Improving Performance of Dense Linear Algebra with Multi-core Architecture

Many computational applications, such as loop current analysis in electric systems and truss analysis, involve solving a huge number of linear equations. These linear equations are considered to consume both time and resources, so finding fast solutions to these equations is a major challenge. Both CPUs and Graphical Processing Units (GPUs) can solve these linear equations, but choosing the best architecture that perfectly fits the computation patterns is not an easy task. In this work, Gaussian Elimination, as an accurate method of solving linear equations, is being conducted on different parallel architectures such as multicore architectures, using different codes, and GPUs in order to find the best architecture for this kind of problems. The overall results of the multicore implementations outperformed those using GPUs because of the data transfer time in GPUs.