Large eddy simulations (LES) of a port-injected 4-valve spark ignited (SI) engine have been carried out with the emphasis on the combustion process. The considered operating point is close to full load at 3,500 RPM and exhibits considerable cyclic variation in terms of the in-cylinder pressure traces, which can be related to fluctuations in the combustion process. In order to characterize these fluctuations, a statistically relevant number of subsequent cycles, namely up to 40, have been computed in the multi-cycle analysis. In contrast to other LES studies of SI engines, here the G-equation (a level set approach) has been adopted to model the premixed combustion in the framework of the STAR-CD/es-ICE flow field solver. Tuning parameters are identified and their impact on the result is addressed. Based on the motivation of decreasing the computational cost for multi-cycle LES and thus bringing such calculations closer to industrial applications, different meshing strategies are subsequently investigated. These include two grid resolutions, and in particular the impact of local refinements in the spark plug vicinity since the predictions showed a high sensitivity of the maximum in-cylinder pressure on the early stage of combustion. Although questionable from an LES fundamental perspective, the impact of employing only one half vs. a full cylinder representation has also been explored. The predictions have been validated by means of pressure and heat release evolutions and in general, good agreement with the experiment in terms of the fluctuation bandwidth has been found. However, the extreme cycles are over predicted by the model resulting in an increased standard deviation of the pressure trace. The dependency of the mean pressure trace on the mesh representation using the Damköhler flame speed close has been illustrated. In addition statistical methods have been applied to characterize the observed cycle to cycle variations (CCV) in the simulation and experiment. The highly resolved results of the LES are probed for correlations between flow variables such as velocity magnitude, subgrid turbulent kinetic energy as well as temperature and residual gas and characteristic combustion parameters. Thus local dependencies have been pointed out which primarily influence the early combustion stage.