Yesterday changed my backend to tensorflow and noticed a bunch of warning like “The TensorFlow library wasn’t compiled to use SSE..”

Hm, I don’t usually run computation on local machine, I use GPU on Google Cloud, but its always nice to spend less time while doing quick model analysis. Whole [compilation process][compilation process] did take about 15 minutes, quite easy. I didn’t make any complex performance tests, but here is my VGG16 quick run example:

Before:

1409/1409 [==================] - 627s
368/368 [====================] - 171s

After:

1409/1409 [==================] - 361s
368/368 [====================] - 91s

It’s works 2x times faster now. For free! :)