RTX 3070
6 records across 2 models (quants: Q4_0, unspecified).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 3070 | llama 7B | Q4_0 | llama.cpp (CUDA) | 84.62 | 2993.73 | – | – | tg128 | 2025-05-29 | community | – | source |
| RTX 3070 | llama 7B | Q4_0 | llama.cpp (Vulkan) | 78.71 | 2113.02 | – | – | tg128 | 2025-05-29 | community | – | source |
| RTX 3070 | llama 7B | Q4_0 | llama.cpp (Vulkan) | 63.64 | 1689.4 | – | – | tg128 | 2025-10-20 | community | – | source |
| RTX 3070 | llama 7B | Q4_0 | llama.cpp (Vulkan) | 62.92 | 1832.07 | – | – | tg128 | 2025-10-20 | community | – | source |
| RTX 3070 | Qwen3-0.6B | unspecified | vllm | 1676 | – | – | – | 512 | 2026-09-30 | community | – | source |
| RTX 3070 | Qwen3-0.6B | unspecified | vllm | 1671 | – | – | – | 512 | 2026-09-30 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085