Best hardware for local LLMs by use case

Every list below is ranked from the same measured benchmark table, with the ranking rule stated. Nothing is hand-picked: the order changes when the data changes.

• Cheapest hardware to run a local LLM

  6 machines qualify · Ranked by cost per million tokens (power + hardware life)

• Best local LLM hardware under $2000

  4 machines qualify · Full rig price of $2000 or less, cheapest first

• Lowest-power local LLM machines

  6 machines qualify · Ranked by wall power draw, lowest first

• Apple Silicon for local LLMs

  9 machines qualify · Apple Silicon only, ranked by cost per million tokens

Ranking rules and assumptions: how we measure