Early Massively Parallel Architectures

Before massively parallel processing became a standard supercomputer design, a small number of research and commercial machines pioneered the approach of coordinating very large numbers of processors working together rather than relying on one or a few fast processors. The University of Illinois's ILLIAC IV, developed beginning in the 1960s, is generally credited as the first massively parallel computer, originally planned to hold 256 processing elements before budget cuts limited the finished machine to 64. Thinking Machines Corporation's Connection Machine series, which grew out of Danny Hillis's doctoral research at the Massachusetts Institute of Technology into alternatives to the conventional von Neumann architecture, pushed the same idea toward commercial artificial intelligence and scientific computing use starting with the CM-1 in the 1980s. Both machines demonstrated, in different decades and for different sponsors, that a supercomputer's performance could scale from coordinating enormous processor counts rather than from a single faster processor.