How we got this number
Each generated token reads the whole model from memory once: 8.1 GB. The Minisforum UM790 Pro (Ryzen 9 7940HS, 32GB) moves 89.6 GB/s, so the ceiling is 11 tokens/sec. Real machines reach a fraction of that ceiling; we use 60%, an assumption until we get measured runs for this kind of machine. Send us your real numbers and we'll replace the estimate.
Faster machines for Gemma 3 12B
Other models on the Minisforum UM790 Pro (Ryzen 9 7940HS, 32GB)
| lfm 2.5 8B (A1B) | Fast | 44 tok/s |
| Gemma 2B | Fast | 32 tok/s |
| smollm 2 1.7B | Fast | 30 tok/s |
| starCoder 2 3B | Fast | 30 tok/s |
| falcon 3 1B | Fast | 30 tok/s |
| Qwen2.5 3B | Fast | 28 tok/s |