DOE calls for upgrade to Los Alamos super

The Energy Department has launched a $65 million upgrade of a supercomputer that is used to monitor the nation's nuclear stockpile, and the agency was expected to begin installation last week of a new machine that is capable of 3 trillion floating point operations per second at Los Alamos National

The Energy Department has launched a $65 million upgrade of a supercomputer that is used to monitor the nation's nuclear stockpile, and the agency was expected to begin installation last week of a new machine that is capable of 3 trillion floating point operations per second at Los Alamos National Laboratory.

Once built, the computer, a cluster of Origin 2000 systems from Silicon Graphics Inc./Cray Research, will be the fastest in the world— capable of performing simulations of nuclear weapons tests in three dimensions. "I really think it's going to be the beginning of a revolution," said Andy White, a Los Alamos project director who heads a program on predictive modeling and simulation. "We are going to be able to do some problems we have not been able to touch before."

A companion machine, which eventually will be equal in speed to the one at Los Alamos, is being assembled by IBM Corp. at Lawrence Livermore National Laboratory. It is not expected to be completed until next year. Los Alamos plans to have the Cray system running in December.

The systems are part of the Accelerated Strategic Computing Initiative (ASCI), through which DOE is paying high-performance computer manufacturers to develop more powerful systems that are made from commercially available components before the broader marketplace starts to demand them. They are being used to simulate underground nuclear tests, which are outlawed under the international Comprehensive Test Ban Treaty.

White said he could not discuss the nuclear weapons applications that DOE will run on the new system, but he said the technology is necessary to add "physics effects" that make computer models more true to life. He said the system also could be used to create more accurate global climate models or to predict the patterns of wildfires.

Mark Goldman, director of business planning at SGI, said the 6,144-processor system, which is made up of 48 computers, has been designed to reduce communications bottlenecks that typically slow down large supercomputers. In addition to providing 1.5 terabytes of memory— three times that of DOE's largest existing machine— ASCI "Blue Mountain" uses new 250 MHz processors, new hub and router technology as well as improved caching algorithms, he said.

Gary Smaby, president of market research firm Smaby Group, said the system is made from the same "high-performance building blocks" that other vendors of superfast systems use. "They're all using the same off-the-shelf parts," he said.

However, he added, neither SGI nor any other firm is likely to sell many systems as large as those for ASCI, although agencies doing scientific research could have applications for them if they had the money to buy them.