RTX 3090 × Llama 3.2 1B Instruct
Llama 3.2 1B Instruct on RTX 3090
377best gen tok/s
15,607best prompt tok/s
4,352max context tested
32results
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 | 377 | 15,607 | LocalScore ↗ | 4,352 | 84 ms | llamafile 0.9.2 |
| Q4_K - Medium | 368 | 15,252 | LocalScore ↗ | 512+128 | 88 ms | llamafile 0.9.2 |
| Q4_K - Medium | 367 | 15,400 | LocalScore ↗ | 4,352 | 86 ms | llamafile 0.9.2 |
| Q4_K - Medium | 361 | 14,974 | LocalScore ↗ | 512+128 | 90 ms | llamafile 0.9.2 |
| Q4_K - Medium | 360 | 15,064 | LocalScore ↗ | 512+128 | 89 ms | llamafile 0.9.2 |
| Q4_K - Medium | 350 | 14,749 | LocalScore ↗ | 4,352 | 93 ms | llamafile 0.9.2 |
| Q4_K - Medium | 348 | 14,452 | LocalScore ↗ | 4,352 | 92 ms | llamafile 0.9.2 |
| Q4_K - Medium | 342 | 13,768 | LocalScore ↗ | 4,352 | 98 ms | llamafile 0.9.2 |
| Q4_K - Medium | 342 | 14,411 | LocalScore ↗ | 512+128 | 93 ms | llamafile 0.9.2 |
| Q4_K - Medium | 340 | 13,805 | LocalScore ↗ | 512+128 | 95 ms | llamafile 0.9.2 |
| Q4_K - Medium | 339 | 14,601 | LocalScore ↗ | 512+128 | 91 ms | llamafile 0.9.2 |
| Q4_K - Medium | 331 | 13,990 | LocalScore ↗ | 512+128 | 96 ms | llamafile 0.9.2 |
| Q4_K - Medium | 330 | 14,061 | LocalScore ↗ | 512+128 | 95 ms | llamafile 0.9.1 |
| Q4_K - Medium | 328 | 14,169 | LocalScore ↗ | 512+128 | 95 ms | llamafile 0.9.2 |
| Q4_K - Medium | 321 | 13,357 | LocalScore ↗ | 512+128 | 99 ms | llamafile 0.9.2 |
| Q4_K - Medium | 319 | 13,125 | LocalScore ↗ | 512+128 | 99 ms | llamafile 0.9.2 |
| Q4_K - Medium | 308 | 12,344 | LocalScore ↗ | 512+128 | 106 ms | llamafile 0.9.2 |
| Q4_K - Medium | 301 | 12,964 | LocalScore ↗ | 512+128 | 105 ms | llamafile 0.9.2 |
| Q6_K | 300 | 13,357 | LocalScore ↗ | 512+128 | 100 ms | llamafile 0.9.2 |
| Q4_K - Medium | 276 | 12,950 | LocalScore ↗ | 512+128 | 99 ms | llamafile 0.9.3 |
| Q4_K - Medium | 216 | 12,560 | LocalScore ↗ | 4,352 | 108 ms | llamafile 0.9.3 |
| Q4_K - Medium | 208 | 12,930 | LocalScore ↗ | 4,352 | 108 ms | llamafile 0.9.3 |
| Q4_K - Medium | 194 | 13,974 | LocalScore ↗ | 512+128 | 97 ms | llamafile 0.9.2 |
| Q4_K - Medium | 193 | 14,107 | LocalScore ↗ | 512+128 | 94 ms | llamafile 0.9.2 |
| Q4_K - Medium | 186 | 13,196 | LocalScore ↗ | 512+128 | 102 ms | llamafile 0.9.2 |
| Q4_K - Medium | 176 | 12,960 | LocalScore ↗ | 512+128 | 104 ms | llamafile 0.9.3 |
| Q4_K - Medium | 151 | 11,382 | LocalScore ↗ | 512+128 | 118 ms | llamafile 0.9.3 |
| Q4_K - Medium | 115 | 11,478 | LocalScore ↗ | 512+128 | 115 ms | llamafile 0.9.2 |
| Q4_K - Medium | 114 | 11,568 | LocalScore ↗ | 512+128 | 114 ms | llamafile 0.9.2 |
| Q4_K - Medium | 112 | 12,246 | LocalScore ↗ | 512+128 | 112 ms | llamafile 0.9.2 |
| Q4_K - Medium | 105 | 11,674 | LocalScore ↗ | 512+128 | 121 ms | llamafile 0.9.2 |
| Q4_K - Medium | 45.3 | 8,635 | LocalScore ↗ | 4,352 | 155 ms | llamafile 0.9.3 |
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.
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.