Accurately predicting the performance of coronagraphs and tolerancing optical surfaces for high-contrast imaging requires a detailed accounting of diffraction effects. Unlike simple Fraunhofer diffraction modeling, near and farfield diffraction effects, such as the Talbot effect, are captured by plane-to-plane propagation using Fresnel and angular spectrum propagation. This approach requires a sequence of computationally intensive Fourier transforms and quadratic phase functions, which limit the design and aberration sensitivity parameter space which can be explored at high-fidelity in the course of coronagraph design. This study presents the results of optimizing the multi-surface propagation module of the open source Physical Optics Propagation in PYthon (POPPY) package. This optimization was performed by implementing and benchmarking Fourier transforms and array operations on graphics processing units, as well as optimizing multithreaded numerical calculations using the NumExpr python library where appropriate, to speed the end-to-end simulation of observatory and coronagraph optical systems. Using realistic systems, this study demonstrates a greater than five-fold decrease in wall-clock runtime over POPPY's previous implementation and describes opportunities for further improvements in diffraction modeling performance.