Slow GPU, underutilized in tensorflow 2.1

I'm learning my own game using pygame in python on windows with GPU, but it's not as fast as I thought, and when I look at GPU utilization in Task Manager, I don't know if it's used properly at 3-5% all the time.

The GPU itself is not using a good one, so I don't expect that speed, but I would like to check if the GPU is being used properly.
In addition, the memory you are learning is more than 90% of the time, according to Task Manager.(You can learn to the end without crashing.)

The learning environment is running DQN using keras and Keras-rl2 in the tensorflow in python3.
The model of learning is not that deep, as it was referenced by Atari's CNN in deepmind.
The following is the running environment:

------- Environment -------
Windows 10
Intel Core i7 16GB RAM NVIDIA GeForce GTX 1050TI CUDA 10.1
cudnn 7.6.5

Python 3.7.6
tensorflow 2.1.0
Keras-rl2 1.0.4
---------------

Is there a way to verify that this GPU is being used properly?
Also, is low utilization normal?
If there is a setting such as limiting GPU performance, I would like to know how to remove it

Please