SIMD: Specialisation Equals Success

Abstract

The paradigm of lock-step parallelism embodied in SIMD computers has proved to be extremely successful at solving a wide range of scientific problems. One reason for this is that many of the problems which arise in more general models of parallelism, such as deadlock, cannot occur on SIMD machines. As a result, there has been greater architectural convergence among SIMD manufacturers, and more progress toward standards such as FORTRAN 90, than anywhere else in the parallel computing field. The increasingly competitive struggle between Thinking Machines Corp. and other manufacturers at the “speed at any price” end of the market, along with AMT’s fight to expand its share of the signal and image processing market, have been made even more intense by MasPar’s recent entry into this field.