Within the past several years, algorithms for realistic descriptions of materials have been developed and implemented on supercomputers. It is reasonable to expect this trend to continue and, most likely, accelerate. The consequences of this 'megatrend' will be accurate numerical experiments which will supplement, and in some cases supplement, laboratory experiments. We review briefly how pseudopotential methods have played an important role in this trend and how such methods are likely to play a leading role in the development of new algorithms. Recent 'experiments' on supercomputers for the structures of amorphous solids, and liquids will be used to illustrate how materials science can be done on the computer.