Can I run Trinity-Large-Thinking on a GeForce RTX 5070?
Not at these settings. No indexed quantization of Trinity-Large-Thinking fits GeForce RTX 5070 at any context we compute, with q4_0 KV. The smallest shipped quantization is 75.64 GiB in weights alone, against 11.16 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| Q8_0 | 394.59 GiB | 395.7 | 395.8 | 395.9 | 396.2 | 396.7 | 397.7 |
| Q6_K_L | 319.94 GiB | 321.0 | 321.1 | 321.3 | 321.5 | 322.0 | 323.1 |
| Q6_K | 319.66 GiB | 320.8 | 320.8 | 321.0 | 321.2 | 321.8 | 322.8 |
| Q5_K_L | 264.48 GiB | 265.6 | 265.7 | 265.8 | 266.1 | 266.6 | 267.6 |
| Q5_K_M | 264.13 GiB | 265.2 | 265.3 | 265.4 | 265.7 | 266.2 | 267.3 |
| Q5_K_S | 256.01 GiB | 257.1 | 257.2 | 257.3 | 257.6 | 258.1 | 259.2 |
| Q4_1 | 232.94 GiB | 234.0 | 234.1 | 234.3 | 234.5 | 235.0 | 236.1 |
| Q4_K_L | 225.66 GiB | 226.8 | 226.8 | 227.0 | 227.2 | 227.8 | 228.8 |
| Q4_K_M | 225.24 GiB | 226.3 | 226.4 | 226.6 | 226.8 | 227.3 | 228.4 |
| Q4_K_S | 217.21 GiB | 218.3 | 218.4 | 218.5 | 218.8 | 219.3 | 220.4 |
| Q4_0 | 210.73 GiB | 211.8 | 211.9 | 212.0 | 212.3 | 212.8 | 213.9 |
| IQ4_NL | 209.91 GiB | 211.0 | 211.1 | 211.2 | 211.5 | 212.0 | 213.1 |
| IQ4_XS | 198.41 GiB | 199.5 | 199.6 | 199.7 | 200.0 | 200.5 | 201.6 |
| IQ3_M | 176.24 GiB | 177.3 | 177.4 | 177.6 | 177.8 | 178.3 | 179.4 |
| Q3_K_L | 176.09 GiB | 177.2 | 177.3 | 177.4 | 177.7 | 178.2 | 179.2 |
| Q3_K_M | 168.90 GiB | 170.0 | 170.1 | 170.2 | 170.5 | 171.0 | 172.1 |
| IQ3_XS | 168.77 GiB | 169.9 | 169.9 | 170.1 | 170.3 | 170.9 | 171.9 |
| Q3_K_S | 161.09 GiB | 162.2 | 162.3 | 162.4 | 162.7 | 163.2 | 164.3 |
| IQ3_XXS | 154.25 GiB | 155.3 | 155.4 | 155.6 | 155.8 | 156.3 | 157.4 |
| Q2_K_L | 130.22 GiB | 131.3 | 131.4 | 131.5 | 131.8 | 132.3 | 133.4 |
| Q2_K | 129.66 GiB | 130.7 | 130.8 | 131.0 | 131.2 | 131.8 | 132.8 |
| IQ2_M | 123.88 GiB | 125.0 | 125.1 | 125.2 | 125.5 | 126.0 | 127.0 |
| IQ2_S | 112.03 GiB | 113.1 | 113.2 | 113.3 | 113.6 | 114.1 | 115.2 |
| IQ2_XS | 110.12 GiB | 111.2 | 111.3 | 111.4 | 111.7 | 112.2 | 113.3 |
| IQ2_XXS | 98.70 GiB | 99.8 | 99.9 | 100.0 | 100.3 | 100.8 | 101.9 |
| IQ1_M | 84.63 GiB | 85.7 | 85.8 | 85.9 | 86.2 | 86.7 | 87.8 |
| IQ1_S | 75.64 GiB | 76.7 | 76.8 | 77.0 | 77.2 | 77.7 | 78.8 |
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.