Performances of Navier-Stokes Solver on a Hybrid CPU/GPU Computing System

Abstract

A computational code for the numerical integration of the incompressible Navier-Stokes equations for the execution of accurate calculations with the approach of the Direct Numerical Simulation (DNS), is implemented on a specially-assembled hybrid CPU/GPU computing system. The computational code is based on a mixed spectral-finite difference numerical technique, and is implemented onto the plane-channel computing domain, for the study of wall-bounded turbulence. The computing system includes one Intel Corei7 (quad-core) processor, and two Nvidia C-1060 Tesla devices. High-resolution numerical simulations of the turbulent flow in the plane-channel domain are executed at wall-shear-velocity Reynolds number 200, and the performances of the code are reported in terms of parallel-machine metrics. Sample results of the simulations are also reported, in which some details are emphasized of the scientific information that have been obtained, mainly due to the high resolution at which the calculations have been executed, in virtue of the availability of such a powerful computing system.