varying numerical output from 2 identical tensorflow CNN models

Question

I have a model similar to [this][1]

[1]: https://github.com/Kulbear/deep-learning-coursera/blob/master/Convolutional%20Neural%20Networks/Convolution%20model%20-%20App