What would be much more interesting is a comparison of Apples frameworks for numerical processing. Like Accelerate. Those do not just use SIMD but actually take

What would be much more interesting is a comparison of Apples frameworks for numerical processing. Like Accelerate. Those do not just use SIMD but actually take advantage of all the additional things they put in their SoC. What we are looking at now is some specific benchmark for a random library that who knows how makes use of things.