Adjustable GPU Acceleration for Hermitian Eigensystems
IDEAS Research Institute, Robert Gordon University, Aberdeen AB25 1HG, United Kingdom
Lecture Notes in Computer Science 01/2013, 7776:150-161, 2013
@article{garba2013adjustable,
title={Adjustable GPU Acceleration for Hermitian Eigensystems},
author={Garba, Michael T and Gonz{‘a}lez–V{‘e}lez, Horacio and Roach, Daniel},
year={2013}
}
This paper explores the early implementation of high-performance routines for the solution of multiple large Hermitian eigenvector and eigenvalue systems on a Graphics Processing Unit (GPU). We report a performance increase of up to two orders of magnitude over the original EISPACK routines with a NVIDIA Tesla C2050 GPU, potentially allowing an order of magnitude increase in the complexity or resolution of a Inelastic Neutron Scattering (INS) modelling application.
September 18, 2013 by hgpu
