Can I run Nidum-Gemma-3-27B-it-Uncensored on a GeForce RTX 3050?
Not at these settings. No indexed quantization of Nidum-Gemma-3-27B-it-Uncensored fits GeForce RTX 3050 at any context we compute, with q4_0 KV. The smallest shipped quantization is 5.83 GiB in weights alone, against 5.58 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| Q8_0 | 26.74 GiB | 27.9 | 28.0 | 28.1 | 28.5 | 29.2 | 30.6 |
| I1-Q6_K | 20.64 GiB | 21.8 | 21.9 | 22.0 | 22.4 | 23.1 | 24.5 |
| Q6_K | 20.64 GiB | 21.8 | 21.9 | 22.0 | 22.4 | 23.1 | 24.5 |
| I1-Q5_K_M | 17.95 GiB | 19.1 | 19.2 | 19.4 | 19.7 | 20.4 | 21.8 |
| Q5_K_M | 17.95 GiB | 19.1 | 19.2 | 19.4 | 19.7 | 20.4 | 21.8 |
| I1-Q5_K_S | 17.48 GiB | 18.6 | 18.7 | 18.9 | 19.2 | 19.9 | 21.3 |
| Q5_K_S | 17.48 GiB | 18.6 | 18.7 | 18.9 | 19.2 | 19.9 | 21.3 |
| I1-Q4_1 | 15.99 GiB | 17.1 | 17.2 | 17.4 | 17.7 | 18.4 | 19.9 |
| I1-Q4_K_M | 15.41 GiB | 16.6 | 16.6 | 16.8 | 17.2 | 17.9 | 19.3 |
| Q4_K_M | 15.41 GiB | 16.6 | 16.6 | 16.8 | 17.2 | 17.9 | 19.3 |
| I1-Q4_K_S | 14.60 GiB | 15.7 | 15.8 | 16.0 | 16.4 | 17.1 | 18.5 |
| Q4_K_S | 14.60 GiB | 15.7 | 15.8 | 16.0 | 16.4 | 17.1 | 18.5 |
| I1-Q4_0 | 14.55 GiB | 15.7 | 15.8 | 15.9 | 16.3 | 17.0 | 18.4 |
| IQ4_XS | 13.87 GiB | 15.0 | 15.1 | 15.3 | 15.6 | 16.3 | 17.7 |
| I1-IQ4_XS | 13.75 GiB | 14.9 | 15.0 | 15.2 | 15.5 | 16.2 | 17.6 |
| I1-Q3_K_L | 13.54 GiB | 14.7 | 14.8 | 14.9 | 15.3 | 16.0 | 17.4 |
| Q3_K_L | 13.54 GiB | 14.7 | 14.8 | 14.9 | 15.3 | 16.0 | 17.4 |
| I1-Q3_K_M | 12.51 GiB | 13.7 | 13.7 | 13.9 | 14.3 | 15.0 | 16.4 |
| Q3_K_M | 12.51 GiB | 13.7 | 13.7 | 13.9 | 14.3 | 15.0 | 16.4 |
| I1-IQ3_M | 11.69 GiB | 12.8 | 12.9 | 13.1 | 13.4 | 14.1 | 15.6 |
| I1-IQ3_S | 11.33 GiB | 12.5 | 12.6 | 12.7 | 13.1 | 13.8 | 15.2 |
| I1-Q3_K_S | 11.33 GiB | 12.5 | 12.6 | 12.7 | 13.1 | 13.8 | 15.2 |
| Q3_K_S | 11.33 GiB | 12.5 | 12.6 | 12.7 | 13.1 | 13.8 | 15.2 |
| I1-IQ3_XS | 10.77 GiB | 11.9 | 12.0 | 12.2 | 12.5 | 13.2 | 14.6 |
| I1-IQ3_XXS | 9.98 GiB | 11.1 | 11.2 | 11.4 | 11.7 | 12.4 | 13.8 |
| I1-Q2_K | 9.78 GiB | 10.9 | 11.0 | 11.2 | 11.5 | 12.2 | 13.6 |
| Q2_K | 9.78 GiB | 10.9 | 11.0 | 11.2 | 11.5 | 12.2 | 13.6 |
| I1-Q2_K_S | 9.09 GiB | 10.2 | 10.3 | 10.5 | 10.8 | 11.5 | 13.0 |
| I1-IQ2_M | 8.84 GiB | 10.0 | 10.1 | 10.2 | 10.6 | 11.3 | 12.7 |
| I1-IQ2_S | 8.18 GiB | 9.3 | 9.4 | 9.6 | 9.9 | 10.6 | 12.0 |
| I1-IQ2_XS | 7.86 GiB | 9.0 | 9.1 | 9.3 | 9.6 | 10.3 | 11.7 |
| I1-IQ2_XXS | 7.16 GiB | 8.3 | 8.4 | 8.6 | 8.9 | 9.6 | 11.0 |
| I1-IQ1_M | 6.33 GiB | 7.5 | 7.6 | 7.7 | 8.1 | 8.8 | 10.2 |
| I1-IQ1_S | 5.83 GiB | 7.0 | 7.1 | 7.2 | 7.6 | 8.3 | 9.7 |
Figures are GiB of total memory: weights plus KV cache plus compute buffer and backend overhead. Weights and KV are near-exact; the overhead term is modeled. Hover any cell for the breakdown.
Why other calculators disagree
A parameters × bits ÷ 8 estimate ignores two things that dominate at long context. First, the weights themselves are not the nominal rate — quantizations are mixtures, so the real file is consistently larger than the label implies. Second, most of this model's layers cache only a 1,024-token window rather than the full context.