GOVERNMENT

Cray Inc reported that recent High Performance Computing (HPC) Challenge benchmark tests show that the Cray XD1 Opteron/Linux-based supercomputer outperforms systems of similar size and price. The Cray XD1 system, which is being demonstrated at the LinuxWorld Conference & Expo in Boston this week, scored first in one of the eight HPC Challenge performance categories and second in two others, resulting in the highest rating overall in its class and some of the best scores among all the systems tested regardless of size and price. "These benchmarks, based on application-oriented rather than theoretical computing models, demonstrate that the Cray XD1 supercomputer provides impressive HPC performance that is usually found only in larger and more expensive systems," said Peter Ungaro, Cray senior vice president for sales, marketing and service. "To suit the economics of the HPC mid-market segment, the system delivers this performance at price points starting at under US$100,000 list." As of February 11, 2005 the Cray XD1 supercomputer surpassed all the other HPC systems tested in random-ring latency, performing more than 3 times faster than the leading Itanium system on this important indicator of real-world application performance that measures latency and bandwidth in small-message codes. In the global parallel matrix transpose (PTRANS) benchmark test, the Cray XD1 system demonstrated a better than fivefold performance increase over Xeon/Infiniband clusters and a 40 percent increase over the SGI Altix system, which was configured with twice as many processors. In the global Fast Fourier Transform (FFT) test, the Cray XD1 system exceeded the performance of a comparably sized IBM p655 with Federation switch by a factor of 8.4, and bettered the SGI Altix by 20 percent. The company's high-end system, the Cray X1 supercomputer, was the top performer on two important HPC Challenge tests: the G-TRANS benchmark, which measures simultaneous communications between processors while attacking a realistic problem; and the G-HPL test, which assesses the efficiency of the Message Passing Interface (MPI). Specifically designed to meet high-end computational needs, the Cray X1 system provides exceptional memory bandwidth, low-latency interconnections and vector-processing capabilities. Complete HPC Challenge results for all listed systems are available at http://icl.cs.utk.edu/hpcc/hpcc_results.cgi. (The Cray XD1 system is identified as "Cray AMD Opteron" in the table.) Going Beyond Traditional Linpack Benchmarks Developed by Jack Dongarra and Piotr Luszczek of the University of Tennessee, with collaborators from the U.S. and Europe, the HPC Challenge benchmark suite debuted in 2003 under the sponsorship of the U.S. Department of Energy, the National Science Foundation and DARPA. The tests are designed to assess those aspects of supercomputing that have the greatest effect on performance under real-world conditions. The HPC Challenge suite includes Linpack, a single test of processor performance that is the basis for the semi-annual TOP500 supercomputer ranking. The suite substantially augments the Linpack measurement with seven more tests. Additional test are expected to be added in the future. "Linpack is useful, but no single test can accurately reflect the overall performance of HPC systems," according to Dongarra. "The HPC Challenge benchmark suite stresses not only the processors, but the memory system and the interconnect. It is a better indicator of how an HPC system will perform across a spectrum of real-world applications." "Because the HPC Challenge benchmark tests are multifaceted, they provide a more comprehensive insight into the performance of today's high-performance computing systems and are a major improvement over traditional single-point benchmarks in predicting how HPC systems will perform on many applications," said Paul Muzio, director of support infrastructure at the U.S. Army's High Performance Computing Research Center (AHPCRC).