,
APLIMAT'2012

Heterogeneous cluster for acceleration of linear algebra computations
Author: Ivan Šimeček, D. Langr
Keywords

General purpose GPU computing, grid computing, remote calls, library for numerical linear algebra


Abstract

Plenty of numerical algebra libraries have been developed in recent years. These libraries are tuned for the given CPU and its memory architecture, fully utilize its memory hierarchy and inner pipelines and achieve impressive computation power. There is also a new trend in the high-performance computing: GPU computing. This paper deals with a new concept of the heterogeneous grid for acceleration of the numerical linear algebra computing. We design this grid with respect to maximal ratio between cost and computational power. It allows a parallelization of scientific codes with minimal programming effort. We also optimize grid concept to be less sensitive to network parameters.

 


BibTex entry:
@article{JA_APLIMAT12_JOURNAL,
author = {{\v S}ime{\v c}ek, I. and Langr, D.},
title = {{Heterogeneous cluster for acceleration of linear algebra computations}},
journal = {Aplimat - Journal of Applied Mathematics (CD ROM)},
year = {2012},
ISSN = {1337-6365},
language = {English}
}


@inproceedings{JA_APLIMAT12_PROC,
author = {{\v S}ime{\v c}ek, I. and Langr, D.},
title = {{Heterogeneous cluster for acceleration of linear algebra computations}},
booktitle = {{Proceedings of International Conference on Applied Mathematics (Aplimat \'2012)}},
publisher = {Slovensk{\' a} technick{\' a} univerzita, Strojn{\'\i}cka fakulta},
address = {Bratislava},
year = {2012},
pages = {299--306},
ISBN = {978-80-89313-58-7},
language = {English}
}

BACK