Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

P: Fachverband Plasmaphysik

P 12: Theory/Modelling II

P 12.1: Fachvortrag

Mittwoch, 10. März 2010, 14:30–14:55, B 302

PIConGPU - A scalable particle-in-cell algorithm for graphic cards — •Michael Bussmann¹, Heiko Burau¹, René Widera², Wolfgang Hönig², Guido Juckeland², Alexander Debus¹, Thomas Kluge¹, Ulrich Schramm¹, Tom Cowan¹, and Roland Sauerbrey¹ — ¹Forschungszentrum Dresden-Rossendorf e.V., D-01328 Dresden — ²Zentrum für Informationstechnologie und Hochleistungsrechned, Technische Universität Dresden, D-01062 Dresden

We present the main features of PIConGPU, the - to our knowledge - first scalable particle-in-cell (PIC) code for relativistic laser plasma interactions written for graphical processing units (GPUs). We show that it is possible to use PIConGPU on standard compute clusters equipped with GPUs and introduce the main features that are important to reach good weak scaling when increasing both the size of the system simulated and the number of GPUs.