The improved system will create more complex models

Aug 25, 2009 20:31 GMT  ·  By
Goddard Space Flight Center recently added 4,128 processors to its Discover high-end computing system, with another 4,128 processors to follow this fall
   Goddard Space Flight Center recently added 4,128 processors to its Discover high-end computing system, with another 4,128 processors to follow this fall

This August, the American space agency made available the first computing hours at its high-end computing system for climate analysis, located at the Goddard Space Flight Center, in Greenbelt, Maryland. The instrument is the centerpiece of NASA's new climate-simulation capabilities, which will contribute with the agency's own reports to the overall assessment on the condition of our environment, requested by the United Nations Intergovernmental Panel on Climate Change (IPCC) and other national and international climate initiatives.

The Discover high-end computing system now has an extra 4,128 Intel Xeon 5500 series processors, which are based on the Nehalem architecture introduced in spring 2009. They were all placed in the IBM iDataPlex “scalable unit,” for a maximum efficiency. The improved supercomputer will run simulations for the IPCC's Fifth Assessment Report by the Goddard Institute for Space Studies (GISS), in New York City, as well as for the Goddard Global Modeling and Assimilation Office (GMAO).

“We are the first high-end computing site in the United States to install Nehalem processors dedicated to climate research. This new computing system represents a dramatic step forward in performance for climate simulations,” Goddard Computational and Information Sciences and Technology Office (CISTO) Chief Phil Webster says. He adds that preliminary simulations showed that the new instrument had about twice as much computing power per processor than any other recognized, high-end installation.

“Nehalem architecture is especially well-suited to climate studies. Speed is an inherent advantage for solving complex problems, but climate models need large memory and fast access. We configured our Nehalem system to have 3 gigabytes of memory per processor, among the highest available today, and memory access is three to four times faster than Discover's previous-generation processors,” CISTO's lead architect, Dan Duffy, adds.

“Once the model goes below 10-kilometer resolution, features such as well-defined hurricane eyewalls and convective cloud clusters appear for the first time. At these cloud-permitting resolutions, the differences are stunning,” Goddard's Software Integration and Visualization Office (SIVO) Advanced Software Technology Group Acting Lead William Putman concludes.