these federally funded computing resources.”

Deems is also the principal investigator for ACCESS Allocation Services, a program established and funded by the NSF to help researchers and educators, with or without supporting grants, to utilize the nation’s advanced computing systems and services at no cost.

The pilot launch of NAIRR will include a select group of six AI computing resources:

  • Neocortex, Pittsburgh Supercomputing Center
  • Delta AI, National Center for Supercomputing Applications
  • Summit, DOE Oak Ridge Leadership Computing Facility
  • Frontera and Lonestar6, Texas Advanced Computing Center
  • ALCF AI Testbed, DOE Argonne Leadership Computing Facility

Future goals for the NAIRR Pilot will include new computing resources, expanded datasets, improved “models” for generating AI output, and new software, among other AI tools. The NAIRR Allocations team is now accepting open proposals from research scientists, as well as input through a potential use-case survey. For more information, see the NAIRR pilot program website.