"Implementation of parallel calculations on graphics processor units in the LOGOS computational fluid dynamics package"
Volkov K.N., Derugin Yu.N., Emelyanov V.N., Karpenko A.G., Kozelkov A.S., Smirnov P.G., Teterina I.V.

A number of tools for the numerical simulation of three-dimensional flows of viscous compressible gas with the use of graphics processor units are developed. The finite volume method implemented in the LOGOS package is used to solve the Navier-Stokes equations on unstructured meshes. Different methods for the flux calculations through faces of finite volumes and features of implementation of a turbulence model are considered. The CUDA technology is used to implement parallel numerical algorithms. The solutions to some benchmark problems of internal and external gas dynamics are discussed. The speed-up of calculations performed on graphics processor units compared to the calculations performed on the corresponding central processor unit is analyzed.

Keywords: graphics processor unit, parallel algorithm, gas dynamics, finite volume method, unstructured mesh, CUDA technology, LOGOS package

Volkov K.N., e-mail: dsci@mail.ru;   Emelyanov V.N., e-mail: vlademelyanov@gmail.com;   Smirnov P.G., e-mail: petr.s.8314@mail.ru;   Teterina I.V., e-mail: yaiv@mail.ru – Voenmekh D. Ustinov Baltic State Technical University, Power Machine Building Faculty; Kransoarmeejskaja ulitsa 1, Sankt-Petersburg, 190005, Russia;
Derugin Yu.N., e-mail: askozelkov@mail.ru;   Kozelkov A.S., e-mail: askozelkov@mail.ru – Russian Federal Nuclear Center, All-Russian Research Institute of Experimental Physics; prospekt Mira 37, Sarov, 607190, Russia;
Karpenko A.G.   e-mail: aspera.2003.ru@mail.ru – Saint-Petersburg State University, Faculty of Mathematics and Mechanics Faculty; prospect Universitetskii 28, St. Petersburg, 198504, Russia