TIMES.txt

C (mine)
4m44s to get to 95.26% accuracy on validation (6 epochs)
-Ofast 24s to get 95% accuracy
-O3 48s to get 95.6% accuracy
-O2 59s to get 95.8% accuracy

Tensorflow
12s to get to 97.5% accuracy on validation (6 epochs)
I think its multithreaded.

TODO:
    implement multithreading
    implement gpu support


Mine on MacOS: 7.25s user 0.08s system 98% cpu 7.439 total - 0.9814 E6, 0.9862 Peak
TF First Run on MacOS: 11.83s user 2.80s system 86% cpu 16.965 total - 0.9742 E6, same peak
TF Second Run on MacOS: 10.79s user 3.03s system 251% cpu 5.491 total - 0.9741 E6, same peak
TF Second Run tells me that it uses multithreading, so mine is actually faster in terms of CPU time, and startup time from TF First Run.
I am also more accurate, but for some reason my accuracy falls off in last epoch. The same is true for any n epochs, where the accuracy peaks in n-1 epoch.