Question about the scale factor layout

Recently I am reading the code of sm100 fp8 GEMM. I have a question:
Why must the scale factor be MN-major? Is it because this provides better performance or is it due to hardware constraints? Thank you for your reply.