Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Perfect strong scaling using no additional energy

James Demmel, Andrew Gearhart, Oded Schwartz and Benjamin Lipshitz

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2012-126
May 30, 2012

http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-126.pdf

Energy efficiency of computing devices has become a dominant area of research interest in recent years. Most of this work is focused on architectural techniques to improve power and energy efficiency; only a few consider saving energy at the algorithmic level. We prove that a region of perfect strong scaling in energy exists for matrix multiplication (classical and Strassen) and the direct (O(n2)) n-body problem via the use of .5D algorithms: This means that we can increase the number of processors by a constant factor, with the runtime (both computation and communication) decreasing by the same factor, and the total energy used remaining constant.


BibTeX citation:

@techreport{Demmel:EECS-2012-126,
    Author = {Demmel, James and Gearhart, Andrew and Schwartz, Oded and Lipshitz, Benjamin},
    Title = {Perfect strong scaling using no additional energy},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {2012},
    Month = {May},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-126.html},
    Number = {UCB/EECS-2012-126},
    Abstract = {Energy efficiency of computing devices has become a dominant area of research interest in recent years. Most of this work is focused on architectural techniques to improve power and energy efficiency; only a few consider saving energy at the algorithmic level. We prove that a region of perfect strong scaling in energy exists for matrix multiplication (classical and Strassen) and the direct (O(n2)) n-body problem via the use of .5D algorithms: This means that we can increase the number of processors by a constant factor, with the runtime (both computation and communication) decreasing by the same factor, and the total energy used remaining constant.}
}

EndNote citation:

%0 Report
%A Demmel, James
%A Gearhart, Andrew
%A Schwartz, Oded
%A Lipshitz, Benjamin
%T Perfect strong scaling using no additional energy
%I EECS Department, University of California, Berkeley
%D 2012
%8 May 30
%@ UCB/EECS-2012-126
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-126.html
%F Demmel:EECS-2012-126