10× MI100
2 records across 1 models (quants: Q3_K_M, Q4_K_XL).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 10× MI100 | GLM-5.3-Flash | Q4_K_XL | llama.cpp (ROCm) | 33.8 | 780 | – | – | tg128 | – | community | – | source |
| 10× MI100 | GLM-5.3-Flash | Q3_K_M | llama.cpp (ROCm) | 32.2 | 730 | – | – | tg128 | – | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085