gemma3 12B
4 records across 2 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX PRO 4500 Blackwell | gemma3 12B | Q4_K - Medium | llama.cpp (CUDA) | 88.38 | 4364.73 | – | – | tg128 | 2025-11-14 | community | – | source |
| RTX PRO 4500 Blackwell | gemma3 12B | Q4_K - Medium | llama.cpp (CUDA) | 86.0 | 4075.46 | – | – | tg128 | 2025-11-14 | community | – | source |
| V100 | gemma3 12B | Q4_K - Medium | llama.cpp (CUDA) | 68.94 | 1771.04 | – | – | tg128 | 2025-09-22 | community | – | source |
| V100 | gemma3 12B | Q4_K - Medium | llama.cpp (CUDA) | 67.34 | 1857.73 | – | – | tg128 | 2025-09-22 | community | – | source |