Architecture and Applications of a Heterogeneous, Massively Parallel Machine
D E Shaw · Annual Review of Computer Science · 1986
The organization of an experimental, massively parallel machine called NON VON is described, along with some of the typical artificial intelligence applications for which the machine is intended to provide significant perform ance and cost/performance advantages over conventional computer systems. The machine incorporates an active memory, which is constructed using custom, very large scale integrated (VLSI) chips. Each chip contains a number of simple processing elements and a small number of larger process ing elements, each capable of controlling the operation of a subset of the active memory. A simplified, preliminary prototype of the NON-VON archi tecture is now operational at Columbia University. Performance projections, derived through detailed analysis and simulation, are summarized for applications in the areas of rule-based inferencing , com puter vision, and knowledge base management. The results, most of which are based on benchmarks proposed by other researchers, suggest that NON VON could improve performance of such tasks by as much as several orders of magnitude, compared to a conventional sequential machine of comparable hardware cost.