Large Eddy Simulation (LES) applications to Internal Combustion Engine (hereafter ICE) flows are constantly growing, due to the increase of computing resources and the availability of suitable CFD codes, methods and practices. LES superior capability to model spatial and temporal evolution of turbulent flow structures with reference to RANS makes it a promising tool to describe, and possibly motivate, ICE cycle-to-cycle variability (CCV) and cycle-resolved events such as knock and misfire. Despite the growing interest towards LES in the academic community, applications to ICE flows are still limited. One of the reasons for such discrepancy is to be found in the uncertainty in the estimation of the computational cost of LES. This in turn is mainly dependent on grid density, CFD domain extent, time step size and overall number of cycles to be run. Grid density is directly linked to the possibility to reduce modelling assumptions for sub-grid scales. The extent of the computational domain influences the impact of boundary conditions on the CFD results. Time-step size needs to be set according to the size of resolved turbulent eddies .It is therefore closely tied to local grid size with the constraint that CFL number should be lower than unity everywhere in the domain for the highest accuracy. Overall number of simulated cycles influences the soundness of statistical analysis of LES outcomes. This paper focuses on the impact of grid density on the LES description of the TCC-III single-cylinder optical engine flow under motored conditions. In particular, attention is focused on the intake and compression stroke portion of the engine cycle, which govern tumble onset and subsequent decay. LES results are at first evaluated by means of well-established quality indices. Secondly, comparison with available PIV measurements are carried out. Finally, Proper Orthogonal Decomposition is adopted to further assess the impact of grid density of the accuracy of the CFD forecast.