NCAR wins award for Best Use of HPC in Physical Sciences

NCAR has won an HPCwire Editor's Choice award for Best Use of HPC in the Physical Sciences for developing the GPU-based FastEddy® model, which can run weather forecasts at a resolution of just five meters (16 feet).

FastEddy, which was developed primarily on CISL’s Casper cluster, is a new resident-GPU microscale large-eddy simulation model coupled with the Weather Research and Forecasting model. FastEddy is capable of providing real-time weather hazard avoidance at the microscale level and allows scientists to predict how weather and buildings in an urban environment affect drones and other small aerial vehicles.

Animation from a coupled WRF-FastEddy multiscale simulation supporting urban air mobility considerations. Covering approximately 81 square kilometers in the Dallas/Fort Worth, Texas, area during a November 2019 cold front passage, it shows near-surface wind speed across an area including street-canyon and building wake flow conditions.

Jeremy Sauer, an NCAR software engineer who co-developed FastEddy with project scientist Domingo Muñoz-Esparza in the Research Applications Lab (RAL), said its novel capacity for accelerated microscale, turbulence-resolving simulation “enables groundbreaking investigations of real-world, heterogeneous boundary-layer processes at the frontier of actionable science based on large-eddy-simulation.” For example, Sauer said, applied research in urban air mobility, fugitive emissions characterization, and wind energy resource assessment involves efforts across multiple RAL programs.

“Inevitably, the adoption of new HPC technologies and addressing computational logistics challenges in the applied research also required key contributions from CISL’s high-performance computing experts,” he added. “It’s great to see NCAR recognized for these kinds of collaborative and cross-laboratory successes.”

While the ongoing initiative makes use of the Casper system’s 64 Nvidia V100 GPUs, additional development efforts have achieved model execution across more than 12,000 Nvidia V100 GPUs of the Oak Ridge Leadership Computing Facility’s Summit architecture,  exposing a technologically disruptive pathway towards ultra-fine scale, hyper-localized weather research and forecasting.

Three additional projects nominated for HPCwire awards demonstrated exceptional NCAR accomplishments in the use of high-performance computing resources:

  • Best Use of HPC in Life Sciences – The ballot describes how NCAR scientists successfully used an interactive global atlas, an extremely high-resolution computer simulation of ocean circulation, to identify possible “thermal refugia” where coral reefs are most likely to survive warming ocean temperatures that threaten their existence. The scientists use Cheyenne and Casper to process and visualize more than 400 terabytes of data from the largest ocean simulation ever conducted.
  • Best HPC Response to Societal Plight (Urgent Computing, COVID-19) – NCAR scientists were nominated for developing a new way to accurately predict summer rainfall in certain regions, enabling better management of dwindling water resources. Using Cheyenne, they applied a specialized algorithm to the long-range forecasts of the leading weather models to determine how to make more accurate predictions months in advance.
  • Best Use of High Performance Data Analytics & Artificial Intelligence – A collaboration involving NCAR scientists and HPE led to the first demonstration of using high-throughput, machine-learning predictions inside an ensemble of realistic ocean climate model simulations, showing a 20% improvement in the predicted quantity of data points over the previous industry best, establishing a new standard.