The year 2002 has brought very significant changes in high performance
computing whose impact will be felt for years to come. At NERSC
the biggest visible change was the decision to upgrade our current
Seaborg platform from 5 to 10 teraflop/s peak performance. This
will create one of the largest systems in the U.S. dedicated
to basic computational science. It will give NERSC users routine
access to an unprecedented 6,556 processors, coupled with one
of the largest memory systems anywhere. DOE-supported computational
scientists again will have access to one of best possible resources
to further the DOE mission in basic sciences.
It is clear to me that resources of this scale are required
if simulations are to make the scientific impact that we all
expect. In 2002 NERSC dedicated a large share of its resources
to a small number of projects that require the full NERSC
capability or special services. This strategy has paid off
with a number of scientific accomplishments that are discussed
in this report.
One of the signposts of change was the arrival of the Earth
Simulator system in Japan. After only a few months of operation,
this system has already made a profound impact on computational
science and high performance computing in the U.S. This system
broadened our horizon and gave us a glimpse of what kinds
of simulations are possible at sustained speeds in the range
of tens of teraflop/s. During the summer and fall of 2002,
our computational scientists held a series of town hall meetings
using the Access Grid and defined their needs for future simulation
capabilities. The Earth Simulator made us suddenly “think
big” again in supercomputing, and I am thrilled that
the DOE computational science community, in particular our
NERSC users, are ready to lead the way to the next level of
computational science.
The Earth Simulator also challenged our dependence in the
U.S. on hardware platforms that are designed primarily for
commercial applications. It exposed the “divergence
problem,” the fact that scientific applications obtain
only a small fraction of the peak performance of current platforms,
and that this ratio seems to be getting even smaller. One
of the highlights of 2002 at NERSC was a workshop with IBM
and Argonne National Laboratory that produced a proposal on
how to change this situation. In particular the workshop defined
the “Blue Planet” system, and more fundamentally
it was able to produce a whole new strategy for creating science-driven
computer architecture that may change the future of high performance
computing in the U.S.
This workshop also brought out what is for me the best of
NERSC: a great team of dedicated staff who within a few weeks
are able to rise to the challenge and produce a new way of
thinking about a problem. My special thanks and congratulations,
as always, go to the NERSC staff for their skill, dedication,
and tireless efforts to make NERSC the best scientific computing
resource in the world.
Horst D. Simon
NERSC Center Division Director
|