There is a long period of idle GPU time in tensorflow timeline as follows:
The GPU is only used by the model training process, not shared with other process. The sys