Several different classes of high-speed computer systems for scientific computing are investigated. These include mainframes, mainframes with vector facilities, minisupercomputers, vector supercomputers, and vector-multiprocessor minisuper and supercomputers. A finite-difference elastic wave code is run on several high-speed systems. The numerical scheme is based on a dimensional splitting that leads in a natural way to efficient parallel and vector implementation on many high-speed systems. Several hardware and software features of these systems that can affect overall performance, if they are not considered by the code developer, are exposed. Performance results are included for some model problems. >