Speaker
Mr
Axel Huebl
(Helmholtz-Zentrum Dresden - Rossendorf)
Description
PIConGPU reportedly is the fastest particle-in-cell code in the world with respect to sustained Flop/s. Written in performance-portable, single-source C++ we constantly push the envelope towards Exascale laser-plasma modeling. However, solving previously week-long simulation tasks in a few hours with a speedy framework is only the beginning.
This talk will present the architecture and recent additions driving PIConGPU. As we speak, we run on the fastest machines and the community approaches a new generation of TOP10 clusters. Within those, many-core computing architectures and severe limitations in available I/O bandwidth demand fundamental rethinking of established modeling workflows towards in situ-processing.
We present our ready-to-use open-source solutions and address scientific repeatability, data-reduction in I/O, predictability and new atomic modeling for XFEL pump-probe experiments.
Primary author
Mr
Axel Huebl
(Helmholtz-Zentrum Dresden - Rossendorf)
Co-authors
Dr
Alexander Debus
(Helmholtz-Zentrum Dresden - Rossendorf)
Mr
Fabian Koller
(Helmholtz-Zentrum Dresden - Rossendorf)
Mr
Heiko Burau
(Helmholtz-Zentrum Dresden - Rossendorf)
Dr
Hyun-Kyung Chung
(International Atomic Energy Agency)
Dr
Jan Vorberger
(Helmholtz-Zentrum Dresden - Rossendorf)
Mr
Marco Garten
(Helmholtz-Zentrum Dresden - Rossendorf)
Dr
Michael Bussmann
(Helmholtz-Zentrum Dresden - Rossendorf)
Mr
René Widera
(Helmholtz-Zentrum Dresden - Rossendorf)
Mr
Richard Pausch
(Helmholtz-Zentrum Dresden - Rossendorf)
Prof.
Thomas COWAN
(Helmholtz-Zentrum Dresden - Rossendorf)
Dr
Thomas Kluge
(Helmholtz-Zentrum Dresden - Rossendorf)
Prof.
Ulrich Schramm
(Helmholtz-Zentrum Dresden - Rossendorf)
