I just replied to someone saying it couldn't run on a Windows laptop with how it can. It's below reading rate. Both are below skimming rate that is more useful for LLMs. I would rather have a 96GB macbook than a windows laptop if my only case was local LLMs and I didn't care about all the hoops around sealed system volume, signed app troubles, etc. that are bringing Macbooks closer to ipads/chromebooks, or the many other things that are better with nvidia GPUs.

For desktop I'd still rather threadripper etc. where it has more competitive CPU memory bandwidth and is upgradable and can run multiple GPUs.