PARALLEL COMPUTER ARCHITECTURES - STATE-OF-THE-ART AND TRENDS

An increasing number of parallel architectures is becoming available for numerically intensive applications. Many chemical problems need intensive calculations due to the complexity of the underlying physical models. Very often these applications show an intrinsic parallelism and therefore can be easily adapted to parallel machines. In the future, in addition to the classical numerically intensive applications, the use of these machines will be extended to a more general purpose use (e.g. data base machines, advanced graphics, AI and expert systems applications, etc.). The principal aim of this paper is to show the state of the art of the commercially available parallel architectures and related trends. A comparison of the main features of shared and distributed memory systems will be presented. The characteristics of coarse and fine grained architectures will be discussed. The analysis will include not only the large-scale machines (usually called "supercomputers"), but also smaller machines (e.g. minisuper and multi-computers) having a very favourable price/performance ratio.