![]() ![]() However, using them efficiently requires different programming methods and algorithms than those employed on standard CPUs. GPUs are built to handle numerous lightweight parallel threads simultaneously and to offer theoretical peak performance far beyond that of current CPUs. The development of relatively inexpensive graphic processing units (GPUs) with large numbers of cores and the development of relatively easily used programming models, OpenCL and CUDA has made the use of GPUs attractive for accelerating scientific simulations. The standard method of speeding up these calculations is to implement parallel algorithms for distributed memory machines using the MPI library ( Hauschildt & Baron 2006 Baron & Hauschildt 2007). The 3DRT framework discussed in the previous papers of this series requires very substantial amounts of computing time owing to the complexity of the radiative transfer problem in strongly scattering environments. 2010, hereafter: Papers I–VII), we have described a framework for solving the radiative transfer equation in 3D systems (3DRT), including a detailed treatment of scattering in continua and lines with a nonlocal operator splitting method and its use in the general model atmosphere package PHOENIX. 2009 Hauschildt & Baron 2010 Seelmann et al. In a series of papers ( Hauschildt & Baron 2006 Baron & Hauschildt 2007 Hauschildt & Baron 2008, 2009 Baron et al. Key words: radiative transfer / methods: numerical ![]() ![]() The latest available GPUs can lead to significant speedups for both small and large grids compared to serial (single core) computations. We present the results of both a small and a large test case and compare the timing of the 3DRT calculations for serial CPUs and various GPUs.Ĭonclusions. We implemented the kernel for solving the 3DRT problem in Cartesian coordinates with periodic boundary conditions in the horizontal ( x,y) plane, including the construction of the nearest neighbor Λ ∗ and the operator splitting step. We discuss an implementation of our 3D radiative transfer (3DRT) framework with the OpenCL paradigm for general GPU computing. of Physics and Astronomy, University ofĮ-mail: Research Division, Lawrence Berkeley NationalĪims. Astronomical objects: linking to databasesĮ-mail: L.Including author names using non-Roman alphabets.Suggested resources for more tips on language editing in the sciences Punctuation and style concerns regarding equations, figures, tables, and footnotes ![]()
0 Comments
Leave a Reply. |