Abstract
To help evaluate the supercomputers, the authors selected a computational neuroscience simulation program that has a bandwidth requirement in excess of 1.5 Gb/s. The simulation runs on the Thinking Machines CM-2 and the target visualization engine is the Convex C3880. The authors benchmarked the various operations required to transmit data from the CM-2 to the Convex C3880. The results show that the parallel to serial data transformation operation on the CM-2 and the network interfaces are the major bottlenecks.