Revolutions in computing are becoming regular enough to be the status quo. In the mid-1970s, DEC Vax made a great contribution to science and engineering by providing distributed computing resources at the laboratory and department level which previously were available only at the institutional or company level. By the mid-1980s, scientific workstations running Unix provided a comparable level of computing power to small groups and individuals, with national computer networks providing a broadened access to supercomputer centers. In the early 1990s the workstations were providing computer power comparable to that of the supercomputers of the 1980s-for about the same or less money than the workstations of the 1980s.
Just what is being done with all this increase in computational power? Much of it is going towards work on problems which could not even be attempted before, such as lattice gauge calculations and hydrodynamic flow around complicated objects; much of it is going towards visualizations, that is, producing two and three dimensional pictures or graphs of the numerical results of computations; and much is going towards the analysis and visualization of experimental data of various forms.
In our view computers are complicated object much like automobiles; there is no one machine which is best for everything and so the more resources a scientist or engineer has available, the better off he or she will be-especially if he or she uses them wisely. In a typical scenario, programs may be developed, debugged, and tested on the smaller, slower, and friendlier machines, and then ported over to the larger ones for production runs, higher precision, or to speed up the process. While it is often hard to beat the response time and convenience of your own machine, this will not help your productivity much (and that of other users on your system) if your programs take hours to run or if you do not have the tools needed to produce or visualize your results. A rough rule of thumb says if you need to wait much of the day for your results, your progress is suffering too much and you are ready for a more powerful machine. But if you are still correcting your poor computer diction, running test cases, looking at results, and then modifying the code and rerunning, you are probably better off on a local machine.