Experimental grid for numerical linear algebra
Author: Ivan Šimeček, J. Krupka, J. Hladík, M. Hovorka

grid; GPGPU; numerical linear algebra; parallel algorithms; CUDA; OpenCL


Time is very often the limiting factor in scientific codes. These codes can be accelerated by parallel executing on special distributed systems (grids). This is usual but very difficult solution. In this paper, we describe a design of the new heterogenous grid for the numerical linear algebra with maximal ratio between prize and computational power. Contributions of this paper is twofold: 1) a design of new parallel routines 2) an approach for parallelization of scientific codes by converting local numerical library calls into remote grid calls.


final version (in .PDF format)

BibTex entry:
author = {{\v S}ime{\v c}ek, I. and Hlad{\'\i}k, J. and Krupka, J. and Hovorka, M.},
title = {Experimental grid for numerical linear algebra},
booktitle = {SNA'11 Seminar on Numerical Analysis. Modelling and Simulation of Challenging Engineering Problems},
publisher = {{\' U}stav geoniky AV {\v C}R},
address = {Ostrava},
year = {2011},
pages = {104--107},
ISBN = {978-80-86407-19-7},
language = {English}