Another Linux supercomputer cluster with Advanced Micro Devices' Opteron chip is in the works, this time for Los Alamos National Laboratory's nuclear weapons testing program.
Linux Networx has been selected to build the cluster of 1,408 dual-processor Opteron servers, it announced Thursday. The system, known as Lightning, will deliver theoretical peak performance of 11.26 trillion FLOPS (floating-point operations per second), the company said.
The deal is worth $10 million, and Linux Networx expects to complete the system in September, said Dean Hutchings, chief operating officer of the Bluffdale, Utah, company. When completed, Lightning will rank slightly ahead of the MCR system the company built for Lawrence Livermore National Laboratories using Intel's Xeon processors and Linux, according to the Top 500 list of the world's fastest supercomputers (http://www.top500.org).
A number of Linux supercomputers have been announced over the past few weeks. Dell Inc. is working on what will become the fastest Linux supercomputer when it is finished later this year, with 17.7T FLOPS of peak theoretical performance. IBM and Fujitsu are also developing Linux supercomputers that will be deployed next year with TFLOPS results expected to be similar to that of the Linux Networx system.
IBM's cluster will use the Opteron processor, while Dell's will be built with the Xeon chip. Fujitsu has not yet announced the chip vendor that it chose for its system.
Last year, Cray announced plans for a Linux supercomputer with Opteron processors that will achieve 40T FLOPS of peak theoretical performance when Sandia National Laboratories receives the system late next year.
Clusters of low-end servers are popular for high-performance computing applications because of the price advantages over large supercomputers, especially when combined with Linux, said Gordon Haff, an analyst with Illuminata in Nashua, N.H.
When compared with the acquisition and maintenance costs of a large supercomputer, a cluster of low-end servers running a free operating system is a bargain, Haff said. These clusters can't do every high-performance computing task, but often provide good-enough performance for most tasks, he said.
The Los Alamos laboratory, in New Mexico, will use Lightning as a supplement to a larger supercomputer built by Hewlett-Packard, the Los Alamos spokesman said. The systems are used to model the behavior of nuclear explosions to ensure the reliability of the U.S. stockpile of nuclear weapons under the lab's Advanced Simulation and Computing program, he said.