AS fast as computer technology is advancing, the Department of Energy's Accelerated Strategic Computing Initiative (ASCI) must advance the technology even faster.
The reason for the push lies in the inexorable aging of the country's nuclear stockpile and DOE's responsibility to keep this stockpile viable. On September 25, 1995, President Clinton directed DOE to undertake the necessary activities to ensure continued stockpile performance in an era of no nuclear testing, no new weapon development, a production complex with reduced capacity and capability, and an aging stockpile of fewer weapons and fewer types of weapons.
The Stockpile Stewardship Program-of which ASCI represents one key component-is DOE's response to this challenge. It must provide the tools researchers need to develop a detailed understanding of the science and technology that govern all aspects of nuclear weapons. It must also proceed quickly so that the necessary tools and scientific understanding are in place within about a decade.
In this race against time, three national laboratories-Lawrence Livermore, Los Alamos, and Sandia-have teamed up with the supercomputing industry to accelerate the development of high-performance supercomputers. Just as fighter planes regularly break the speed-of-sound barrier, ASCI supercomputers are breaking speed barriers of a different sort set by Moore's law. That is, they are doubling computing speeds in terms of teraops (trillions of floating-point operations per second) faster than every 18 months.
The current high-end computer at Livermore is the ASCI Blue Pacific machine built by IBM and delivered in the fall of 1998. It was used to perform the first-ever three-dimensional simulation of an exploding nuclear weapon primary. This calculation, completed in November 1999, represented the first successful completion of an ASCI milepost application. In addition, this machine has performed a series of first-principles simulations detailing the molecular interactions of the highly corrosive compound hydrogen fluoride, which occur in some high explosives.
The newest candidate for this innovative lineup is ASCI White, scheduled for delivery to Lawrence Livermore this summer. In terms of pure speed, ASCI White will be at least two and a half times faster than the Blue Pacific machine, which is itself an impressive system. One of the big triumphs of ASCI Blue Pacific-which will become routine with White-is performing detailed three-dimensional simulations of complex physical phenomena. Clocking in at more than 12 teraops, White is the next step in the ASCI plan to produce a 100-teraops system by 2004. One hundred teraops is the entry-level performance needed to perform full-scale simulations of exploding nuclear weapons. The results from these incredibly complex three-dimensional simulations will be combined with existing nuclear test data and new, nonnuclear experiments to ensure the safety, reliability, and performance of U.S. nuclear weapons.
The national weapons laboratories and the U.S. high-performance computer industry are not the only entities engaged in the ASCI challenge. Through DOE's Academic Strategic Alliance Program, the U.S. academic community also draws on the power and speed of these supercomputer systems to advance unclassified science-based modeling and simulation technologies applicable to all of ASCI's research areas.
At Livermore, ASCI efforts engage talents across the Laboratory's organizations, with the Defense and Nuclear Technologies and the Computation directorates providing the focus and the leadership. This issue's feature article, beginning on p. 4, highlights the accomplishments of ASCI Blue Pacific and the promise of ASCI White. The progress in computing inherent in these machines has placed researchers in the DOE community and in academia on the verge of being able to simulate first-principle physics without resorting to oversimplified models. It's an exciting prospect, one that promises breakthroughs not only for stockpile stewardship but also for areas as diverse as biochemistry, materials science, and astrophysics.


Back to June 2000