Large Eddy Simulation (LES) of premixed turbulent combustion in a confined cylinder setup at engine relevant conditions has been carried out for three different initial turbulence intensities, mimicking different flame propagation regimes. Direct Numerical Simulation (DNS) of the setup under investigation provides the reference data to be compared against. The DNS fields have been filtered on the LES grid and are used as initial conditions for the LES at onset of combustion, guaranteeing a direct comparability of the single realizations between the modeled and reference data. Two different combustion models, the G-Equation and LES-CMC premixed are compared with respect to their predictive capabilities as well as their usability and computational cost. While the G-Equation is a widely adopted approach for industrial applications and usually relies on a tunable turbulent flame speed closure, the novel LES-CMC comes as a tuning parameter free model. Different turbulent flame speed closures for the G-Equation have been assessed. Both modeling approaches were in agreement with the reference data. However, LES-CMC leads to improved prediction accuracy for changing turbulence intensities. A sweep in the initial pressure in the LES has been performed in order to investigate numerically the effect of increasing the contribution of the subgrid models by decreasing the resolved part of the flame caused by the decreasing flame thickness for higher pressures.