6x RTX 3090
12 records across 1 models (quants: UD-Q4_K_XL).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | 41.9 | – | – | – | 51200 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | 33.7 | – | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | 27.3 | – | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | 25.7 | – | – | – | 51200 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | 10.2 | – | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | 2.3 | – | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | – | 568 | – | – | 51200 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | – | 2127 | – | – | 51200 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | – | 263 | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | – | 2111 | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (CUDA) | – | 265 | – | – | 256000 | 2026-10-06 | community | – | source |
| 6x RTX 3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp (multigpu fork) | – | 2060 | – | – | 256000 | 2026-10-06 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085