HIGH-EFFICIENCY COMPUTING SYSTEMS

Scalability of CFD-applications on hybrid computing systems

A. A. Davidov

Keldysh Institute of Applied Mathematics of Russian Academy of Sciences, Moscow

Abstract: One of important characteristics of computing appendices is their scalability — that is opportunity to beaccelerated at increase in volume of the involved computing resources. This characteristic not so much to the appendix, how many to set of the appendix and the computing system. With the advent of powerful computing systems of hybrid architecture with a large number of graphic processors (GPU) there is a question of possibility of their effective use for the solution of big tasks.
In this article the parallel efficiency of the “Express–3D” program complex [4] is investigated at the solution of the Euler equations and of the quasigasdynamic equations [1] on K-100 [6] computing systems and "Lomonosov" system [7] with use of graphic NVIDIA Fermi processors (Tesla C20XX).

Keywords: CFD, HPC, GPGPU, CUDA, quasigasdynamic equations.