The Livermore, Calif.-based supercomputing cluster will consist of 962 nodes running on 1,920 of Intel's 2.4GHz Xeon processors, with a theoretical peak of 9.2 teraflops (trillions of calculations per second). Each node will have 4GB of DDR SDRAM memory and 120GB of hard-disk space.
The cluster, to be completed by Linux NetworX this fall, would be the fastest Linux or Intel-based supercomputer, and one of the world's five fastest supercomputers.
The Evolocity cluster, as it will be known, will give Linux a significant boost up the charts of supercomputing prowess, if the claims are true. In last month's Top500 ranking of supercomputers, the fastest Linux cluster ranked at only the 35th spot--a 512-node "Beowulf" cluster at the University of Heidelberg in Germany.
It would also be a significant demonstration of the use of commonly available technology, such as the mass-market Intel hardware and freely available Linux software, in high-performance computing. Many supercomputers use proprietary hardware and operating systems such as Unix.
Linux is popular with scientists, because it is cheap and stable and because its license allows them to freely modify the software code.
"This Intel-based Linux NetworX system is historic in that it represents a viable method of using standards-based technologies to create some of the fastest supercomputers in the world," Lisa Hambrick, director of enterprise processor marketing for Intel, said in a statement.
The fastest supercomputer is currently NEC's Earth Simulator, which35.9 trillion calculations per second, more than the next 12 systems combined, according to Top500.
Linux NetworX said that the cluster will be seven times more powerful than Deep Blue, the machine used to beat world chess champion Garry Kasparov in 1997. It will have the same amount of processing power as 9,200 average desktop PCs, the company said.
The company, whose customers include the U.S. National Security Agency, Sandia National Laboratories and Los Alamos National Laboratory, is using custom software for the nodes' BIOS and for cluster management.
Big iron for e-science
In other supercomputing news, IBM plans to build a supercomputer for the Engineering and Physical Sciences Research Council (EPSRC) in England. The supercomputer will be used to support the country's e-science initiative, which will focus on "large-scale science" and research by harnessing massive computing power and the Internet, according to the U.K. Research Council's Web site.
IBM's machine will be made up of a cluster of its P690 servers. Each P690 can be configured with up to 32 Power4 chips, each with two Power4 processor cores. The supercomputer will run IBM's AIX operating system.
The supercomputer will initially be able to process information at a speed of 6.72 teraflops, IBM said. This performance will be upgraded to more than 11 teraflops in 2004 and more than 22 teraflops in 2006. Such speeds, if achieved, would rank the supercomputer among the fastest in the world by today's standards.
ZDNet U.K.'s Matthew Broersma and Matt Loney reported from London.