Umm, running a 1990s fortran code that's a CFD simulation is a "real" workload. Seems relatively likely that any floating point heavy code would act similarly. Hard to say if it's the matrix multiple or the memory bandwidth that's giving apple such a large lead.

Normally I'd discount using a 28 core Intel CPU from 2019, but from what I can tell Intel hasn't improved much since then. Keep in mind that Intel has a specialized vector unit (AVX256 or AVX512 depending on the model), and the listed CPU is pretty high end (with 6 memory channels) where the normal i5/i7/i9 is only 2.

So sure it's not a video compression, gaming, or web browing benchmark, but some folks do run floating point heavy codes. Unlike CUDA, which requires a rewrite, this code wasn't specifically optimized for the M1.