The evolution of the approach to Scientific Computing: a Survey

This article surveys relevant present and promising future methods for scientific computing. General concepts, like scalability, latency, saturation, and fault tolerance are recalled. A scientific application of distributed heterogeneous computing to the numerical solution of certain partial differential equations problems is given. Basing on the idea of grid computing, we outline mainstream distributed and parallel environments such as GPU computing and cloud computing. CUDA and OpenCL languages, used to implement algorithms on GPU systems, are also briefly described and compared. The main advantages and potential issues of leveraging such resources for scientific computing are described. In addition, a few novel promising approaches are outlined, such as crowd computing, quantum computing, and frequency-fractal computing. The latter, for instance, can indeed be considered a new paradigm for computing.