Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 41: Poster - Pattern/ Nonlinear Dyn./ Fluids/ Granular/ Critical Phen.
DY 41.16: Poster
Donnerstag, 3. April 2014, 17:00–19:00, P3
POCLMD: A Flexible GPU Accelerated Molecular Dynamics Code — •Chuanfu Luo1 and Jens-Uwe Sommer1,2 — 1Leibniz-Institut für Polymerforschung Dresden, Germany — 2Technische Universität Dresden, Germany
Recently, heterogeneous computing becomes a trend in high performance computing. Many newly built super-computers are equipped with powerful GPUs or coprocessors, which can deliver up to 5 Tflops per unit. The OpenCL is an open standard programing language which is supported by all vendors of GPU or coprocessor. Code written in OpenCL can run on multiple-core CPUs, GPUs, and coprocessors without any changes. POCLMD is designed to be a fast, flexible and easy-to-use MD code, which is programed in Python and OpenCL through PyOpenCL [1]. All time-consuming calculations are performed on GPU, and CPU only deals with data transfer and pre-/post- processing. The first released version of POCLMD will support basic simulations of NVE/NVT/NPT for coarse-graining polymeric models, including bond, angle and pair-wised LJ potentials. Further potentials of dihedral and long-range electric interactions will be supported later. The performance of POCLMD can catch up >80% of HOOMD-blue [2] (the fastest CUDA based MD code as we know) on the same NVIDIA's GPUs. POCLMD on one AMD's HD7970 card runs 60-80 times faster than LAMMPS on a single core Q6600 CPU (2.4GHz).
[1] http://mathema.tician.de/software/pyopencl
[2] http://codeblue.umich.edu/hoomd-blue/