HIGH-PERFORMANCE COMPUTING

RISC and CISC: When Slower Is Faster

The central processing units (CPUs) of early computers were relatively simple: a few circuits capable of performing a few functions, such as get, a figure, put it somewhere, add it to another, and so on. The speeds of the calculations seemed phenomenal then; what was slow was the process of getting the data and the instructions to the CPU so it could do its work. The solution was to hard-wire more and more instruction sets into the CPU itself: multiply, divide, square root, or whatever; once a set of figures was in the CPU, the complex instruction set was run at blazing speeds, and the CISC (complex-instruction-set computing) machine was born, and kept getting more complex. More and more instructions were hardwired into the chips, and the CPUs got hotter and hotter - both figuratively and literally.

Ultimately, though, chip designers began to see a point of diminishing returns: as more and more functions were added, the chip as a whole was gaining less speed. Since some of the complex functions were only rarely used, the designers figured that it made more sense to leave them out, to be handled by the software. Those rare functions would be handled more slowly, but the reduced-instruction-set computer (RISC) could perform the simpler operations faster, and the system as a whole would operate faster. [ C ]

Posted 20 May 1996