Published in:Parallel Computing 2017, ParCo2017, Bologna, Italy, 2017-09-12 - 2017-09-15
Language:
English
Origination:
Footnote:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Description:
Satellite-based remote sensing in the mid-infrared spectral region can deliver a wealth of information on pressure, temperature, clouds and aerosols, and trace gas concentrations in the atmosphere. Interpreting the satellite measurements requires to solve an inverse modelling problem based on variational methods and a forward model evaluating the radiative transfer equations. As state-of-the-art satellite measurement campaigns require Petascale systems to process the data in due time, graphical processing units are employed for the high-throughput problem of computing the forward model for a given atmospheric state. We explore features of the considered architecture as well as relevant performance signatures of the different implementations to improve our understanding on opportunities for efficient exploitation of GPU-accelerated architectures based on the POWER8 processor for this class of applications. Scalability is a key aspect as the application is known to scale well on massively-parallel architectures.