I’m an AMD fanboy, and just built a new Ryzen box. I’d be curious to see if there is a difference in accuracy running the two kernels on AMD hardware.

Maybe it is faster but less accurate to run the intel kernel on AMD hardware.

There is no difference. The operations these kernels use are all well-defined and work identically on all CPUs that implement them. In general, FP/SIMD is not nearly as much of a crapshoot as people seem to expect it to be. Beyond timings, there are generally no user-space visible differences in operation between any AVX/SSE instructions in Intel/AMD cpus.

It's early days. Maybe they are still working on tuning the kernels?

But why not start with the kernels they already have which already perform better on the target hardware?

And thats exactly the key question. The original kernel that Intel uses on their own CPUs works perfectly fine on AMD. Otherwise Matlab wouldn't have validated it in their production release in 2020a. So the key question is: Why is Intel taking efforts in implementing a specific kernel for AMD that is somewhat slower. I guess there aren't too many answers out there that make sense.

The entire Industry should develop a true interest to push OSS alternatives. And all OSS projects should honestly think twice about implementing such a CSS like the MKL as a standard.