community pool / model answer
gemma-3-12b-it
chat · google/gemma-3-12b-it
44 tok/s · model median · 41 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| RTX 2080 8GB | 4-bit | 21 tok/s | reported | - | 8.1 GB | 2,048 | ollama |
| RTX 3060 12GB | 4-bit | 40 tok/s | reported | - | 8.1 GB | 2,048 | ollama |
| RTX 3060 12GB ×2 | 4-bit | 34.3 tok/s | measured | 8,289 | 12.2 GB | 131,072 | llama.cpp |
| RTX 3080 12GB | 4-bit | 51.5 tok/s | measured | - | 8.6 GB | 2,048 | ollama |
| RTX 3090 24GB | 4-bit | 59.6 tok/s | measured | 3,320 | - GB | 131,072 | llama.cpp |
| RTX3060+RTX2080-pooled-20GB 20GB | 4-bit | 44 tok/s | measured | - | 8.1 GB | 2,048 | ollama |
| Ryzen AI Max 395 64GB | 4-bit | 24.6 tok/s | reported | 1,771 | 9.1 GB | 8,192 | llama.cpp |
Context tested
The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.