2x RTX 3090 + EPYC 7663
4 records across 1 models (quants: UD-IQ4_XS).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2x RTX 3090 + EPYC 7663 | Qwen3.8-Flash-Next 177B | UD-IQ4_XS | llama.cpp (CUDA) | 65.00 | – | – | – | 81920 | 2026-10-05 | community | – | source |
| 2x RTX 3090 + EPYC 7663 | Qwen3.8-Flash-Next 177B | UD-IQ4_XS | llama.cpp (CUDA) | 62.02 | 73.97 | – | – | 262144 | 2026-10-05 | community | – | source |
| 2x RTX 3090 + EPYC 7663 | Qwen3.8-Flash-Next 177B | UD-IQ4_XS | llama.cpp (CUDA) | 41.01 | 30.83 | – | – | 32768 | 2026-10-05 | community | – | source |
| 2x RTX 3090 + EPYC 7663 | Qwen3.8-Flash-Next 177B | UD-IQ4_XS | llama.cpp (CUDA) | 39.55 | – | – | – | 32768 | 2026-10-05 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085