Intel Corp yesterday unveiled in the US its second generation supercomputer, the Paragon XP/S, based on the Intel 80860XP RISC and scalable from 66 to 4,000 processor nodes, delivering between 5 GFLOPS and 300 GFLOPS. The new multiple instruction multiple data massively parallel machine, which is compatible with Intel’s existing iPSC/860 parallel machine but which embodies the architecture Intel plans to use in future machines capable of TeraFLOPS perfomance, will cost from $2m to $55m, to ship in the first half 1992. The scalable Unix running on the Paragon XP/S is based on a Mach microkernel combined with Open Software Foundation OSF/1 server technology. Aimed at technical applications, Paragon uses technology from the Intel/DARPA Touchstone Research Programme. Each processing node consists of several computer chips including two 80860XP microprocessors. Intel’s older line of supercomputers offers up to 128 processors, delivering 7.6 GLOPS. The new machine, which aggressively challenges the likes of Cray – which doesn’t expect to have a 200 GFLOPS machine out until 1997, doesn’t require a front-end or system manager, thus eliminating the possibility of an input-output bottleneck. Orders have flooded in from Boeing, Seattle and Grant Tensor Geophysical, Houston, Texas, Merrill Lynch, and the Research Centre Julich (KFA) in Germany. And both Argonne National Laboratory, Chicago, and the Oak Ridge National Laboratory, Oak Ridge, Tennessee plan to serve as beta sites.
