Computational models, besides their relatively low cost, offer the benefit of complete control over all testing variables (e.g., atmospheric conditions, loading rates, applied stress states), which can be difficult to control in experiments. This control can be used to identify key controlling parameters and improve our understanding of the deformation and failure processes. This dissertation investigates how modern computational resources can be leveraged to improve both the understanding and prediction of material deformation and failure. Because of its widespread use and variability of application, aluminum and its alloys are the focus of the investigation. Specifically, the applications of ductile failure and grain boundary shear strength were chosen for this dissertation. Though these phenomena are quite different, the same theme is present in the approach to both problems. In both cases, we were able to run numerous simple simulations of the phenomenon in concern. A large computational effort was required in all cases to run the sets of simulations, but the results of these simulations were synthesized into a simple model which can be used at larger scales. In the case of ductile failure, a unit cell was designed to simulate the microstructural evolution of an aluminum alloy. The population of second phase particles in the alloy was represented as a spherical void surrounded by an alu- minum matrix. Many different loadings were applied to the cell which were characterized by a stress state and orientation. The results of these tests were used to form a simple model for the dependence of ductile failure on applied stress state. By refining the model microstructure, it was found that increasing the fidelity of the model microstructure leads to increased predictive capability of the model. In the case of grain boundary shear strength, atomistic models of interface structures were subjected to shear in many directions in the boundary plane. The simulation of a large number of these interface structures showed that shear yield strengths were relatively independent of the macroscopic parameters describing each interface. Subsequently, it was shown that a statistical approach to predicting grain boundary shear strengths could be used.