2× TU106
15 records across 1 models (quants: IQ3_S).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 698 | – | – | 8000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 439 | – | – | 8000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 657 | – | – | 10000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 683 | – | – | 10000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 658 | – | – | 16000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 664 | – | – | 16000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 544 | – | – | 32000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 650 | – | – | 32000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 255 | – | – | 230000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 384 | – | – | 230000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 699 | – | – | 8000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 664 | – | – | 16000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 650 | – | – | 32000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 387 | – | – | 230000 | 2026-09 | community | – | source |
| 2× TU106 | Qwen3.8-27B | IQ3_S | llama.cpp (CUDA) | – | 385 | – | – | 230000 | 2026-09 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085