https://towardsdatascience.com/building-efficient-data-pipelines-using-tensorflow-8f647f03b4ce

Having efficient data pipelines is of paramount importance for any machine learning model. In this blog, we will learn how to use TensorFlow’s Dataset module tf.data to build efficient data pipelines.

Most of the introductory articles on TensorFlow would introduce you with the feed_dict method of feeding the data to the model. feed_dict processes the input data in a single thread and while the data is being loaded and processed on CPU, the GPU remains idle and when the GPU is training a batch of data, CPU remains in the idle state. The developers of TensorFlow have advised not to use this method during training or repeated validation of the same datasets.