|
The Cray-built Red Storm computer uses a high-speed 3D torus interconnect and 64-bit AMD Opteron processors. Its high-speed I/O and advanced MPP operating system scale applications from 200 to more than 30,000 processors without performance losses from communications bottlenecks, asynchronous processing, or memory access delays. |
|
Images show winds aloft predicted by the Red Storm computer. Sandia researchers have also run a Parallel Ocean Program from Los Alamos National Laboratory on 10,000 processors of the Red Storm to produce a 10-year simulation of global ocean circulation, one element in climate modeling. The high-resolution simulation uses average grid spacing of 10 km (0.1°) for 345 million grid points. |
"The upgrade to the Red Storm computer lets us carry out unprecedented simulations," says William Camp, director of computation, computers, information and mathematics at Sandia. "For example, we can resolve climate calculations to within 10 km or about 0.1° of Earth's circumference, which is beyond the current state of the art. We can carry out larger simulations that are faster than those of previous systems by an order of magnitude. The added capacity will let us undertake new and more complex research."
The upgraded Red Storm supercomputer will contain 14,348 AMD Opteron processors, more than 72 terabytes/sec of system memory bandwidth, more than 125 terabytes/sec of sustained bandwidth, and 400 terabytes of disk storage.
Several challenging problems are already running at large scale on Sandia's Red Storm supercomputer. For example, the Sandia-NCAR Spectral Element Atmospheric Model runs on 10,000 processors of Red Storm for 36 hr to simulate 20 days of global weather. The billion-gridpoint simulation of the entire Earth uses an average grid spacing of 13 km.
Sandia and Cray designed the Red Storm computer as part of a $90 million contract under the DOE's Advanced Simulation & Computing program. The computer design became the basis for the Cray XT3 massively parallel supercomputer.