Sorry for the mess, just getting started.
I've written several Neural Network applications over the years my first being back in 1992, based on the Scientific American aritcle "How Neural Networks Learn from Experience" which gave a brief outline of the Back-Propagation Algorithm. This first progam was written in C and I was probably running on a 386 class processor. No artifacts of the original code or test results survive today.
My most recent NN framework is written in C# and I'm running on an AMD Ryzen 7 at 3.6 GHz. I recently tested the training ability on the MNIST dataset and so far my best result is 2.8% accuracy using two hidden layers and TANH in 4.55 hours. I has really just testing my NN design to prove that it was successfully training and a bit surprised at the result.