Abstract

The commercial availability of powerful 32-bit microprocessors provides an unprecedented opportunity for researchers to construct inexpensive but very powerful multiprocessor computer systems that are optimally designed for specific computational tasks. However, to obtain good performance from such systems it is essential that both the processor interconnection architecture and the system software be appropriate for the problem. Two multiprocessor system designs being developed at the Optical Sciences Center illustrate the potential of such systems. The first is an image- data-driven dynamically reconfigurable multiprocessor using thirty Motorola 68000 microprocessors interconnected by multiple global data buses. This system is fully operational and is being used to develop machine vision systems capable of very high speed biomedical image acquisition and image understanding. The second system is still in the design phase and will be used for image reconstruction using simulated annealing. It consists of an array of sixty-four Inmos Transputers with fourway or six-way nearest-neighbor interconnections.

© 1986 Optical Society of America