Industry-standard technology continues to change the face of high-performance computing, making supercomputers more affordable not only for research institutions and federal agencies, but also for enterprises.
The latest evidence of this trend was on display last week at the Supercomputing show in Seattle, where vendors launched a host of servers and compute clusters based on chips from Intel Corp. and Advanced Micro Devices Inc., and Sun Microsystems Inc. announced its largest supercomputing deal involving its AMD Opteron-based systems.
“You are seeing at this Supercomputing show Microsoft [Corp.] entering HPC [with the second beta for its Windows Compute Cluster Server 2003],” said John Zepper, manager of capacity and capability of computing at Sandia National Laboratories. “So you know its kind of evolving into a commodity field.”
The Albuquerque, N.M., laboratory in September launched “Thunderbird,” a cluster of 4,096 PowerEdge 1850 servers from Dell Inc. powered by 8,000 Intel Xeon chips. It also uses Ethernet switches from Force10 Networks Inc. and InfiniBand connectivity from Cisco Systems Inc. Thunderbird was listed as the fifth-fastest supercomputer in the world, and it was the most affordable, with the highest teraflops-per-dollar ratio, Zepper said. Its capable of almost 60 teraflops, or 60 trillion floating-point operations per second.
“The only way you can put together [a cluster such as Thunderbird] and not have it be way off the charts [in price] is because of industry-standard technology,” Zepper said. “The line between the enterprise IT side and the HPC side is all blurred now. If you can get a bunch of PowerEdge 1850s and a fast interconnect, hey, youve got a supercomputer. Its the volume you get with industry-standard [technology]. Thats where you get the advantage.”
Two-thirds of the systems on the Top 500 list use Intel chips, and an additional 55 run on AMD processors.
Suns deal with the Tokyo Institute of Technology will involve building a cluster of Sun Fire systems running 10,480 Opteron chips. The Santa Clara, Calif., companys cluster will be Japans fastest supercomputer, surpassing NEC Corp.s Earth Simulator, which until last year was the worlds fastest system.
At the same time, vendors continue to roll out new HPC systems based on industry-standard technology. Linux Networx, of Bluffdale,Utah, last week unveiled a family of Linux-based systems—the LS Series Supersystems—powered by Opterons.
San Francisco-based Penguin Computing Inc. is rolling out a series of preconfigured Linux clusters. The Application-Ready Cluster Portfolio comes in three product families, all with bundled hardware and software and featuring the companys Scyld Beowulf Linux clustering product. Dell, of Round Rock, Texas, is bulking up the services it offers to its HPC customers and over the next few months will start increasing the vertical-specific software offered in its HPC bundles. “The technology is proven,” said Reza Rooholamini, director of Dells Enterprise Solutions group. “Cluster-based [systems are] very viable.”