Oak Ridge National Laboratory (ORNL) has deployed a new suite to enhance its research capabilities with secure, trustworthy, and energy-efficient AI. The deployment aims to tackle problems of national importance, making ORNL the latest US national lab to adopt this advanced AI technology. ORNL is renowned for its world-changing energy, security, medicine, and climate science research.
The lab will use this technology to create an AI Center of Excellence, elevating the use of AI in its key research areas. ORNL plans to deploy two DataScale nodes, each powered by 16 advanced AI chips. Prasanna Balaprakash, Director of AI Programs at ORNL, said, “Our research is focused on secure, trustworthy, and energy-efficient AI.
We intend to leverage these systems’ capabilities to enhance our work. Their fast inference, energy efficiency, and Composition of Experts architecture will significantly boost our AI for science portfolio.
The platform allows multiple models to run and be queried in parallel, combining their outputs to improve prediction accuracy. This transformative capability is beneficial for a range of problems across multiple domains.
ORNL will utilize models trained on diverse data types—such as scientific texts and images—to aggregate predictions for better results. Scientists at ORNL conduct research in areas including material science, climate modeling, neutron science, and nuclear fusion and fission.
AI advancements at ORNL
The platform will be utilized for parallel inferencing across models trained on data from all of these fields. One notable application will involve inference on a foundational climate model initially trained on ORNL’s Frontier supercomputer. Initial steps will focus on evaluating various models to assess their behavior and trustworthiness under different conditions.
Eventually, ORNL scientists can query diverse models across multiple scientific domains, facilitating cross-disciplinary research. For example, those working on stable fusion reactor designs will query models trained in different branches of materials science to uncover connections that would have otherwise gone unnoticed. The fast inference capabilities and enhanced energy efficiency were key factors in ORNL’s decision.
As the lab evaluates different models, the platform can scale up reasoning tasks while using less energy compared to Frontier. “It’s much more efficient to do the inference on a platform specialized for faster inference,” Balaprakash noted. “Achieving this with a fraction of the energy costs is a significant advantage.”
Balaprakash expects the deployment to boost ORNL’s productivity and lead to more scientific discoveries.
The ability to extract information across multiple domains is incredibly exciting. It could significantly enhance the scientific work we do at ORNL.”
Cameron is a highly regarded contributor in the rapidly evolving fields of artificial intelligence (AI) and machine learning. His articles delve into the theoretical underpinnings of AI, the practical applications of machine learning across industries, ethical considerations of autonomous systems, and the societal impacts of these disruptive technologies.
