Apple M4 Pro (20-core GPU) × Meta Llama 3.1 8B Instruct
Meta Llama 3.1 8B Instruct on Apple M4 Pro (20-core GPU)
34.2best gen tok/s
369best prompt tok/s
640max context tested
17results
Measured · all numbers below come from the linked sources
llama.cpp
| Quant | Gen tok/s | Prompt tok/s | Source | Context tested | TTFT | Build |
|---|---|---|---|---|---|---|
| Q4_K - Medium | 34.2 | 367 | LocalScore ↗ | 512+128 | 3,386 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.1 | 366 | LocalScore ↗ | 512+128 | 3,381 ms | llamafile 0.9.2 |
| Q4_K - Medium | 33.0 | 367 | LocalScore ↗ | 512+128 | 3,371 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.9 | 362 | LocalScore ↗ | 512+128 | 3,421 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.8 | 369 | LocalScore ↗ | 512+128 | 3,383 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.8 | 366 | LocalScore ↗ | 512+128 | 3,411 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.7 | 351 | LocalScore ↗ | 512+128 | 3,512 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.7 | 363 | LocalScore ↗ | 512+128 | 3,441 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.7 | 352 | LocalScore ↗ | 512+128 | 3,510 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.7 | 360 | LocalScore ↗ | 512+128 | 3,423 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.6 | 364 | LocalScore ↗ | 512+128 | 3,431 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.6 | 362 | LocalScore ↗ | 512+128 | 3,405 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.5 | 357 | LocalScore ↗ | 512+128 | 3,488 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.5 | 358 | LocalScore ↗ | 512+128 | 3,406 ms | llamafile 0.9.2 |
| Q4_K - Medium | 32.1 | 358 | LocalScore ↗ | 512+128 | 3,492 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.8 | 329 | LocalScore ↗ | 512+128 | 3,794 ms | llamafile 0.9.2 |
| Q4_K - Medium | 31.6 | 351 | LocalScore ↗ | 512+128 | 3,445 ms | llamafile 0.9.2 |
Ollama
No measured Ollama results yet. Submit yours →
MLX
No measured MLX results yet. Submit yours →
"Context tested" is the largest prompt+generation length actually benchmarked (tokens), not a memory-fit estimate.
What else can Apple M4 Pro (20-core GPU) run? →
Includes results from LocalScore, a Mozilla Builders project by cjpais, mirrored in the cn0303/local-llm-speed-benchmarks dataset. LocalScore runs use llamafile, which is built on llama.cpp.