Honest question (as a person outside of AI industry): I understand the title "M5 Ultra Mac Studio Review: The Dream Mac for Local AI Agents" requires ... but if

Honest question (as a person outside of AI industry): I understand the title "M5 Ultra Mac Studio Review: The Dream Mac for Local AI Agents" requires ... but if those LLM benchmarks are so settings-sensitive, why not to dedicate at least half of the article to some general benchmarks? Maybe ultra-large matrix/tensor/N-body calculations, some memory bandwidth measurements, which all contribute to LLM performance, but which allow to compare more-or-less apples to apples for those different devices (M3 Ultra, M5 Ultra, RTX 5090, et al.)