H200 NVL
3 records across 1 models (quants: IQ2_M).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| H200 NVL | Qwen3.8-27B 27B | IQ2_M | llama.cpp (CUDA) | 99.8 | – | – | – | 8192 | 2026-10-05 | community | – | source |
| H200 NVL | Qwen3.8-27B 27B | IQ2_M | llama.cpp (CUDA) | 97.7 | – | – | – | 8192 | 2026-10-05 | community | – | source |
| H200 NVL | Qwen3.8-27B 27B | IQ2_M | llama.cpp (CUDA) | 75.4 | – | – | – | 8192 | 2026-10-05 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085