DARPA puts millions in U.S. vendors' pockets
Need for speed
DARPA has done its part to keep U.S. computer makers alive and kicking by handing out millions in grants to Cray, Sun Microsystems and IBM.
As part of its High Productivity Computing Systems program, DARPA has kicked down more than $100 million to vendors. The government agency hopes the server makers can come up with some high performance systems by 2009 that will replace today's equipment and serve as stopgaps before quantum computers arrive.
Cray and its New Technology Endeavors subsidiary received $43.1 million to continue with their Cascade computing concept. The idea is to make a peta-scale computer that pays special attention to memory hierarchy, processor-in-memory technology and high memory bandwidth. On the software side, the two companies hope to create code tuned for shared and distributed memory programming models.
Sun, not usually on DARPA's short list, will pull in $49.7 million to further work on the Hero project. This effort seems to play into many of Sun's strengths as its focuses on improving programming tools, boosting system security and supporting legacy software. Sun is to create a system that takes all of these factors into account and that can crank through quadrillions of calculations per second. It's meant to be a a programmers' dream.
IBM, DARPA's darling, will see $53 million pop into its coffers to advance PERCS (Productive, Easy-to-use, Reliable Computing Systems). IBM's main mission is to design systems that have high performance on both scientific and commercial software workloads.
The U.S. government bowed its head in shame when NEC stole the top supercomputer crown with its Earth Simulator System. NEC is poised to hold its top spot for some time, but companies such as IBM are working hard to return prestige to U.S. high performance computing.
Speedy machines are just one facet of the computing efforts being pushed at DARPA. The organization also pays close attention to developing hulking servers that consume as little power as possible. ®
Sponsored: Beyond the Data Frontier