The company, as reported, has also been showing off – no, not a cardboard replica but an engineering model of its forthcoming scalable, parallel entry system, currently in development at IBM’s Highly Parallel Supercomputing Systems Laboratory at the Supercomputing ’92 event in Minneapolis. It was accompanied by Model 3 Power Visualization System running Wavefront visualisation software, and an eight-way cluster of RS/6000 Powerstation 560s. Fibre Channel Standard technology was also featured for the first time in a cluster and high-speed networking environment. The Highly Parallel System will have eight to 64 RS/6000 processors scaling up to 6 GFLOPS peak with an optional high-performance switch interconnect. Each RISC can be configured with a maximum of 256Mb of memory and 2Gb of disk; a 64-way system can have a maximum memory of 16Gb and 128Gb of disk. IBM says it plans to have Linda, Express, PVM and the IBM Parallel Environment supported by the system. The 16-processor machine on show was running a number of applications, including a three dimensional depth migration seismic parallel code and a protein bending application developed by IBM, and biomechanical applications from the Cornell Theory Center. A three-dimensional aerodynamics simulation using FLO67no within a parallel computational environment, developed by IBM’s European Centre for Scientific and Engineering Computing was also running. The Highly Parallel system has garnered support from Cornell University, New York, and IBM is working with the Cornell Theory Center to develop scientific and technical applications for the machine, the aim being to produce software that will run across the entire family of Power RISC-based high-performance computers. Initial application areas are computational fluid dynamics and plasma physics. And the Halliburton Energy Services Group’s Halliburton Geophysical Services will develop seismic processing applications for the machine, applying its expertise in seismic processing. IBM says the Highly Parallel system will be formally announced shortly. It declined to specify exactly when the product will be unveiled and admitted that it had not yet taken any orders, but said that there is a queue of interest out there. It does say that the highly-parallel system will be generally available sometime next year with up to 64 processors; by 1995, it expects to expand to hundreds of CPUs and by 1997, to thousands, which will then define it as a massively parallel supercomputer.
