Features of a TeraFlop supercomputer
Geoffrey Fox, Andreas Frey · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 1987
We consider architectural, application, and software issues for the coarse grain distributed and shared memory concurrent computers, currently typified by the hypercubes, transputer arrays, the RP3 and BBN Butterfly, and summarize some of the lessons learned from several years of study. Based on these considerations we discuss the possibility of building a supercomputer with a maximum configuration that could sustain performance in the range of 10/sup 12/ floating point operations per second with a memory of order 10/sup 12/ bytes. We show how lessons from the use of current machines indicate that good performance would be obtained on a wide range of scientific problems. We suggest how such a machine could be built in the next few years with current technology. 31 refs., 1 fig., 3 tabs.