The computation of large systems' atomistic dynamics is required in fields such as biochemistry, electrochemistry and many others. Fully quantum molecular dynamics is a powerful tool, but can have a high computational cost. An approach that was developed in the last decade is to use ML algorithms to build on the fly computationally cheap predictors for the energy, forces, and other physical properties. This approach enables the performance of calculations with an accuracy that is close enough to fully quantum molecular dynamics but with running speeds that are more than 100 times faster. We describe and analyze the construction and use of a DNN based model for the forces in solids.