A parallel architecture comes of age at lastl hypercube machines reach the market on the promise of executing billions of operations a second to solve difficult simulation problems

Scientists and engineers have grown to expect the performance of their computers to increase by an order I of magnitude about every five years. But that dizzy pace has slowed recently, and supercomputers built around a single processing unit — the Cray-1, the NEC SX-2, or the Fujitsu VP200 — may already be within an order of magnitude of their technological limit. This theoretical upper boundary, some 3 gigaflops (billions of floating-point operations per second), is established by the length of time it takes I electrical signals to propagate, traveling through the wires at about half the speed of light.