Scientific Research Transformed by NVIDIA AI Infrastructure, Achievements of the NAIRR Program

The U.S. National Science Foundation (NSF)'s NAIRR (National Artificial Intelligence Research Resource) pilot program has supported over 700 projects over the past two years, leading to innovative research across various fields such as protein prediction and infectious disease management. NVIDIA provided cloud-based resources giving researchers dedicated access to NVIDIA DGX nodes and offered technical support.

A key research case is the physics simulation work by Polymathic AI. Using NVIDIA GPUs and NVLink technology, they developed 'Walrus,' a large-scale foundation model for fluid dynamics. This model was released to overcome the scale and diversity limitations of existing physics pretraining environments.

A research team at the University of Michigan developed the MIST (Molecular Insight SMILES Transformers) model for energy storage technology. This model is a framework that combines domain-specific molecular AI with general-purpose LLMs, helping to explore chemical space and identify next-generation energy materials.

The research team developed the model using a 40-GPU NVIDIA DGX cluster and 200,000 GPU hours secured from ALCF's Polaris cluster. They also utilized NVIDIA NGC PyTorch containers to build a reproducible, GPU-accelerated development environment across different clusters.