GOVERNMENT

SEATTLE--Global supercomputer leader Cray Inc. today reported that its new Cray X1 supercomputer demonstrated superior speed and efficiency on the industry-standard Linpack test. The results were recently published in the Linpack report maintained by Professor Jack Dongarra, University of Tennessee. On Cray X1 systems varying from 4 to 60 processors, the per-processor speed averaged 11.55 billion calculations per second (gigaflops), 2.5 to 10.6 times faster than the published per-processor speeds of HPC products from other U.S. vendors. In addition to its absolute speed advantage, the Cray X1 system demonstrated efficiency (actual speed as a percentage of theoretical "peak" speed) averaging 90 percent, substantially better than HPC systems from major U.S. competitors. Cray officials said Linpack is a processor-centric benchmark test and was not designed to measure other system attributes that can contribute significantly to real-world performance, including internal memory speed, disks, and external networks. "Early results indicate that on demanding real-world applications, the Cray X1 system's performance advantage typically is much larger than on the less-challenging Linpack benchmark," said Cray Chairman and CEO Jim Rottsolk. "These initial Linpack results confirm the strong capability of the Cray X1 processors, however." "The Cray X1 system has demonstrated its ability to perform on Army applications in computational fluid dynamics, computational mechanics and battlefield weather forecasting that are of great importance to the defense of the United States," said Paul Muzio, NCSI VP-Government Programs and AHPCRC Support Infrastructure Director. "The Cray X1 is an excellent high-capability system that provides impressive sustained performance for large, complex applications." The semi-annual ranking of the world's "Top500 Supercomputer Sites" is based on Linpack test results. Cray said it expects to have a growing presence at the upper end of this list as larger versions of the Cray X1 product are installed during 2003 and 2004. The "Top500 Supercomputing Sites" list is compiled and published by Jack Dongarra, Hans Meuer and Erich Strohmaier from the universities of Tennessee and Mannheim (Germany), and NERSC/Lawrence Berkeley National Laboratory.