Published June 3, 2013
| Version v1
Working paper
Open
GPU Implementation of the DP code
Creators
- 1. Laboratoire des Solides Irradies, Ecole Polytechnique, CNRS, CEA, UMR 7642, 91128 Palaiseau cedex, France
Contributors
- 1. Laboratoire des Solides Irradies, Ecole Polytechnique, CNRS, CEA, UMR 7642, 91128 Palaiseau cedex, France
- 2. Scientic Computing Laboratory, Institute of Physics Belgrade, University of Belgrade, Pregrevica 118, 11080 Belgrade, Serbia
- 3. Maison de la Simulation, USR 3441, bat. 565, CEA Saclay, 91191 Gif-sur-Yvette cedex, France
Description
Main goal of this PRACE project was to evaluate how GPUs could speed up the DP code { a
linear response TDDFT code. Pro ling analysis of the code has been done to identify computational
bottlenecks to be delegated to the GPU. In order to speed up this code using GPUs, two di erent
strategies have been developed: a local one and a global one. Both strategies have been implemented
with cuBLAS and/or CUDA C. Results showed that one can reasonably expect about 10 times speedup
on the total execution time, depending on the structure of the input and the size of datasets used,
and speedups up to 16 have been observed for some cases.
Files
WP67.pdf
Files
(961.6 kB)
Name | Size | Download all |
---|---|---|
md5:5ea3979298e5d1f9f005a1b10eabc39a
|
961.6 kB | Preview Download |