RTX 5070
6 records across 3 models (quants: Q4_K, UD-IQ3_XXS).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 5070 | Qwen3 14B 14B | Q4_K | llama.cpp | 54.2 | 2144.2 | – | – | 4096 | – | sourced | – | source |
| RTX 5070 | Qwen3 14B 14B | Q4_K | llama.cpp | 40.6 | 1315.2 | – | – | 16384 | – | sourced | – | source |
| RTX 5070 | Qwen3 8B 8B | Q4_K | llama.cpp | 85.8 | 3487.7 | – | – | 4096 | – | sourced | – | source |
| RTX 5070 | Qwen3 8B 8B | Q4_K | llama.cpp | 59.1 | 1600.8 | – | – | 16384 | – | sourced | – | source |
| RTX 5070 | Qwen3 8B 8B | Q4_K | llama.cpp | 43.6 | 898.8 | – | – | 32768 | – | sourced | – | source |
| RTX 5070 | Qwen3.8-Flash-Next | UD-IQ3_XXS | llama.cpp (CUDA) | 11.5 | – | – | – | – | 2026-10-04 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085