I recently upgraded to a Mac Studio M2 Ultra with 64GB (the 128GB model was too expensive).
My existing M1 Max 32GB is used for coding and tool work, while the M2 will be used only for models and a few necessities.
Since I'm not running anything else, I can give all the memory except the basic OS to the LLM.
I should have separated my equipment before, but I only recently did. It seems like separating things is always a good idea.
Considering performance, it would be better to just use a paid service, but I plan to run local models with lower performance for a while.
Thinking about Qwen in the past, local models have improved a lot, so I'm looking forward to the future.
