Note however, that on Intel it's actually slower than run off the mill float32 linear algebra library like Eigen or OpenBLAS. Its main forte seems to be ARM.