Nuke systems get $2.5b in race against time

The Clinton administration considers numerical simulation so important to nuclear
safety and reliability that it plans to spend nearly $2.5 billion on development through
2004.


Energy Department managers in charge of the Accelerated Strategic Computing Initiative
see it as a race against time.


The scientists who designed the nation's nuclear weapons are aging right along with the
weapons, said Gilbert Weigand, deputy assistant secretary for strategic computing and
simulation.


The department wants a computer capable of 100 trillion floating-point operations per
second developed by 2004, when half the weapons in the U.S. nuclear stockpile will have
aged beyond their design lifetimes. Scientists don't know what to expect when that
happens, Weigand said.


Also by 2004, Energy will have lost half the institutional expertise in its research
labs, based on the number of scientists who retire and the number of nuclear weapons tests
each scientist conducted in Nevada before 1993. That was the year the United States
stopped building and testing nuclear weapons.


The retiring scientists are the only researchers who understand the weapons systems
well enough to validate simulation tools and introduce them into the culture, Weigand
said.


The simulation project calls for the 100-teraFLOPS machine 10 years in advance of
industry plans to develop such a powerful computer, Weigand said.


Energy needs that much processing power to create 3-D simulations from its weapons
test, physics and engineering data, spending no more than one or two weeks per simulation.


The three Energy-funded research laboratories involved in the accelerated computing
initiative are Lawrence Livermore National Laboratory in California and the Los Alamos and
Sandia National Laboratories in New Mexico.


Since 1996, Energy has awarded contracts to three computer companies to build teraFLOPS
computers: IBM Corp., Intel Corp. and Cray Research Inc., a subsidiary of Silicon Graphics
Inc.


Energy has since awarded additional contracts to IBM, Silicon Graphics, Sun
Microsystems Inc. and Digital Equipment Corp., now part of Compaq Computer Corp., to
accelerate development of hardware switches and software for teraFLOPS computers.


"The computers are on schedule and on budget," Weigand said.


In February, the department awarded another $85 million contract to IBM for a
10-teraFLOPS computer for delivery to Lawrence Livermore in the first quarter of 2000. IBM
also is building a 3-teraFLOPS machine for the simulation project; the 10-teraFLOPS
machine will be an upgrade, Weigand said.


IBM in March delivered 250 RS/6000 Scalable Processor nodes to Lawrence Livermore. They
are fast symmetrical-multiprocessing nodes based on the 332-MHz 604E PowerPC processor and
PCI bus.


"We've worked with Energy to develop nodes appropriate for the simulations it
wants to run," said David Gelardi, program director of business intelligence
solutions for IBM's RS/6000 division.


A 332-MHz PowerPC balances integer and floating-point performance, Gelardi said. The
nodes are packaged in wide and thin footprints so that more processors can fit into the
CPU drawers of the SP frame than was possible for previous-generation nodes, he said.


Energy's accelerated computing schedule initially required a 1.8-teraFLOPS computer,
which Intel delivered in December 1996. This December, Weigand said, DOE will receive
3-teraFLOPS computers from IBM and Silicon Graphics.


inside gcn

  • data science (chombosan/Shutterstock.com)

    4 steps to excellence in data analysis

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