Accelerating Scientific Computations with Mixed Precision Algorithms
Department of Mathematics, University of Coimbra, Coimbra, Portugal
Computer Physics Communications, Volume 180, Issue 12, December 2009, Pages 2526-2533 (20 Aug 2008)
@article{baboulin2009accelerating,
title={Accelerating scientific computations with mixed precision algorithms},
author={Baboulin, M. and Buttari, A. and Dongarra, J. and Kurzak, J. and Langou, J. and Langou, J. and Luszczek, P. and Tomov, S.},
journal={Computer Physics Communications},
volume={180},
number={12},
pages={2526–2533},
issn={0010-4655},
year={2009},
publisher={Elsevier}
}
On modern architectures, the performance of 32-bit operations is often atleast twice as fast as the performance of 64-bit operations. By using acombination of 32-bit and 64-bit floating point arithmetic, the performance ofmany dense and sparse linear algebra algorithms can be significantly enhancedwhile maintaining the 64-bit accuracy of the resulting solution. The approachpresented here can apply not only to conventional processors but also to othertechnologies such as Field Programmable Gate Arrays (FPGA), GraphicalProcessing Units (GPU), and the STI Cell BE processor. Results on modernprocessor architectures and the STI Cell BE are presented.
November 9, 2010 by hgpu
