Can I Run Llama 4 Scout 17B on Apple M3 Ultra?
作者： Jakub Rusinowski · 最后更新： 2026年7月12日
Yes. Llama 4 Scout 17B needs 68.2 GB at 8K context (65.8 GB weights + 2.4 GB KV cache/overhead) at Q4_K_M and the Apple M3 Ultra has 384 GB usable, leaving about 315.8 GB of headroom, at roughly 36 tok/s (estimated) with room for up to 1M of context.
See what else this hardware can run →
The numbers
| Model | Llama 4 Scout 17B |
| Parameters | 17B active / 109B total |
| Quantization | Q4_K_M |
| VRAM needed (8K context) | 68.2 GB |
| Apple M3 Ultra usable memory | 384 GB |
| Verdict | Runs |
| Max context | 1M |
| Estimated speed | ~36 tok/s |
Llama 4 Scout 17B on the Apple M3 Ultra: VRAM by quantization
| Quant | VRAM needed | Fits 512 GB? | Max context | Whole PC |
|---|---|---|---|---|
| F16 | 220.4 GB | ✓ Yes | 512K | — |
| Q8_0 | 118.2 GB | ✓ Yes | 1M | — |
| Q6_K | 91.8 GB | ✓ Yes | 1M | — |
| Q5_K_M | 79.7 GB | ✓ Yes | 1M | — |
| Q4_K_M | 68.2 GB | ✓ Yes | 1M | — |
| Q3_K_M | 48.9 GB | ✓ Yes | 1M | — |
| Q2_K | 38.2 GB | ✓ Yes | 1M | — |
Assumes an 8K-token context with an f16 KV cache. A longer window needs more; a quantized KV cache needs less. “Max context” is the largest window that still fits in 512 GB of unified memory. Figures are estimates from parameter count, quantization and memory bandwidth — the analyzer lets you tune KV-cache quant and context.
Context costs VRAM too. At Q4_K_M on the Apple M3 Ultra: 8K 68.2 GB · 32K 73.1 GB · 128K 92.4 GB. Every window this model publishes still fits 512 GB.
What to do next
Nothing to change — run it at Q4_K_M. If you want more capability, check whether the next size up also fits.
No compatible GPU? Llama 4 Scout 17B on 48 GB of system RAM, CPU only: runs, but it is tight.
Running Llama 4 Scout 17B on this card:
Llama 4 Scout 17B needs ~68 GB but this GPU has 512 GB. Rent a A100 (80 GB)-class GPU by the hour instead of buying one:
Affiliate links — we may earn a commission if you sign up, at no extra cost to you.
Cloud rates verified 2026-07 — estimates, and marketplace prices vary. Buying price is GPU MSRP only, not a full PC.
Other Llama 4 sizes on the Apple M3 Ultra
- Llama 4 Maverick 17B — runs (243.9 GB)
Llama 4 Scout 17B on nearby hardware
- AMD Ryzen AI Max+ 395 (128 GB) — runs
All Llama 4 sizes on the Apple M3 Ultra | Llama 4 Scout 17B model page | Apple M3 Ultra GPU page | VRAM calculator | Check your hardware