IBM Corp has been granted a further $85m under the US Department of Energy’s Accelerated Strategic Computing Initiative to develop a 10 TFLOPS supercomputer for simulating nuclear explosions. Like the $93m contract it signed in 1996 (CI No 2,965), the new system will be based on IBM’s RS/6000 SP technology – this time around with up to 8,000 processors – and is due to be installed in 2000. IBM claims the system, commissioned by Lawrence Livermore Labs, will be the world’s fastest supercomputer. IBM was also one of the beneficiaries of the expanded ASCI PathForward initiative, a separate, longer-term research effort (CI No 3,341). Under PathForward, IBM is developing high-speed switching technology to support the 100 TFLOPS performance rates. It expects to demonstrate prototypes in 2001. Other US strategic development laboratories, including as Sandia and Los Alamos have won ASCI funding to buy supercomputer technology from other vendors, including Intel, Sun and Silicon Graphics. The $85m award is IBM’s third under the ASCI initiative. Big Blue has already delivered a 1 TFLOP system, and is to deliver a 3 TFLOPS SP-based system with 4,096 processors – ASCI Blue – next year. It should give the DOE an eight-fold increase in simulation detail compared with what is available today. ASCI’s program calls for 1, 3, 10, 30, and 100 TFLOPS systems, with a 100 TFLOPS system due to be delivered in 2004. The US weapons modeling initiative, which uses massively parallel processing technology, is pitched squarely against a rival Japanese national supercomputer program. Japan’s Science and Technology Agency Earth Simulation Program has commissioned NEC Corp to build a 32 TFLOPS machine by 2002 that will model the earth’s weather and environmental systems (CI No 3,333). NEC’s supercomputer will use a traditional vector processing architecture.
