Supercomputing at FSU

Over the last several months the experimental physicists at FSU and SCRI have investigated the use of properties of the CYBER-205 on pattern recognition and Monte Carlo calculations. We have found that extremely large performance enhancements can be achieved with such machines if the algorithms are assembled out of intrinsicly vector constructs. This usually requires changes of algorithms. These changes almost always produce codes of a simpler nature emphasizing the rule rather than the exception as is the case in scalar codes. The use of predefined look-up tables allows the exceptions to be handled within the structure of the more common cases making the structure of the algorithm more apparent. This gives speed increases as large as a factor of ten. 2 refs., 5 figs.