Кудрявцев А.Н.   Kashkovsky A.V.   Borisov S.P.   Шершнёв А.А.  

A numerical code for the simulation of non-equilibrium chemically reacting flows on hybrid CPU-GPU clusters

Reporter: Шершнёв А.А.

Problems including non-equilibrium chemically reacting flows are considered among the most complicated problems of aerothermodynamics. The need to resolve characteristic time of chemical processes results in stiffness of the equations and greatly increases the computational cost of numerical simulation of such flows. One of the promising ways to improve the efficiency and reduce wall-clock time of the computations is the employment of hybrid computational systems, which include both central processing units (CPUs) and graphical processing units (GPUs). The latter ones are often referred as graphical accelerators and act as powerful coprocessor. Usage of hybrid CPU-
GPU systems is currently a rapidly growing area of computing. However, when designing the program for hybrid computational clusters, one can encounter certain difficulties caused by specific features of the GPU architecture.

In the present work a computer code for numerical simulation of chemically reacting compressible flows on hybrid CPU/GPU supercomputers is developed. It solves 3D unsteady Euler equations for multispecies chemically reacting flows in general curvilinear coordinates using 3rd order shock-capturing TVD schemes. Time advancement is carried out using approach, based on the explicit 2nd order Runge-Kutta TVD scheme. Program implementation uses CUDA
application programming interface to perform GPU computations. Data between GPUs is distributed via domain decomposition technique. Within one computational node data exchange is carried out using OpenMP threads. Inter-node exchange uses MPI library. The developed code is verified on the number of test cases and its performance is measured against the existing hybrid code for the single-species perfect gas.


To reports list