localLLMriglocal, on your table

landing page

Best Mac for a local LLM

On a Mac, unified memory is the limit. What does not fit in that pool does not fit anywhere. Bandwidth decides how fast the tokens arrive.

A Mac has no separate VRAM. The model, the KV cache and macOS share the same memory. We therefore leave several GB free for the system. What remains is what the model may use.

16 GB is an 8B machine. 32–36 GB is 14B and 32B Q4 territory. From 64 GB, 70B Q4 comes into view. FP16 of a 70B needs a much larger pool, in this list the M5 Ultra with 256 GB.

The M4 mini and the M4 Max Studio are here as the previous generation, partly with an estimated bandwidth because we did not re-open that product page. The M5 and M6 memory and bandwidth figures come from Apple's spec pages as of October 2026. The euro amounts are an estimate throughout.

Open in the calculator

Three setups

Llama 3.3 70B · Q4 · 8,192 ctx

Estimated memory

43.0 GBestimate

39.5 GB weights · 2.7 GB KV cache · 0.8 GB overhead

  1. 01 · Fastest

    Mac Studio M5 Ultra 96GB

    High bandwidth and enough memory for 70B Q4/Q5. FP16 70B does not fit.

    $7,266€6,490

    estimate

    tok/s

    13.2

    estimate

    needed

    43.0 GB

    full fit

    usable

    84.0 GB

    unified

    Why this one

    • The model fits entirely in VRAM or unified memory. No CPU offload.
    • Memory left for a longer context or a heavier quantisation.
    • Unified memory: weights and KV cache share one pool.
    • High memory bandwidth, so a higher tokens-per-second estimate.
    • One box, little noise, no discrete GPU.
    • Highest estimated speed in this shortlist.

    Parts

    • Mac Studio M5 Ultra, 96 GB, 1.2 TB/s

  2. 02 · A step up

    Mac Studio M5 Ultra 256GB

    The machine in this list that can still hold 70B in FP16. The price is a rough estimate.

    $11,744€10,490

    estimate

    tok/s

    13.2

    estimate

    needed

    43.0 GB

    full fit

    usable

    244.0 GB

    unified

    Why this one

    • The model fits entirely in VRAM or unified memory. No CPU offload.
    • Memory left for a longer context or a heavier quantisation.
    • Unified memory: weights and KV cache share one pool.
    • High memory bandwidth, so a higher tokens-per-second estimate.
    • One box, little noise, no discrete GPU.

    Parts

    • Mac Studio M5 Ultra, 256 GB unified memory, 1.2 TB/s

  3. 03 · A step up

    Mac Studio M5 Max 128GB

    70B up to Q8 has room here. Bandwidth still trails a dual high-end GPU.

    $5,810€5,190

    estimate

    tok/s

    6.9

    estimate

    needed

    43.0 GB

    full fit

    usable

    116.0 GB

    unified

    Why this one

    • The model fits entirely in VRAM or unified memory. No CPU offload.
    • Memory left for a longer context or a heavier quantisation.
    • Unified memory: weights and KV cache share one pool.
    • One box, little noise, no discrete GPU.

    Parts

    • Mac Studio M5 Max 18-core/40-core GPU, 128 GB, 614 GB/s

Tokens per second are estimated from memory bandwidth, not from a measured benchmark. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop source. Run llama-bench yourself before you buy.

Want a build tailored to you? Custom report

Get new builds when GPU prices move

End of results

Best Mac for a local LLM · localLLMrig