The Floating Point Systems T Series

Abstract

The Floating Point Systems T-Series is a massively parallel supercomputer for scientific and engineering applications. Each processing node combines the performance of fully pipelined vector oriented floating point arithmetic with the connectivity and parallel processing capability of the INMOS Transputer. Because each processor has its own memory rather than a switched path to a global memory, very large homogeneous configurations are possible. This paper describes the new architecture and gives an example of distributed processing with associated performance.