5x RTX3090
8 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 56.5 | – | – | – | – | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 54.4 | – | – | – | – | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 50.8 | – | – | – | 42000 | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 48.0 | 716.0 | – | – | 111000 | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 41.8 | – | – | – | 250000 | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 37.2 | – | – | – | 42000 | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 22.5 | 609.0 | – | – | 111000 | 2026-09 | community | – | source |
| 5x RTX3090 | Qwen3.8-Flash-Next | UD-Q4_K_XL | llama.cpp | 11 | – | – | – | 250000 | 2026-09 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085