M3 Ultra (80 GPU)
3 records across 1 models (quants: F16, Q4_0, Q8_0).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M3 Ultra (80 GPU) | llama 7B | Q4_0 | llama.cpp (Metal) | 92.14 | 1471.24 | – | – | – | 2023-11-22 | community | – | source |
| M3 Ultra (80 GPU) | llama 7B | Q8_0 | llama.cpp (Metal) | 63.93 | 1487.51 | – | – | – | 2023-11-22 | community | – | source |
| M3 Ultra (80 GPU) | llama 7B | F16 | llama.cpp (Metal) | 39.78 | 1538.34 | – | – | – | 2023-11-22 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085