The fast development of Graphics processing units (GPUs) and the software libraries needed for their usage in scientific computing allows highly efficient calculations using not only a single but also multiple GPUs per task. One of the fields in which graphic cards are widely used is molecular dynamics (MD). By including a large number of GPUs in a single MD simulation, systems with the size of several millions of atoms can be simulated. However, the most common sizes of systems studied via MD rarely exceed 300 k atoms. The computational resources of a single GPU might not be fully occupied by such systems, making the use of further graphic cards inefficient. To estimate at which system size the resources of a given GPU are saturated, in this work, we studied the performance of data center-class graphic cards from different vendors as a function of the number of atoms in the simulated system. We found out that for all graphic cards, the resource saturation is reached at system sizes of approximately 250 k atoms, confirming the initial assumption that the computational capabilities of a single GPU are sufficient for the most commonly studied system sizes.