Scientific Computing Seminar

Date:
Friday, February 6, 2004
Time:
1:00pm-2:00pm
Location:
50A-5132
Seminar Speaker:
L. Oliker, J. Carter, and A. Canning
CRD and NERSC
LBNL
Title:
Trip report from the Earth Simulator Center
Abstract:
The rapidly increasing peak performance and generality of superscalar cache-based microprocessors long led researchers to believe that vector architectures hold little promise for future large-scale computing systems. Due to their cost effectiveness, an ever-growing fraction of today’s supercomputers employ commodity superscalar processors, arranged as systems of interconnected SMP nodes. However, the growing gap between sustained and peak performance for scientific applications on such platforms has become well known in high performance computing.

The recent development of parallel vector systems offers the potential to bridge this performance gap for a significant number of scientific codes, and to increase computational power substantially. This was dramatically highlighted by the recent availability of the Japanese Earth Simulator (ES) , the world's most powerful supercomputer both in terms of peak (40.96 TFlops/s) and sustained LINPACK (87.5% of peak) performance. More important than the raw performance, is what this new capability entails for the scientific community that relies on computational modeling and simulation. It is therefore critical to evaluate these architectural approaches in the context of demanding computational algorithms of importance to the Office of Science.

Thanks to a recent cooperative agreement between NERSC and ES, we had the opportunity to visit the center in December 2003 to conduct a preliminary performance evaluation. This talk will present our experience on the ES system and include benchmarking results in the application areas of astrophysics, energy fusion, and material science.

Sponsor of Seminar:
Scientific Computing

Contact Esmond G. Ng EGNg@lbl.gov