DOD signs on for CPU-GPU hybrid supercomputer
Appro machines will have clusters at five HPC centers
- By Kevin McCaney
- Dec 08, 2010
As graphics processing units continue to catch on as the power behind supercomputers, organizations also are finding use for CPU-GPU hybrid systems.
The Defense Department this week hired Appro International to build an Appro 1U-Tetra supercomputer, with clusters in five of DOD’s supercomputing centers, Timothy Prickett Morgan writes in the Register. Appro is a subcontractor on the contract to Lockheed Martin, which will operate the machines and applications. John Lee, of vice president of advanced technology solutions at Appro, told the Register that Appro’s part of the deal is worth about $7 million.
About a third of the nodes on each cluster will have Nvidia M2050 GPU co-processors, accounting for about 57 per cent of the machines’ performance, Morgan writes.
The Nvidia GPUs are the same graphics processors Amazon is using in its Elastic Compute Cloud, through which it is offering supercomputing processing as part of its cloud computing service. The chips also are used in China’s Tianhe-1A, which recently claimed the title of world’s fastest supercomputer.
Supercomputing as a commodity, available on Amazon
There’s a new supercomputing champ in town
The graphic processing units provide better overall performance and faster simulation, according to a release from the company. The entire system will have an aggregate peak of 147.5 teraflops, with 83.4 teraflops powered by the GPUs.
Purchased as part of DOD's High Performance Computing Modernization Program, the systems will be located at the Army Research Laboratory at Aberdeen Providing Ground, Md.; Air Force Research Laboratory at Wright Patterson Air Fo4rce Base, Ohio; Army Engineer Research and Development Center in Vicksburg in Miss.; the Navy Supercomputing Resource Center at Stennis Space Center in Mississippi; and the Arctic Region Supercomputing Center in Alaska.
Kevin McCaney is a former editor of Defense Systems and GCN.