A Comparison of Some Current Parallel Computer Architectures

Summary

Ten or twelve years ago parallel computing was a topic for research and parallel computers were mostly built in laboratories [1]. In 1986, Intel started to market a hypercube, the iPSC/1, a commercial product built after the original design of the Cosmic Cube at CalTech, [2]. From there on there was a continuous growth in the spectrum of parallel computers being offered, and also the maximum number of processors a machine could have was steadily increasing. Typically, the manufacturers were either start-ups or electronic companies that had the technology to embark on such development. For a survey of the status of 1990, see [3].

This situation has changed radically. The number of mostly smaller companies that offer parallel computers have now been joined by the traditional manufacturers of vector computers: Cray, Convex, Fujitsu, IBM. At the same time there is a growing number of companies that offer software for parallel computers. In this paper the architectures of some current parallel computers will be reviewed.