Data Centers

Solving Linear Recurrences on Hybrid GPU Accelerated Manycore Systems

Date Added: Oct 2011
Format: PDF

The aim of this paper is to show that linear recurrence systems with constant coefficients can be efficiently solved on hybrid GPU accelerated manycore systems with modern Fermi GPU cards. The main idea is to use the recently developed divide-and-conquer algorithm which can be expressed in terms of Level 2 and 3 BLAS operations. The results of experiments performed on hybrid system with Intel Core i7 and NVIDIA Tesla C2050 are also presented and discussed. Graphical Processing Units (GPUs) have recently been widely used for scientific computing due to their large number of parallel processors which can be exploited using the Compute Unified Device Architecture (CUDA) programming language.