Yes this is the case, and also the CPU is the one responsible for loading batches and scheduling the GPU, so even if there is no dataprep, the CPU will be used to certain usage, and exactly what operations are performed by deep learning frameworks in the background we dont know that either.