8× H20
8 records across 1 models (quants: MXFP8).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 2014.9 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 1983.8 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 739.0 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 709.9 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 621.4 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 601.1 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 338.0 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| 8× H20 | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 335.2 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085