We accelerate energy technologies through supercomputing and scientific collaboration.

Designing Advanced Catalysts With Boost From Artificial Intelligence

To design better catalysts, a collaborative research effort paired GPU-accelerated algorithms with AI to achieve an unprecedented description of electronic structure in speed and scale. Researchers leveraged Kestrel to perform high-fidelity calculations for a range of electrocatalysts and adsorbates, including previously unachievable calculations. Kestrel enabled the running of standard catalytic model sizes on 8 GPU nodes and up to 64 GPU nodes for complex models.

Funded by the U.S. Department of Energy Office of Science's Beyond-DFT Electrochemistry with Accelerated Solvated Techniques

National Transmission Planning Study Defines Long-Term Planning Scenarios For Growing Power System

For the National Transmission Planning Study, the Kestrel supercomputer's capabilities helped develop new grid-scale planning tools and methods to assist industry in planning for interregional transmission capacity needs.

NREL and Pacific Northwest National Laboratory collaborate on the multiyear National Transmission Planning Study in a quest to understand the changes needed to serve customers as the power sector evolves.

Funded by the U.S. Department of Energy Grid Deployment Office

First Macromolecular Model of Woody Plant Cell Walls Will Improve Biomass Conversion

After deploying a multidisciplinary approach to define the macromolecular arrangement of polymers in Populus wood, NREL researchers used Kestrel to develop molecular models of lignocellulosic biopolymer assemblies. A molecular-level understanding of lignocellulosic biomass will inform the design and optimization of pretreatment strategies for maximizing the conversion performance of the material.

Funded by the U.S. Department of Energy Bioenergy Technologies Office

More Resilient Polymer Membrane Options for Scalable Water and Biofuel Filtration

University of Connecticut researchers used the Kestrel supercomputer to model the resiliency of polymeric membranes for organic solvent nanofiltration—a promising option for purifying water, biofuels, and hydrogen. Using this model, they identified accurate and computationally efficient forcefields, enabling machine learning combined with simulations to explore the desalination performance of approximately 1,000 polyamide membranes.

Funded by the U.S. Department of Energy Industrial Technologies Office

Artificial Intelligence Algorithm Blends Atmospheric Measurements With Physics-Rich Simulation Data

NREL researchers developed an innovative machine learning approach that leverages large-eddy simulations, enabling the reconstruction of wind speeds across an entire turbine rotor disk from turbine hub height measurements. The enhanced reconstruction complements the physics-based computational models used to extrapolate atmospheric dynamics from measured to unmeasured regions.

Funded by the U.S. Department of Energy Wind Energy Technologies Office

Enhancing Energy Sector Readiness for Extreme Weather Events

Using Kestrel's new H100 GPUs, the research team designed new model architectures with larger effective receptive fields that greatly improve the quality of the downscaled climate data. Datasets available in 2025 will feature these improvements, helping the energy sector better plan for and manage severe weather events.