Apple M3 Ultra (80-core GPU, 512GB) × Llama 4 Scout (17B active, 16 experts)
Can the Apple M3 Ultra (80-core GPU, 512GB) run Llama 4 Scout (17B active, 16 experts)?
Yes
Yes. The Apple M3 Ultra (80-core GPU, 512GB) can run Llama 4 Scout (17B active, 16 experts) at up to F16 (needs ~227 GB of 384 GB usable), at an estimated ~20.7 tokens/sec at Q4_K_M.
Speed is estimated from memory bandwidth calibrated with a measured 92.14 tok/s Llama 2 7B Q4_0 result (source). No direct public benchmark of this exact pair is recorded yet.
By quantization (8K context)
| Quant | Memory needed | Fits? | Est. tok/s |
|---|---|---|---|
| Q3_K_M | 57.1 GB | ✅ | ~25.4 |
| Q4_K_M | 70.5 GB | ✅ | ~20.7 |
| Q5_K_M | 82.4 GB | ✅ | ~17.7 |
| Q6_K | 95.0 GB | ✅ | ~15.3 |
| Q8_0 | 122 GB | ✅ | ~12.0 |
| F16 | 227 GB | ✅ | ~6.4 |
Fastest hardware for Llama 4 Scout (17B active, 16 experts)
- Apple M5 Max (40-core GPU, 128GB) — ~26.9 tok/s at Q4_K_M
- Apple M2 Ultra (76-core GPU, 192GB) — ~21.1 tok/s at Q4_K_M
- Apple M4 Max (40-core GPU, 128GB) — ~18.6 tok/s at Q4_K_M
- NVIDIA DGX Spark — ~12.8 tok/s at Q4_K_M
No suitable GPU? Rent one by the hour on Vast.ai or RunPod (affiliate links).
Get new models & benchmarks in your inbox
No spam. Unsubscribe anytime by replying to any email.