This is probably already well known, but if you're using Python and want to exploit SIMD instructions, one of the simpler ways is to reformulate stuff as matrix

This is probably already well known, but if you're using Python and want to exploit SIMD instructions, one of the simpler ways is to reformulate stuff as matrix operations and let numpy handle them. The speed-up can be staggering. TensorFlow also can use these SIMD instructions, just that the standard installation (the one you get with pip) does not [1]. This is why you might see this message when you run it on your machine "Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2". [1] https://stackoverflow.com/questions/47068709/your-cpu-supports-instructions-that-this-tensorflow-binary-was-not-compiled-to-u https://stackoverflow.com/questions/47068709/your-cpu-suppor...