4× H20
6 records across 1 models (quants: FP8).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 2045.35 | – | – | – | 16384 | 2026-09-28 | community | – | source |
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 1619.19 | – | – | – | 16384 | 2026-09-28 | community | – | source |
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 1137.92 | – | – | – | 16384 | 2026-09-28 | community | – | source |
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 962.39 | – | – | – | 16384 | 2026-09-28 | community | – | source |
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 903.60 | – | – | – | 16384 | 2026-09-28 | community | – | source |
| 4× H20 | GLM-5.3-Flash | FP8 | vLLM | 820.00 | – | – | – | 16384 | 2026-09-28 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085