8×H100 (640 GB HBM)
1 records across 1 models (quants: IQ1_S-XS).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 8×H100 (640 GB HBM) | Kimi-K3 | IQ1_S-XS | llama.cpp (CUDA) | 6.8 | – | – | – | 8192 | 2026-08-10 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085