Deep learning models have been highly successful at breast cancer histopathology image classification, but are computationally expensive and time-consuming to train. We introduce a GPU-accelerated deep learning system that automatically exploits available GPUs as well as the best number of CPU cores, spreading the training workload across all available hardware. Our strategy utilizes parallelization of data loading and augmentation pipelines, and supports multi-GPU training when multiple GPUs are present automatically. These parallelization methods highly speed up training as well as hardware resource utilization, but at no expense to model accuracy. The classification system revolves around a powerful Python implementation of a convolutional neural network (CNN), incorporating a modern React-based user interface for ease of end-user usage. The model has high accuracy in malignant versus benign classification of breast histopathology images, reflecting both technical innovation in accelerating computation as well as significant clinical utility in enhanced diagnostic workflow.