Don't Try to Screw in a Nail.

The article focuses on the diversity of computational approaches and mathematical algorithms that came forward over the years during the development of high performance computer technology. It has been assessed that the most beneficial, cost-effective computing resource for a simulation differs considerably from one application to another. The majority of super computer centers choose to have a single general-purpose, homogeneous system. The net result is that some application suffers: the memory intensive calculations do not have optimal amounts memory, while the greater compute-intensive calculations do not have optimal numbers of processors. In addition, different microprocessors are more appropriate for specific types of scientific applications than for others. The lifetime of scientific codes is expressed in decades while the lifetime of computers is expressed in years, which is also a major problem for users of parallel supercomputers.