Can I run Trinity-Mini on a GeForce RTX 3050?
Not at these settings. No indexed quantization of Trinity-Mini fits GeForce RTX 3050 at any context we compute, with q8_0 KV. The smallest shipped quantization is 6.11 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◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 48.68 GiB | 49.6 | 49.6 | 49.7 | 49.8 | 50.1 | 50.6 |
| Q8_0 | 25.88 GiB | 26.8 | 26.8 | 26.9 | 27.0 | 27.3 | 27.8 |
| Q6_K_L | 20.22 GiB | 21.1 | 21.1 | 21.2 | 21.3 | 21.6 | 22.1 |
| Q6_K | 20.04 GiB | 20.9 | 21.0 | 21.0 | 21.2 | 21.4 | 22.0 |
| Q5_K_L | 17.59 GiB | 18.5 | 18.5 | 18.6 | 18.7 | 19.0 | 19.5 |
| Q5_K_M | 17.36 GiB | 18.2 | 18.3 | 18.3 | 18.5 | 18.7 | 19.3 |
| Q5_K_S | 16.86 GiB | 17.8 | 17.8 | 17.9 | 18.0 | 18.3 | 18.8 |
| Q4_1 | 15.37 GiB | 16.3 | 16.3 | 16.4 | 16.5 | 16.8 | 17.3 |
| Q4_K_L | 15.12 GiB | 16.0 | 16.0 | 16.1 | 16.2 | 16.5 | 17.0 |
| Q4_K_M | 14.84 GiB | 15.7 | 15.8 | 15.8 | 16.0 | 16.2 | 16.8 |
| Q4_K_S | 14.36 GiB | 15.2 | 15.3 | 15.3 | 15.5 | 15.7 | 16.3 |
| Q4_0 | 14.11 GiB | 15.0 | 15.0 | 15.1 | 15.2 | 15.5 | 16.0 |
| IQ4_NL | 13.92 GiB | 14.8 | 14.8 | 14.9 | 15.0 | 15.3 | 15.8 |
| IQ4_XS | 13.19 GiB | 14.1 | 14.1 | 14.2 | 14.3 | 14.6 | 15.1 |
| Q3_K_L | 12.66 GiB | 13.6 | 13.6 | 13.7 | 13.8 | 14.1 | 14.6 |
| Q3_K_M | 11.68 GiB | 12.6 | 12.6 | 12.7 | 12.8 | 13.1 | 13.6 |
| IQ3_M | 11.27 GiB | 12.2 | 12.2 | 12.3 | 12.4 | 12.7 | 13.2 |
| Q3_K_S | 10.80 GiB | 11.7 | 11.7 | 11.8 | 11.9 | 12.2 | 12.7 |
| IQ3_XS | 10.23 GiB | 11.1 | 11.2 | 11.2 | 11.4 | 11.6 | 12.1 |
| IQ3_XXS | 9.82 GiB | 10.7 | 10.7 | 10.8 | 10.9 | 11.2 | 11.7 |
| Q2_K_L | 9.15 GiB | 10.0 | 10.1 | 10.1 | 10.3 | 10.5 | 11.1 |
| Q2_K | 9.01 GiB | 9.9 | 9.9 | 10.0 | 10.1 | 10.4 | 10.9 |
| IQ2_M | 7.90 GiB | 8.8 | 8.8 | 8.9 | 9.0 | 9.3 | 9.8 |
| IQ2_S | 7.04 GiB | 7.9 | 8.0 | 8.0 | 8.2 | 8.4 | 9.0 |
| IQ2_XS | 6.97 GiB | 7.9 | 7.9 | 8.0 | 8.1 | 8.4 | 8.9 |
| IQ2_XXS | 6.11 GiB | 7.0 | 7.0 | 7.1 | 7.2 | 7.5 | 8.0 |
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 2,048-token window rather than the full context.