The earliest versions of Tensorflow could be built to support the nodes with or without GPUs, but starting from version r1.2 we have to build 2 separate versions of the tensorflow. Is there something we are missing. Is it still possible to build Tensorflow in such a way that would work in both environments - with and without GPUs (without creating 2 separate sets of binary files)?
Please see:
https://www.tensorflow.org/install/pip?lang=python3
`tensorflow` —Latest stable release with CPU and [GPU support]
(https://www.tensorflow.org/install/gpu) *(Ubuntu and Windows)**
Tensorflow 2 is capable of working on both GPUs and CPUs with the same pip install (without 2 sets of binary files) as far as I am aware and from what I can tell with a VERY basic test.
You can compile TensorFlow for GPU support but configure it to run on CPU at runtime as explained in this thread (for both 1.x and 2.x releases) : https://github.com/tensorflow/tensorflow/issues/31135.
If I may ask, is there a reason for doing this ?