Computer Architectures for Artificial Intelligence Processing
Kai Hwang, Ghosh, Chowkwanyun · Computer · 1987
This article assesses different architectural approaches to the design of computers for artificial intelligence (AI) applications. The authors provide a classification scheme that highlights the relationship between the application domains and hardware realization of AI-oriented computer systems, as well as a comparison of AI machines and numerical computers. The emphasis is on three architectural classes: multiprocessors that support interactive MIMD (multiple-instruction stream and multiple-data stream) operations through a shared memory space; multicomputers that support multiple SISD (single-instruction stream and single-data stream) operations by means of message passing among distributed processors with local memories; and multipurpose computers consisting of a large number of fine-grain processor-memory nodes operating in an SIMD (single-instruction stream and multiple-data stream), a multiple SIMD, or an MIMD fashion.