Gemma 3 on Apple M4 Pro · 48GB
✅ Runs (tight) — Gemma 3 27B @ Q8_0
Gemma 3 27B at Q8_0 needs ~32.4 GB (weights 27.1 GB + KV 3.9 GB + overhead 1.4 GB @ 8K ctx) of your 33.0 GB usable (of 48.0 GB unified memory) — fits, but little headroom — close other apps or trim context. Expect ~4–7 tok/s (slow).
context @ Q8_0: runs to 16K · won't fit past that on this rig
| quant | needs | speed |
|---|---|---|
| ✓ Q8_0 | 32.4 GB | ~4–7 tok/s (slow) |
| ✓ Q5_K_M | 23.0 GB | ~6–10 tok/s (usable) |
| ✓ Q4_K_M | 20.1 GB | ~7–12 tok/s (usable) |
| ✓ Q3_K_M | 17.0 GB | ~8–14 tok/s (usable) |
$runlocal install gemma3:27b
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 M3 Max
Also runs on a Apple M4 Pro: Qwen3 · DeepSeek-R1 (distill) · Qwen2.5 · Mixtral 8x7B · Mistral Small 3
Best models for a Apple M4 Pro · 48GB →
Architecturally similar
These share Gemma 3's KV-cache geometry — they size memory the same way, so here's how they fit on a Apple M4 Pro · 48GB too:
| model | shares | verdict here |
|---|---|---|
| Gemma 4 | 8kv/256hd | ✅ Runs comfortably |
| Gemma 2 | 16kv/128hd4kv/256hd8kv/256hd | ✅ Runs (tight) |
| Falcon3 7B | 4kv/256hd | ✅ Runs comfortably |
Check any combo yourself: open the checker →