UW–Madison home to weather-predicting supercomputer

A new supercomputer designed to run weather prediction models is now the most powerful computer of its kind on the University of Wisconsin–Madison campus, where researchers will help make those models more accurate.

The National Atmospheric and Oceanic Administration approached UW–Madison’s Space Science and Engineering Center with a $1 million grant to design and install the Supercomputer for Satellite Simulations and Data Assimilation Studies (known as S4).

NOAA maintains an array of complex computer models translating data on wind, moisture and temperature into weather and climate predictions that guide farmers, pilots, ships at sea and more. Juggling all the required data is a demanding and intensely complicated computing task.

“We will be running the NOAA Global Forecast System model, which provides weather forecast information for the United States every day,” says Liam Gumley, an SSEC researcher and principal investigator for the project. “Every grid point in the model has to represented in the computer’s memory, and for every grid point there are dozens of parameters to track.”

Researchers like those at SSEC are often proposing improvements to NOAA models based on evolving atmospheric science and new satellite capabilities. But too much rides on the model’s smooth operation to incorporate new ideas without thorough testing. And the other two NOAA supercomputing test beds — both on the East Coast — are already heavily used, or needed to run the official model.

“To get their work accepted into the code base for the national model, our scientists need to prove that the model will run with their changes in a stable way, and that the changes improve the accuracy of the model’s outlook,” says Scott Nolin, the SSEC’s technical lead on S4 design. “You can’t just go into the operational system to try things out.”

Nolin and SSEC colleagues worked with Dell’s High Performance Computing group to optimize the final system design.

Ninety percent of the computing time is dedicated to NOAA researchers in Madison and elsewhere, with the remaining time available for other computing-intensive SSEC research.

The S4 system incorporates 3,072 computer processor cores — larger by half than any previous computer on campus — eight terabytes of memory and 450 terabytes of disk space, all connected by a 40 gigabit-per-second network.

Operating since June, S4 has run thousands of hours of computing tasks — much of it shake-down work to get the purpose-built system up to speed.

Building the computer in Madison gives scientists direct access to the people who designed and operate it, no small thing for such complicated modeling. That NOAA chose SSEC is a testament to SSEC and UW–Madison’s reputation for collaborative science and research computing, according to Nolin.

“We get an opportunity like this because we have experience making it work well,” he says. “It’s not for everyone on campus to use, but it’s here because of the way everyone on campus operates.”