For the purpose of a deeper understanding of thin film growth, in the last two decades several groups developed models for simulation on the atomistic scale. Models using molecular dynamics as their simulation method already give results comparable to experiments, however statistical analysis of the simulations themselves are lacking so far, reasoned by the limits imposed by the computational power and parallelization that can only be used in lateral dimensions. With advancements of software and hardware, an increase in simulation speed by a factor of up to 10 can be reached. This allows either larger structures and/or more throughput of the simulations. The paper analyses the significance of increasing the structure size in lateral dimensions and also the repetition of simulations to gain more insights into the statistical fluctuation contained in the simulations and how well the coincidence with the experiment is. For that, glancing angle incidence deposition (GLAD) coatings are taken as an example. The results give important insights regarding the used interaction potential, the structure size and resulting important differences for the density, surface morphology, roughness and anisotropy. While larger structures naturally can reproduce the real world in more detail, the results show which structure sizes are needed for these aspects without wasting computational resources.