Our catalog tracks 4 Gemma models (1B–27.4 parameters). For each, the table shows the cheapest retail GPU that fully fits it at Q4 / 8K context and the fastest retail GPU, with decode tok/s. Fit means weights + KV cache + 1 GB runtime overhead within 95% of VRAM — not weights alone. Methodology.
| # | Model | Params | Q4/8K needs (GB) | Best value GPU (fits) | Fastest retail GPU | Buy |
|---|---|---|---|---|---|---|
| 1 | gemma-3-1b-it | 1.0B | 1.75 | Arc B570 10GB ≈241.8 tok/s est. | RTX 5090 32GB ≈1140.4 tok/s est. | Search Amazon (affiliate) |
| 2 | gemma-3-4b-it | 4.3B | 4.15 | Arc B570 10GB ≈62.0 tok/s est. | RTX 5090 32GB ≈292.4 tok/s est. | Search Amazon (affiliate) |
| 3 | gemma-3-12b-it | 12.2B | 10.11 | Arc B580 12GB ≈24.2 tok/s est. | RTX 5090 32GB ≈95.0 tok/s est. | Search Amazon (affiliate) |
| 4 | gemma-3-27b-it | 27.4B | 21.51 | RX 7900 XTX 24GB ≈22.5 tok/s est. | RTX 5090 32GB ≈41.9 tok/s est. | ASRock Radeon RX 7900 XTX Phantom Gaming 24GB OC (affiliate) · More listings |