Running a model's predictions using a computer's central processor rather than a specialized graphics card, which is slower but requires less specialized hardware.