Introducing Replicated VLSI to Supercomputing: the FPS-164/MAX Scientific Computer

• Published in: Computer (Long Beach, Calif.) Calif.), 1986-03, Vol.19 (3), p.10-23
• Main Authors: Charlesworth, A E, Gustafson, J L
• Format: Article
• Language: English
• Subjects: Adders; Atmospheric modeling; Business; Central Processing Unit; Computational modeling; Digital arithmetic; Floating-point arithmetic; Logic; Scientific computing; Very large scale integration
• ISSN: 0018-9162; 1558-0814
• DOI: 10.1109/MC.1986.1663175

Before the advent of VLSI, the implementation of the computational operations required for performing very fast calculations with great accuracy was very expensive. This situation began to change with the development of VLSI circuits for fast 64-bit arithmetic by an American company in mid-1984. This advance made it possible to reduce the size of a high-speed floating point data unit from several thousand chips to nine chips. The present paper is concerned with a computer, which, initially delivered in April 1985, represents the first computer to be implemented from replicated VLSI arithmetic parts. It is also the computer to provide several hundred MFLOPs in a supermini-priced processor. Attention is given to the impact of VLSI on number crunching, matrix supercomputing, a matrix supercomputer, and parallel linear algebra. Applications of the new computer are related to structural analysis, computational chemistry,and radar cross sections. (G.R.)