IBM Corp has secured a contract from the Lawrence Livermore National Laboratory in California to build what’s dubbed as the world’s fastest supercomputer. For $93m, IBM will build an RS/6000 SP-based system to be installed as part of the Department of Energy’s Accelerated Strategic Computing Initiative (ASCI). This is all necessary because of the various nuclear test-ban treaties around the world denying scientists the pleasure of blowing the things up in the desert and recording the results. The supercomputers in the ASCI program will be used to construct three-dimensional simulations that scientists will use to analyze the effects of aging and other problems associated with storing large numbers of nuclear warheads: what is known as stockpile stewardship. The eventual machine will have 4,096 RISC shared-memory processors clustered together, with the first components delivered next month. The full 3-teraflop device is slated for December 1998 delivery. Another 3 teraflop device contract is currently being negotiated with a vendor other than IBM for a Los Alamos lab, also involved in the ASCI program, along with the DOE’s Sandia National Laboratories, which has facilities in both California and New Mexico. There are five teraflop stages involved in the program: 1.8, 3, 10, 30 and 100-teraflops. The 1.8 teraflop machine is the one Intel Corp won the contract to build last year for $45m (CI No 2,745) at Sandia. But each stage may take more than one computer, as is the case with the 3-teraflop machine.
