The Dataparallel Computer MasPar MP - 1

By using CMOS VLSI and replication of components effectively, massive parallel computers can achieve extraordinary performance at low cost, Key issues are how processors and the memory are partitioned and replicated, and how interprocessor communication and I/O are accomplished. From the user’s point of view, the ease of programming and the programming environment are of major importance. Dataparallelism is the natural extension of vector processing and allows for a transparent programming and debugging. This paper describes the design of a modern SIMD parallel computer that achieves peak rates above one billion floating point operations per second at minicomputer prices.