In recent years, Computational Fluid Dynamics (CFD) is used more and more as a design tool for industrial applications, such as the medical, automotive and renewable energy in- dustries. The main constraint of the use of CFD in industrial applications nowadays lies in the computational cost and the wall-clock simulation time. This constraint inevitably leads to a cost-versus-accuracy trade-off when simulations are performed. This work is part of a larger project [1] in which the viability of conducting overnight LES simulations on GPU-accelerated supercomputers is evaluated, aiming to combine a highly-portable al- gebraic framework with a symmetry-preserving discretisation for unstructured collocated grids. The former part aims to cut down on the cost side, whereas a carefully chosen discretisation can greatly impact accuracy of the solution. The symmetry-preserving discretisation aims to conserve energy, momentum and mass of the simulation by mimicking properties of the continuous operators of the Navier-Stokes equations in their discrete counterparts. This property is deemed essential in accurately depicting the motion of fluids at any scale, which has to be carried out properly in turbu- lent flow simulations. The effect on the accuracy of applying this scheme will be compared to the use of non-symmetry-preserving schemes using the open-source code OpenFOAM, for which the method was previously implemented by [2]. Using several benchmark cases and monitoring higher order turbulent statistics, while also monitoring computational costs, an extensive cost-versus-accuracy analysis is performed for this method. Analysing and improving the resulting accuracy, while simultaneously decreasing computational cost, these combined efforts will form a step towards feasible high-fidelity overnight simulations.