Intel Corp. said Monday that it has developed for the Energy Department the world's fastest supercomputer, one able to perform 1 trillion operations a second.
That speed comes close to tripling the previous record in computing speed achieved by Hitachi Ltd. in 1995 with a supercomputer capable of doing 368 billion operations a second, Intel executives said.
The Intel supercomputer will be used at the government's Sandia National Laboratory to simulate the performance of U.S. nuclear weapons, replacing live tests of weapon stockpiles.
The supercomputer also will be used to predict weather changes and natural disasters and in other Energy Department projects that require a simulation of a series of events with massive mathematical calculations, Energy Department officials said.
"One trillion operations per second. . . . This was once thought unachievable," Energy Secretary Hazel O'Leary said during a conference call announcing Intel's fulfillment of its supercomputer contract.
"The U.S. can reestablish its dominance in this market as an innovator . . . in the business of ultra-computing," she said.
Santa Clara, Calif.-based Intel built the supercomputer using 9,624 standard desktop microprocessors, or Pentium Pro chips, and linking them to process data simultaneously in an architecture the industry calls "massive parallel processing."
Craig Barrett, Intel's executive vice president, said one of the breakthroughs achieved with this supercomputer is that it uses "standard building blocks."
"We were trying to demonstrate that with standard building blocks you can deliver world record-breaking power," he said.
However, Intel does not see much of a market for the supercomputers beyond a few government agencies, he said.
The Energy Department will pay Intel about $50 million for the machine under terms of the contract set in September 1995.
Barrett said he believes there is a reasonable market for intermediate-size parallel processing supercomputers using Pentium Pros. Intel said the supercomputers could be used for molecular modeling by pharmaceutical companies and manufacturing planning by auto makers.
Supercomputers traditionally have been built with processor chips designed for the purpose of supercomputing.