The most robust way to work with the specific version of Tensorflow with GPU support

Question

The project I am working on uses the Tensorflow 2.0, which is not the most recent version of this Deep Learning Framework and I would like to enable GPU support.