From: cwi-all-bounce(a)cwi.nl [mailto:cwi-all-bounce@cwi.nl] On Behalf Of Joost Batenburg
Sent: maandag 18 oktober 2010 15:39
To: cwi-all(a)cwi.nl
Subject: [cwi-all] GPU computing seminar - 17 november
What: CWI GPU computing seminar
When: Wednesday, 17 November
Where: CWI, room TBA
Dear colleagues,
You are all invited to join the first CWI GPU computing seminar on
Wednesday, 17 November. A description of the topic and the
program can be found below.
If you are interested in participating, please register before
3 November by sending an email to Paul de Zeeuw (Paul.de.Zeeuw(a)cwi.nl),
so we can estimate the number of participants.
With best regards,
Joost Batenburg
Paul de Zeeuw
===================================================
CWI Seminar: GPU computing
Within a PC, the Graphics Processing Unit, or GPU, is responsible for 2D
and 3D visualization tasks. Stimulated by the 3D gaming industry, GPUs
have evolved over the years into powerful parallel processors. GPU
computing refers to the use of a GPU for general purpose computing. In
recent years, the GPU has become fully programmable to the point where
it can now be utilized for a wide range of computations, and in
particular for scientific computing. However, the programming model
exposed by a GPU is not as general as the sequential CPU model and
implementing algorithms on the GPU is a delicate task. Nevertheless,
for suitable applications, using a GPU instead of a CPU can result in
impressive speedups of 25 or more.
This seminar aims to serve as an introduction to the emerging field of
GPU computing and provide a global overview of the GPU architecture. A
range of programming models will be covered, ranging from transparent
use of the GPU through Matlab, to direct low-level programming of the
GPU. No background knowledge on GPUs will be assumed.
After this introductory seminar, one or more additional meetings will be
planned that will be focused on particular scientific problems that can
benefit from GPU computing.
Program:
13:30 - 14:15 Introduction GPU computing
Joost Batenburg
14:15 - 14:45 Transparent GPU computing through Matlab (Jacket)
Paul de Zeeuw
14:45 - 15:00 Coffee Break
15:00 - 15:30 Automatic parallellization of program code using the
Portland Group Accelerator
Joke Blom
15:30 - 16:00 Direct GPU programming with CUDA
Willem Jan Palenstijn (University of Antwerp)
16:00 - 16:15 Starting with GPUs: availability and practical
considerations
Joost Batenburg
16:15 - Drinks and informal discussion