Introduction. Code_Saturne is a multipurpose Computational Fluid Dynamics (CFD) software package, which has been developed by EDF (France). Code_Saturne's open-source status allows for answers to specific needs that cannot easily be made available in commercial "black box" packages. It is also one of the 12 solvers selected by the Unified European Application Benchmark Suite (UEABS) inside PRACE and has been thoroughly proved to scale on large systems.Goal.The goal of this project is to use high performance computing to assess how Code_Saturne is able to characterize and solve turbulent flows, using the Taylor-Green vortex as a benchmark problem.

Nowadays, turbulent flows are of great interest in many applications because they mix fluid much more effectively than a comparable laminar flow due to its diffusivity. The conservation of the properties of the Navier-Stokes equations is of extremely importance for an accurate description of turbulent flows. This thesis analyzes the behavior of Code_Saturne, a multipurpose open-source CFD software package that is included in the Partnership for Advanced Computing in Europe (PRACE), in terms of kinetic energy conservation in turbulent flows. In order to do so, Code_Saturne is compared with a selfmade spectro-consistent 2D code and another spectro-consistent 3D code that exactly preserve the symmetries of the underlying differential operators of the Navier-Stokes equations, i.e. the convective operator is approximated by a skew-symmetric matrix and the diffusive operator by a symmetric, positive-definite matrix. The well known benchmark cases of the 2D Taylor vortex and the 3D Taylor-Green vortex are solved. A sensitivity analysis is performed in order to assess the best parameters of Code_Saturne that yield the best performance for both structured and unstructured meshes. The results show that Code_Saturne strictly conserves kinetic energy on regular Cartesian grids if an appropriate fully centered scheme is used together with the deactivation of the Rhie Chow interpolation. On 3D grids, the fully turbulent flow reveals that the presence of three dimensional effects due to vortex stretching causes a numerical dissipation that can be overcomed by refining the mesh size. However, this formulation does not allow to strictly preserve kinetic energy on unstructured grids due to the instabilities generated on the pressure gradient term. Finally, a set of configuration parameters that follow the aforementioned properties for Code_Saturne are provided.