RTX 5070 Ti
21 records across 5 models (quants: MXFP4, Q4_0, Q4_K, Q4_K_M).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 5070 Ti | gpt-oss 20B 20B | MXFP4 | llama.cpp | 156.0 | 6178.3 | – | – | 4096 | – | sourced | – | source |
| RTX 5070 Ti | gpt-oss 20B 20B | MXFP4 | llama.cpp | 133.1 | 4940.7 | – | – | 16384 | – | sourced | – | source |
| RTX 5070 Ti | gpt-oss 20B 20B | MXFP4 | llama.cpp | 115.9 | 3903.5 | – | – | 32768 | – | sourced | – | source |
| RTX 5070 Ti | gpt-oss 20B 20B | MXFP4 | llama.cpp | 91.3 | 2644.9 | – | – | 65536 | – | sourced | – | source |
| RTX 5070 Ti | gpt-oss 20B 20B | MXFP4 | llama.cpp | 66.6 | 1363.7 | – | – | 131072 | – | sourced | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (CUDA) | 168.2 | 7802.37 | – | – | tg128 | 2025-05-10 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (CUDA) | 153.55 | 7623.55 | – | – | tg128 | 2025-05-23 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (CUDA) | 149.28 | 6870.08 | – | – | tg128 | 2025-05-23 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (Vulkan) | 135.63 | 6213.63 | – | – | tg128 | 2025-05-23 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (Vulkan) | 133.94 | 6614.86 | – | – | tg128 | 2025-05-23 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (Vulkan) | 126.67 | 5164.6 | – | – | tg128 | 2025-05-10 | community | – | source |
| RTX 5070 Ti | llama 7B | Q4_0 | llama.cpp (Vulkan) | 125.89 | 6801.97 | – | – | tg128 | 2025-05-12 | community | – | source |
| RTX 5070 Ti | Qwen3 14B 14B | Q4_K | llama.cpp | 74.3 | 3264.7 | – | – | 4096 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 14B 14B | Q4_K | llama.cpp | 58.0 | 2303.9 | – | – | 16384 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 14B 14B | Q4_K | llama.cpp | 45.5 | 1658.2 | – | – | 32768 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 8B 8B | Q4_K | llama.cpp | 120.5 | 5557.3 | – | – | 4096 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 8B 8B | Q4_K | llama.cpp | 87.5 | 3653.8 | – | – | 16384 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 8B 8B | Q4_K | llama.cpp | 63.3 | 2269.0 | – | – | 32768 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3 8B 8B | Q4_K | llama.cpp | 40.4 | 1078.7 | – | – | 65536 | – | sourced | – | source |
| RTX 5070 Ti | Qwen3.6-35B-A3B 35B-A3B | Q4_K_M | llama.cpp (CUDA) | 101.0 | 1265.61 | – | – | – | 2026-08 | community | – | source |
| RTX 5070 Ti | Qwen3.6-35B-A3B 35B-A3B | Q4_K_M | llama.cpp (CUDA) | 87.92 | 923.45 | – | – | – | 2026-08 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085