Comment by bitL

They commit patches to OpenBLAS which is still 2-3x slower on Zen 2 than MKL on Intel hardware. Intel has better software engineers for high-performance math it seems.