Supercomputer Exceeds Previous Performance Numbers on Graph 500 Benchmark Test by 3.5 Times
MILPITAS, CA -- (Marketwired) -- SGI (NASDAQ: SGI), the trusted leader in high performance computing, announced today that the SGI® UV? 2000 became the first single node SMP to exceed 100 GTEPS, as verified by the Graph 500 benchmark test for measuring large data volume processing by supercomputers. In addition to this new performance world record, the Green Graph 500 test also revealed the SGI UV 2000 to be the commercial supercomputer with the lowest power-consumption figures.
The Kyushu University Institute of Mathematics for Industry (IMI), in conjunction with the Institute of Statistical Mathematics (ISM) and SGI Japan, Ltd., have conducted the Graph 500 (*1) benchmark test on the SGI UV 2000, which is officially known as the Data Assimilation Supercomputer System at ISM. It recorded a figure of 131.427 GTEPS, becoming the first single-node SMP system to exceed 100 GTEPS. The test measures the performance of supercomputers in processing large amounts of data. The Green Graph 500 (*2) benchmark test showed that it was also the commercial supercomputer with the lowest power consumption performance of 12.481 MTEPS/W.
A research team at the IMI Laboratory of Advanced Software in Mathematics was responsible for running the dual benchmark tests on one rack of SGI UV 2000, which had 640 cores per node.
The exceptional performance of the SGI UV 2000 was primarily due to its inherent capability of providing high-frequency memory access and wide memory spaces required by the Graph 500 benchmark.
"We are delighted to have one of our most innovative HPC solutions achieve these kinds of results on such a widely respected industry test," said Jorge Titinger, president and CEO of SGI. "And by doing so in such an energy-efficient manner, we have proven it is possible to accomplish high-performance computations reliably and stably at a lower power consumption. It truly is possible to have one single supercomputer provide high-density, industry-leading performance, and do so within a stringent power envelope."
"Attention is increasingly being turned to the use of new supercomputers in big data computing, such as processing massive amounts of graph data," said Katsuki Fujisawa, professor, Institute of Mathematics for Industry, Kyushu University. "This requires large-scale memory utilization and high-speed parallel processing, making supercomputers more important than the cloud for big data processing. The software developed at the Kyushu University Institute of Mathematics for Industry has shown even greater than expected processing performance and energy savings when processing massive graph data on the SGI UV 2000. Other major advantages of the SGI UV 2000 are the ability to migrate existing software to it, and the ability to apply past experience and expertise."
"The Institute of Statistical Mathematics is a leading inter-university research institute engaged in data centric science research in the big data age," said Yoshiyasu Tamura, vice director-general, The Institute of Statistical Mathematics. "We have chosen the SGI UV to be our 'extreme' supercomputer, which is providing the institute's researchers access to the world's largest coherent shared memory of 64 TB optimum performance in analyzing large amounts of data on graph theory and is essential to our research. This system is a critical factor in our university's output and is used in HPCI work and through the ISM Cooperative Research Program arranged by our institute."
SGI is a global leader in high performance solutions for compute, data analytics and data management that enable customers to accelerate time to discovery, innovation, and profitability. Visit sgi.com (sgi.com/) for more information.
© 2014 Silicon Graphics International Corp. All rights reserved. SGI, the SGI logo and SGI UV are trademarks or registered trademarks of Silicon Graphics International Corp. or its subsidiaries in the United States and/or other countries. All other product and service names mentioned are the trademarks of their respective companies.
(*1) The Graph 500 is a benchmark for testing graph search performance of supercomputers and other computers. The test uses the number of edges found in a 1-second breadth first search of a graph as an index. TEPS (Traversed Edges Per Second) is the unit used by this index. A high TEPS value indicates faster graph search capability. A GTEPS (Giga TEPS) value of 131.427 indicates capability to search approx. 131.4 billion edges in a graph per second. Visit the Graph 500 website at: http://www.graph500.org/
(*2) The Green Graph 500 is a benchmark for testing graph search performance similar to the Graph 500 that uses TEPS/W (Traversed Edges Per Second/W) as an index unit. A high TEPS/W value indicates a high graph search capability per unit of power consumption, or in other words low power consumption. A 12.418 MTEPS/W (Mega TEPS/Watt) value indicates that the computer can search 12.41 million edges in a graph using just one watt of power. Learn more at the Green Graph 500 website: http://green.graph500.org/
Contact Information: Grayling Public Relations Jessie Adams-Shore (415) 593-1400 [email protected]