Séminaire de groupe
|GPU computing for Molecular Dynamics - and Monte Carlo Simulations|
|jeudi 12 septembre 2013 , 14h00|
|Salle de séminaire du groupe de Physique Statistique|
In the last decade GPU computing became a promising way to deal with computing expensive simulations. The GPU is able to execute a code fragment (thread) for different input data in parallel (up to a few hundred parallel threads). Therefore the GPU can be a lot faster than a single CPU which deals with one input after another. But since the memory exchange between GPU and CPU also costs time it is necessary to put in some thought how to develop an efﬁcient code which is avoiding memory exchange. This talk will present, as a simple case, how GPUs are used for molecular dynamics simulations. Also a more difﬁcult approach for Monte Carlo simulations by the use of Open CL (also applicable for multi-core CPU computation) will be presented. The test case for the simulations was a Lenard-Jones gas, simulated by NVE Molecular Dynamics simulations and NVT Monte Carlo simulation.