RTX 4070 Ti SUPER × Meta Llama 3.1 8B Instruct
Meta Llama 3.1 8B Instruct on RTX 4070 Ti SUPER
94.5best gen tok/s
4,546best prompt tok/s
4,352max context tested
16results
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 | 94.5 | 4,546 | LocalScore ↗ | 512+128 | 291 ms | llamafile 0.9.2 |
| Q4_K - Medium | 82.5 | 4,285 | LocalScore ↗ | 512+128 | 305 ms | llamafile 0.9.2 |
| Q4_K - Medium | 66.3 | 4,258 | LocalScore ↗ | 4,352 | 307 ms | llamafile 0.9.3 |
| Q4_K - Medium | 55.7 | 3,834 | LocalScore ↗ | 512+128 | 352 ms | llamafile 0.9.2 |
| Q4_K - Medium | 49.9 | 3,833 | LocalScore ↗ | 512+128 | 332 ms | llamafile 0.9.2 |
| Q4_K - Medium | 49.1 | 3,366 | LocalScore ↗ | 512+128 | 381 ms | llamafile 0.9.2 |
| Q4_K - Medium | 49.1 | 3,661 | LocalScore ↗ | 512+128 | 356 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.8 | 3,305 | LocalScore ↗ | 512+128 | 383 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.8 | 3,342 | LocalScore ↗ | 512+128 | 385 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.3 | 3,239 | LocalScore ↗ | 512+128 | 389 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.2 | 3,231 | LocalScore ↗ | 512+128 | 391 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.1 | 3,061 | LocalScore ↗ | 512+128 | 409 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.1 | 3,235 | LocalScore ↗ | 512+128 | 391 ms | llamafile 0.9.2 |
| Q4_K - Medium | 48.0 | 3,240 | LocalScore ↗ | 512+128 | 389 ms | llamafile 0.9.2 |
| Q4_K - Medium | 47.9 | 3,227 | LocalScore ↗ | 512+128 | 390 ms | llamafile 0.9.2 |
| Q4_K - Medium | 29.6 | 3,313 | LocalScore ↗ | 512+128 | 381 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.
What else can RTX 4070 Ti SUPER 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.