Cosmological gravitational microlensing is a useful technique for understanding the structure of the inner parts of a quasar, especially the accretion disk and the central supermassive black hole. So far, most of the cosmological microlensing studies have focused on single objects from the ~90 currently known lensed quasars. However, present and future planned all-sky surveys (e.g. LSST) are expected to discover thousands of new lensed systems. The advent of graphics processing units (GPUs)for general computation in astronomy has enabled the theoretical study of tens of thousands of microlensing magnification patterns. I examine the systematic properties of magnification maps at high resolution ($4092^2$) across the entire convergence ($\kappa$) and shear ($\gamma$) parameter space of interest to microlensing. I present a number of statistical properties of the magnification distribution averaged among 15 different sets of lens positions per $\kappa,\gamma$ point. Finally, I give an update on the status of the GPU-Enabled High-Resolution Micro-Lensing parameter survey (GERLUMPH) and results of the first applications to quasar accretion disks. All of the simulations have been carried out on gSTAR, the new GPU supercomputer at Swinburne University.