The practicality of SIMD for scientific computing
J.R. Fischer, L.E. Harnet, Clark M. Mobarry, J. A. Pedelty, J. S. Cohen, R.K. de Fainchtein, Bruce A. Fryxell, P. J. MacNeice, T. Olson, Thomas Sterling · 2002
While not popular at present for scientific computing, SIMD has proven to be a practical option for tackling a variety of scientific problems. The MasPar MP-2 at NASA/Goddard Space Flight Center is being successfully employed for Grand Challenges in Earth and space science. This paper considers the applicability, specific implementation and performance of five algorithms-astrophysical fluid dynamics, particle-in-cell techniques, N-body simulations, finite element simulations of fluids, and image processing. When compared to a single-processor CRAY C90, the MasPar MP-2 exhibits substantially better performance-to-cost and fully competitive sustained performance.>