[dcrazy]

How does CPU-based inference compare to GPU-based inference, performance-wise? And aren’t these machines likely to be used for training?