[gov3009-l] Ramey on "How GPUs Can Help Solve Large Optimization Problems"

Matt Blackwell mblackwell at iq.harvard.edu
Mon Oct 4 16:47:41 EDT 2010

Hi all,

We hope you can join us for the Applied Statistics Workshop this
Wednesday, October 6th, when we will be happy to have Adam Ramey, who
is currently a Statistician/Analyst at the Harvard Business School,
presenting a talk on utilizing graphical processing units for large
optimization problem. You will find details about the talk and an
abstract below. We will serve lunch (sandwiches) at the workshop. Hope
you can make it.

"Graphic Materiel: How GPUs Can Help Solve Large Optimization Problems."
Adam Ramey
Harvard Business School
October 6th, 2010, 12 noon
K354 CGIS Knafel (1737 Cambridge St)

The advent of multi-core processing has led to the implementation of
explicit and implicit multithreading in most modern statistical
packages. Unfortunately, in many of these cases, the user is required
to "hard code" how the processes are to be divided up across multiple
cores. Perhaps more problematic, the utility of this parallelization
is restricted by the number of threads/cores in a user's machine. To
address this issue in a cost-effective way, GPU computing has emerged
as a powerful alternative. Harnessing the power of hundreds of small
cores, GPU's are able to parallelize at a level virtually impossible
in a CPU framework. In this talk, I discuss how MATLAB and R can take
advantage of the GPU and present Monte Carlo experiments verifying the
sheer power of the method.

Matthew Blackwell
PhD Candidate
Institute for Quantitative Social Science
Department of Government
Harvard University
url: http://people.fas.harvard.edu/~blackwel/

More information about the gov3009-l mailing list