DARPA aims to increase productivity in high-end computing

DARPA aims to increase productivity in high-end computing

Under a new program led by the Defense Advanced Research Projects Agency, five vendors are competing for a chance to design a more productive supercomputer for the year 2010.

The High Productivity Computing Systems (HPCS) program seeks new ideas for easier-to-use high-performance computers that will meet the needs of the defense and intelligence communities, said Robert B. Graybill, HPCS program manager in DARPA's Information Processing Technology Office.

Today's massively parallel high-performance computers have their architectural roots in the technology of the late 1980s, Graybill said in a panel discussion yesterday at the SC2002 conference in Baltimore. HPCS seeks to bridge the gap between that technology and the promise of quantum computing'the faster performance of multiple computations simultaneously.

Graybill said DARPA also seeks to double the value of high-end computers every 18 months'though value and productivity could have many meanings to the user community.

Supercomputing specialists need to develop a way to measure the capability of their machines other than calculating theoretical performance from processor clock speeds, Graybill said.

"We may want to rethink how we populate the Top 500," Graybill said, referring to a semiannual list of the world's fastest computers [see story at www.gcn.com/vol1_no1/daily-updates/20531-1.html].

Representatives of the five HPCS participants, Cray Inc. of Seattle, Hewlett-Packard Co., IBM Corp., SGI and Sun Microsystems Inc., largely agreed that supercomputers should:

  • become easier to program

  • improve compiler performance

  • increase bandwidth to reduce memory and I/O bottlenecks

  • become more robust

  • decrease the so-called idea-to-solution time.


  • In this first phase of HPCS, competitors are being asked to develop a concept study on how the ideal high-performance computer of 2010 should be built, Graybill said. As many as three vendors will be asked to provide more detailed studies in the second phase of the competition, and one or two vendors will be chosen to do a detailed engineering design in Phase 3.

    Graybill declined to say how much money DARPA is investing in the program.

    inside gcn

    • machine learning

      Mitigating the risks of military AI

    Reader Comments

    Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

    Please type the letters/numbers you see above

    More from 1105 Public Sector Media Group