I always thought it's Apple's on-package DRAM latency that contributes to its speed relative to x86 especially for local LLM (generative but not necessarily tra

I always thought it's Apple's on-package DRAM latency that contributes to its speed relative to x86 especially for local LLM (generative but not necessarily training) usage but with the answers here I'm not so sure.