Campus Event Calendar
Event Entry
What and Who
Mixed Precision Iterative Refinement - High Accuracy Results from Low Precision Computations
Robert Strzodka
Stanford University
AG4 Group Meeting
AG 4
AG Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Thursday, 20 July 2006
13:00
45 Minutes
E1 4
019
Saarbrücken
Abstract
Recently evolved highly parallel architectures like GPU, FPGA, Cell offer very high single precision
floating point performance, but suffer a huge slowdown when emulating double precision operations. For the
solution of linear systems of equations mixed precision iterative refinement methods are powerful
techniques which allow to concentrate double precision operations in only very few most relevant places
and perform 99% of computations in single precision, while still obtaining the same accuracy as a full
double precision solver. Combining these methods with the fast parallel single precision co-processors
allows to exploit the superior bandwidth and computation power of these devices also in applications with
high accuracy requirements.
Contact
Bodo Rosenhahn
--email hidden
passcode not visible
logged in users only
Bodo Rosenhahn, 05/23/2006 11:10
Bodo Rosenhahn, 05/22/2006 12:39
Bodo Rosenhahn, 05/19/2006 09:59 -- Created document.
Bodo Rosenhahn, 05/22/2006 12:39
Bodo Rosenhahn, 05/19/2006 09:59 -- Created document.