Two Cray supercomputer products that masterfully leverage the AMD Opteron processor and HyperTransport technology, the Cray XT3 and Cray XD1 systems, have posted leading overall results on the HPC Challenge benchmark tests, AMD and Cray reported today.
HPC Challenge results are gaining importance as customers increasingly use them to help decide which high-performance computers to buy. For example, CSCS, the Swiss National Supercomputing Center, relied heavily on HPC Challenge results when it recently selected a Cray XT3 system with 1,100 AMD Opteron processors that will be one of Europe's most powerful supercomputers.
According to CSCS Director Marie-Christine Sawley, "We chose the HPC Challenge benchmark suite for our recent 'Horizon' procurement because we can measure and analyze the characteristics of a given supercomputer architecture with it. It lowers the burden on the bidders, speeds up your procurement project, and still allows you to gauge the effects of a given architecture on your key user applications by mapping their characteristic requirements onto the individual HPCC benchmark results."
In comparing customer-reported HPC Challenge results for three large-scale systems of about the same size, an 1,100-processor Cray XT3 supercomputer had the best scores on seven of the 10 "condensed results" tests, compared to an SGI Altix 3700 system with 1,008 processors and an IBM Blue Gene system with 1,024 processors. In the seven tests, the Cray XT3 typically outperformed the next-best system by a factor of two to five times, and was up to 17 times faster than the third-ranking system.
Among 128-processor scalar systems, the Cray XD1 supercomputer demonstrated leading results on four tests, more than any other microprocessor-based system. The Cray XD1 system did especially well in the random ring latency and global FFT tests.
In addition, a Cray X1E vector supercomputer with 248 multistreaming processors was more than 10 times faster than the nearest competitor on the important global random access test, measuring random updates of memory.
Results cited are those posted as of June 15, 2005 on the HPC Challenge website: http://icl.cs.utk.edu/hpcc/.
"When systems do well across the board on the HPC Challenge benchmark tests, as Cray supercomputers do, it's clear that they were purpose-built for high-performance computing," said Steve Scott, Cray chief technology officer. "Many HPC systems today were designed for other markets and do well on only one or two HPC Challenge tests. AMD Opteron processors and HyperTransport technology provide a powerful foundation for the direct connect, balanced system architectures of our Cray XT3 and Cray XD1 supercomputers."
"Cray supercomputers based on AMD Opteron processors with Direct Connect Architecture give customers remarkable performance for their money on real-world problems," said Rich Oehler, Corporate Fellow at AMD. "AMD64 technology is being designed into many of the world's most powerful computers, including the world's largest AMD Opteron processor-based system, 'Red Storm,' a Cray supercomputer located at Sandia National Laboratories, that when analyzed on real-world problems and applications, has no match for providing architectural balance in a high-performance system."
About the HPC Challenge Benchmark Tests
Assembled by Jack Dongarra and Piotr Luszczek of the University of Tennessee, with collaborators from the U.S. and Europe, the HPC Challenge benchmark suite tests multiple capabilities that can make a major difference in the real-world performance of HPC systems. The test suite includes High Performance Linpack, a single test primarily of processor performance that is the basis for the semi-annual TOP500 supercomputer ranking, and substantially augments this with six additional tests. More tests may be added over time.
"Linpack is useful, but no single test can accurately reflect the overall performance of HPC systems," Dongarra said. "The HPC Challenge benchmark test 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."
The new set of tests, co-sponsored by the National Science Foundation, U.S. Department of Energy and DARPA HPCS (High Productivity Computing Systems) program, was introduced at the SC2003 annual supercomputing conference in November 2003.