Gemma 2 on Apple M4 Max · 128GB
✅ Runs comfortably — Gemma 2 27B @ Q8_0
Gemma 2 27B at Q8_0 needs ~31.2 GB (weights 26.9 GB + KV 2.9 GB + overhead 1.3 GB @ 8K ctx) of your 93.0 GB usable (of 128 GB unified memory) — plenty of headroom. Expect ~7–12 tok/s (usable).
context @ Q8_0: comfortable to 8K
| quant | needs | speed |
|---|---|---|
| ✓ Q8_0 | 31.2 GB | ~7–12 tok/s (usable) |
| ✓ Q5_K_M | 21.8 GB | ~11–18 tok/s (usable) |
| ✓ Q4_K_M | 19.0 GB | ~12–20 tok/s (usable) |
| ✓ Q3_K_M | 15.9 GB | ~14–25 tok/s (usable) |
$runlocal install gemma2:27b
for your own README — links back here
Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro
Also runs on a Apple M4 Max: DeepSeek-R1 (distill) · Qwen3 · Qwen2.5 · Llama 3.1 · Llama 3.3
Best models for a Apple M4 Max · 128GB →
Architecturally similar
These share Gemma 2's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M4 Max · 128GB too:
| model | shares | verdict here |
|---|---|---|
| Gemma 4 | 8kv/256hd | ✅ Runs comfortably |
| Gemma 3 | 16kv/128hd4kv/256hd8kv/256hd | ✅ Runs comfortably |
| Falcon3 7B | 4kv/256hd | ✅ Runs comfortably |
Check any combo yourself: open the checker →