SPIKE GPU – An Implementation of a Recursive Divide-And-Conquer Parallel Strategy for Solving Large Systems of Linear Equations

This project proposes to investigate, produce, and maintain a methodology and its software implementation that leverage emerging heterogeneous hardware architectures to solve billion-unknowns linear systems in a robust, scalable, and efficient fashion. The two classes of problems targeted under this project are banded dense and sparse general linear systems. Preliminary results suggest that the adopted methodology displays a good strong-scaling attribute and its early implementation, called SPIKE, is one order of magnitude faster than competitive software solutions.

Contributors: Ang Li, Radu Serban and Dan Negrut