In an earlier deep learning article, we talked about how inference workloads—the use of already-trained neural networks to analyze data—can run on fairly cheap hardware, but running the training ...