<p>One of the major problems of the investigation of the flow field in internal combustion engines is the analysis of the impact of cyclic variations on the overall flow structure. Implicit large-eddy simulations (LES) are conducted to quantitatively analyze cold in-cylinder flow with associated cyclic variations in an internal combustion engine. To validate the simulation results, a comparison with stereoscopic PIV measurements in the tumble plane is performed and a grid convergence study is conducted in which five consecutive engine cycles are simulated using meshes ranging from 4.2 to 212 million cells. To quantitatively assess the agreement with the experimental data, weighted relevance and magnitude indices are evaluated for the instantaneous and ensemble averaged velocity vector during the intake and compression strokes. The influence of the number of simulated intake and compression cycles on the results is assessed. The grid convergence study demonstrates that the tumble spin and breakup processes are not accurately resolved by the coarse mesh, although the single-grid estimator suggests sufficient mesh resolution. Overall, satisfactory agreement with experimental data is achieved with sufficient mesh resolution. To analyze cyclic variations, a 2-D noise-assisted multivariate EMD method is applied to the experimental and numerical results. A novel mode separation of the velocity scales, which utilizes the identical decomposition of the ensemble averaged and instantaneous cycle data, is presented. This method simplifies the extraction of the tumble vortex core, enables the estimation of the ensemble mean flow and the analysis of cycle-to-cycle variations despite the limited number of available simulation cycles. The EMD analysis reveals a significant cyclic variance in the position of the tumble vortex core at crank angles between 90 to 120 degrees and 180 to 200 degrees. A correlation between the vortex core position and vortex strength at top dead center is established.</p>