A $1.2 million award from the National Science Foundation will support the addition of a powerful cluster of Graphics Processing Units (GPUs) with Artificial Intelligence (AI) and Machine Learning (ML) capability, to NCShare’s existing suite of services.
The NCShare Accelerating Impact — GPUs-as-a-Service (NCShare AI-GaaS) — platform will consist of 32 high-end GPUs and provide an environment that will serve researchers and educators from participating universities, including HBCUs.
New Partners
NCShare also welcomes two new participating universities. North Carolina Agricultural and Technology University (NC A&T) and University of North Carolina, Chapel Hill (UNC-CH) now join Davidson College, North Carolina Central University, and Duke University, all in conjunction with MCNC as the primary service provider.
In addition, four other universities have signed on as early adopters of NCShare AI-GaaS to advance their AI/ML research and education: Fayetteville State University, North Carolina State University, UNC-Charlotte, and Winston-Salem State University.
Each NCShare participant is committed to advancing scientific computing and enhancing STEM education. Together, the nine schools who participate in the NCShare computational platform aim to serve high-end research needs on their campuses, and to democratize educational access to GPUs across small and large schools in North Carolina.
New Service – NCShare AI-GaaS
The NCShare AI-GaaS environment will enhance existing services, enabling research into foundational AI models and support applied AI research. It will also provide undergraduate and graduate students with access to modern computational environments and give them firsthand experience with AI tools and technologies.
AI-GaaS becomes the latest in a suite of services NCShare has developed, which already include the shared Science network (NCShare DMZ) and a shared computational facility (NCShare CaaS).
“We already have a lot of GPUs in individual labs throughout the university,” said Kaushik Roy, chair of Computer Science at North Carolina A&T University. “NCShare will give our researchers access to a larger concentration of high-end GPUs that they can use when their research needs exceed what’s available in their lab.”
The platform will employ a novel architecture and shared-services model, adapting and extending mature techniques used at R1s that enable virtualization of powerful GPUs, and applying those same virtualization techniques within a shared, multi-institutional environment. This will create an affordable and scalable platform for AI and ML research and for advancing Foundational AI Research and Applied Research in AI.
New Collaborations
In addition to enhancing teaching and learning, NCShare AI-GaaS provides opportunities for deep and meaningful collaborations across disciplines, and between technologists and faculty. The structure will offer possibilities to create shared methodologies in fields such as computer science, engineering, health sciences, statistics, and physics, and, ultimately, a shortened path to discovery.
Beyond the initial participating institutions, the North Carolina School of Science and Mathematics, the state’s residential high school and STEM magnet school, has expressed interest in the NCShare AI-GaaS project. Secondary school participation has the potential to usher in a novel set of new collaborators.
As needs continue to grow, and opportunities for innovation develop, NCShare invites future collaborators to the NCShare project ecosystem. To learn more, contact info@ncshare.org.
NSF 2430141 – NCShare AI-GaaS
Duke, UNC, NC A&T, MCNC
The Challenge
- Deploy a shared GPU infrastructure to support research and education at both small and large institutions
- Use virtualization techniques to support equally well two underserved extremes on the GPE demand cure: (1) Small uses at under-resourced colleges and MSIs and (2) large aggregated GPU capacity to support AI research by R1s and others.
Solutions/Deliverables
- Automated provisioning, containerization, GPU partitioning, advanced job scheduling, and federated login via either InCommon or bilateral federation with NCShare
- Horizontally scalable, shared environment of 32 H100s, with R1s and others able to buy-in with additional nodes
- Ensure sustainability /cost-effectiveness via a shared service from NC’s mature network operator (MCNC)
Scientific Impact/Broader Impact
- Support at least 30 researchers across nine NC institutions, with advanced GPU access
- Enable development of richer AI methods, further research, and strengthen CI practices to secure shared GPU resources
- Broaden access to modern GPU resources and AI techniques for college students; to improve education outcomes; increase science literacy
- Democratize access to GPUs for NC MSIs and smaller colleges