Highlights of Alliant’s Parallel Supercomputer Generation FX/2800 Series

Abstract

Alliant Computer Systems’ new generation of supercomputers, the FX/2800 formally represents a new level of integration in supercomputing. To the parallel processing and visualization technologies, Alliant adds the power of RISC processing with advanced compiler algorithms. Alliant’s use of the Intel 1860 RISC processor reflects a commitment to use high-performance VLSI components to create an open, non-proprietary architecture. This design approach not only puts supercomputing on a much more aggressive price/performance curve (akin to the price/ performance curve for PCs and workstations), but allows engineering efforts to focus on value-added technology rather than reinventing the processor. Alliant’s Concurrency Control Architecture (CCA), which applies parallel processing to graphics as well as computationally intensive applications, has been adopted as the model for the Parallel Architecture Extended (PAX) standard published by Intel Corporation. CCA could become the standard for parallel processing not only at the supercomputer level but also at the workstation and PC levels. This could increase exponentially the number of off-the-shelf applications for supercomputing.

Alliant’s FX/2800 series of supercomputers achieves a peak performance of 1120 Double Precision MFLOPS, 672 Double Precision Whetstone MIPS or 1148 Dhrystone MIPS by using up to 28 processors in the system. The same processors work in parallel on graphical applications, thus reducing the need for special graphics hardware and costs. The FX/2800 can be equipped with up to 1GB of dual ported main memory that is accessed by all processors using shared memory techniques via 8xS12 KB cache modules. The Concurrency Control Architecture provides hardware support for loop-level parallelism.