2× 20 GB
9 records across 1 models (quants: IQ1_M main / Q8_0 head).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 70.5 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 67.1 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 61.5 | – | – | – | – | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 59.6 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 58.9 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 55.4 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 50.3 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 49.4 | – | – | – | 65536 | 2026-10-03 | community | – | source |
| 2× 20 GB | Qwen3.8-Flash-Next | IQ1_M main / Q8_0 head | llama.cpp (CUDA) | 41.2 | – | – | – | 65536 | 2026-10-03 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085