GPU’S STORM INTO NASA’S CLIMATE SIMULATIONS

by Will Park

NVIDIA GPU technology isn’t just for gamers; it will soon help make your local weather forecast more accurate.

NASA meteorologists in the Goddard Space Flight Center’s Sciences and Exploration Directorate are leveraging GPU computing technologies to create more accurate global climate models – and is the focus of today’s CUDA Spotlight. Scientist Bill Putman’s team has dramatically improved the amount of information from NASA’s Goddard Earth Observing System atmospheric general circulation model version 5 (GEOS-5) thanks to GPUs.

Left to right: Matt Thompson, Bill Putman, Max Suarez

Most climate models in use today are typically run at global resolutions of 100 to 200 km. This level of resolution limits the ability to model cloud-system behaviors, and requires that these models essentially make approximations (also known as “parameterization”) of cloud-system convection currents and other movements. Simulating climate interactions is no easy task – just ask your local weatherman.

GEOS-5, on the other hand, has the power to simulate cloud-systems at much finer resolutions. Combined with GPU acceleration, Bill expects to be able to resolve cloud systems down to 3.5 km. At this level, NASA can examine direct representations of convection behaviors rather than rely on deep convective parameterization. Ordinarily, such a model would require hundreds of thousands of CPU cores.

Enter the GPU. Putman says, “GPU computing presents an opportunity to improve the efficiency of these leading edge Earth system simulations.” Bill goes on to explain that GPU acceleration will make it possible to bring some of the power of GEOS-5 to academic researchers who only have access to “a small box of GPUs.”

Next stop: more accurate local weather predictions using GPU computing. Stay tuned.