Can I run Trinity-Large-Preview on a GeForce RTX 5050?
Not at these settings. No indexed quantization of Trinity-Large-Preview fits GeForce RTX 5050 at any context we compute, with q4_0 KV. The smallest shipped quantization is 76.06 GiB in weights alone, against 7.44 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 742.60 GiB | 743.7 | 743.8 | 743.9 | 744.2 | 744.7 | 745.8 |
| Q8_0 | 394.59 GiB | 395.7 | 395.8 | 395.9 | 396.2 | 396.7 | 397.7 |
| Q6_K_L | 305.35 GiB | 306.4 | 306.5 | 306.7 | 306.9 | 307.5 | 308.5 |
| Q6_K | 305.07 GiB | 306.2 | 306.3 | 306.4 | 306.6 | 307.2 | 308.2 |
| Q5_K_L | 264.22 GiB | 265.3 | 265.4 | 265.5 | 265.8 | 266.3 | 267.4 |
| Q5_K_M | 263.87 GiB | 265.0 | 265.0 | 265.2 | 265.4 | 266.0 | 267.0 |
| Q5_K_S | 255.88 GiB | 257.0 | 257.1 | 257.2 | 257.5 | 258.0 | 259.0 |
| Q4_1 | 232.76 GiB | 233.9 | 233.9 | 234.1 | 234.3 | 234.9 | 235.9 |
| Q4_K_L | 225.46 GiB | 226.6 | 226.6 | 226.8 | 227.0 | 227.6 | 228.6 |
| Q4_K_M | 225.04 GiB | 226.1 | 226.2 | 226.3 | 226.6 | 227.1 | 228.2 |
| Q4_K_S | 217.12 GiB | 218.2 | 218.3 | 218.4 | 218.7 | 219.2 | 220.3 |
| Q4_0 | 213.32 GiB | 214.4 | 214.5 | 214.6 | 214.9 | 215.4 | 216.5 |
| IQ4_NL | 209.71 GiB | 210.8 | 210.9 | 211.0 | 211.3 | 211.8 | 212.9 |
| IQ4_XS | 198.19 GiB | 199.3 | 199.4 | 199.5 | 199.8 | 200.3 | 201.4 |
| Q3_K_M | 176.43 GiB | 177.5 | 177.6 | 177.7 | 178.0 | 178.5 | 179.6 |
| Q3_K_L | 175.82 GiB | 176.9 | 177.0 | 177.1 | 177.4 | 177.9 | 179.0 |
| UD-IQ3_XXS | 170.01 GiB | 171.1 | 171.2 | 171.3 | 171.6 | 172.1 | 173.2 |
| IQ3_M | 168.61 GiB | 169.7 | 169.8 | 169.9 | 170.2 | 170.7 | 171.8 |
| Q3_K_S | 160.91 GiB | 162.0 | 162.1 | 162.2 | 162.5 | 163.0 | 164.1 |
| IQ3_XS | 151.27 GiB | 152.4 | 152.5 | 152.6 | 152.8 | 153.4 | 154.4 |
| IQ3_XXS | 146.20 GiB | 147.3 | 147.4 | 147.5 | 147.8 | 148.3 | 149.4 |
| UD-IQ2_M | 143.68 GiB | 144.8 | 144.9 | 145.0 | 145.3 | 145.8 | 146.8 |
| UD-IQ2_XXS | 142.94 GiB | 144.0 | 144.1 | 144.3 | 144.5 | 145.0 | 146.1 |
| Q2_K_L | 135.17 GiB | 136.3 | 136.4 | 136.5 | 136.7 | 137.3 | 138.3 |
| Q2_K | 135.04 GiB | 136.1 | 136.2 | 136.3 | 136.6 | 137.1 | 138.2 |
| IQ2_M | 116.50 GiB | 117.6 | 117.7 | 117.8 | 118.1 | 118.6 | 119.7 |
| IQ2_S | 102.49 GiB | 103.6 | 103.7 | 103.8 | 104.1 | 104.6 | 105.7 |
| IQ2_XS | 102.27 GiB | 103.4 | 103.5 | 103.6 | 103.8 | 104.4 | 105.4 |
| IQ2_XXS | 88.58 GiB | 89.7 | 89.8 | 89.9 | 90.2 | 90.7 | 91.7 |
| IQ1_M | 79.29 GiB | 80.4 | 80.5 | 80.6 | 80.9 | 81.4 | 82.4 |
| IQ1_S | 76.06 GiB | 77.1 | 77.2 | 77.4 | 77.6 | 78.2 | 79.2 |
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 4,096-token window rather than the full context.