Alliant Computer Systems Corp, Littleton, Massachusetts, has cut down its FX/8 minisupercomputer to create a parallel vector machine that sells for under $100,000, making multi-user supercomputing affordable for the first time to workgroups in industrial and commercial markets. The announcement also includes the company’s first vectorising C compil-er. The FX/4 is a one- to four-processor system fully compatible with the FX/8 and FX/1. In four-processor configuration, it is rated at a peak 64-bit performance of 47.2MFLOPS. All processors, memory, disks and a new VMEbus input-output subsystem fit into a single, 28.25 wide by 43.5 high cabinet. Prices start at $99,900 for a system with one computational compute-intensive vector/scalar processor element, one interactive processor, 32Mb memory and 550Mb disk. Up to four computat-ional elements can work in parallel to execute a single program automatically, with little or no recoding of source programs. In applications where maximum throughput is required, up to four computational elements and six interactive procesors can be applied to the execution of independant jobs. The system can alternate dynamically between using the processors in parallel to accelerate single jobs and using them independently to maximize total system throughput. New software includes FX/Skyline Solver, a highly optimised package for solving systems of linear algebraic equations commonly found in finite element analysis and computational fluid dynamics applications. The new FX/C Release 1.0 typically provides twice the compilation rate and execution performance of the standard Unix C compiler – on the Alliant FX/8, applications are claimed to out-perform a Cray 1S by 20% to 40% by one user. FX/C is the first Alliant implementation of C that incorporates much of the technology of the FX/Fortran compiler, and the full parallelising and vectori-sing compiler technology of FX/Fortran will be integrated into FX/C ver-sion 2.0 for mid-1988. Version 2.0 will automatically detect the poten-tial for parallel and vector processing in standard C code and generate instructions that use them. New FX/Linpack and FX/Eispack scientific libraries are claimed to offer performance typically five times, and in special cases up to 30 times, that of existing mathematical subroutines.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.