Berlin 2015 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 32: P4: Computational Physics of Soft Matter

CPP 32.7: Poster

Dienstag, 17. März 2015, 14:00–16:00, Poster B

Ewald summation on GPU — •Sascha Ehrhardt and Axel Arnold — Institute for Computational Physics, Stuttgart, Germany

We implemented the Ewald summation algorithm for the computation of electrostatic energies in molecular dynamics simulations for graphic cards (GPUs) using CUDA. Our implementation is part of the open source molecular dynamics package ESPResSo (http://espressomd.org).

The Ewald summation splits up the Coulomb energy sum in two sums, where one is calculated in real space and the other in reciprocal space. Because the two sums can be computed independently of each other, the real space part will be computed on the CPU using the efficient short range routines of ESPResSo, whereas the reciprocal space part will be computed on the GPU. The results are compared to the existing particle-particle particle-mesh (P3M) algorithm without GPU support. For moderate densities, the GPU-accelerated Ewald summation is comparable in efficiency with the better scaling P3M algorithm up to about 3000 particles, but allows for much higher accuracies without drastic penalties if desired.