Can I run Trinity-Large-Thinking on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of Trinity-Large-Thinking fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 75.64 GiB in weights alone, against 3.72 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 | 395.8 | 395.9 | 396.0 | 396.3 | 396.8 | 397.8 |
| Q6_K_L | 319.94 GiB | 321.1 | 321.2 | 321.4 | 321.6 | 322.1 | 323.2 |
| Q6_K | 319.66 GiB | 320.9 | 320.9 | 321.1 | 321.3 | 321.9 | 322.9 |
| Q5_K_L | 264.48 GiB | 265.7 | 265.8 | 265.9 | 266.2 | 266.7 | 267.7 |
| Q5_K_M | 264.13 GiB | 265.3 | 265.4 | 265.5 | 265.8 | 266.3 | 267.4 |
| Q5_K_S | 256.01 GiB | 257.2 | 257.3 | 257.4 | 257.7 | 258.2 | 259.3 |
| Q4_1 | 232.94 GiB | 234.1 | 234.2 | 234.4 | 234.6 | 235.1 | 236.2 |
| Q4_K_L | 225.66 GiB | 226.9 | 226.9 | 227.1 | 227.3 | 227.9 | 228.9 |
| Q4_K_M | 225.24 GiB | 226.4 | 226.5 | 226.7 | 226.9 | 227.4 | 228.5 |
| Q4_K_S | 217.21 GiB | 218.4 | 218.5 | 218.6 | 218.9 | 219.4 | 220.5 |
| Q4_0 | 210.73 GiB | 211.9 | 212.0 | 212.1 | 212.4 | 212.9 | 214.0 |
| IQ4_NL | 209.91 GiB | 211.1 | 211.2 | 211.3 | 211.6 | 212.1 | 213.2 |
| IQ4_XS | 198.41 GiB | 199.6 | 199.7 | 199.8 | 200.1 | 200.6 | 201.7 |
| IQ3_M | 176.24 GiB | 177.4 | 177.5 | 177.7 | 177.9 | 178.4 | 179.5 |
| Q3_K_L | 176.09 GiB | 177.3 | 177.4 | 177.5 | 177.8 | 178.3 | 179.3 |
| Q3_K_M | 168.90 GiB | 170.1 | 170.2 | 170.3 | 170.6 | 171.1 | 172.2 |
| IQ3_XS | 168.77 GiB | 170.0 | 170.0 | 170.2 | 170.4 | 171.0 | 172.0 |
| Q3_K_S | 161.09 GiB | 162.3 | 162.4 | 162.5 | 162.8 | 163.3 | 164.4 |
| IQ3_XXS | 154.25 GiB | 155.4 | 155.5 | 155.7 | 155.9 | 156.4 | 157.5 |
| Q2_K_L | 130.22 GiB | 131.4 | 131.5 | 131.6 | 131.9 | 132.4 | 133.5 |
| Q2_K | 129.66 GiB | 130.8 | 130.9 | 131.1 | 131.3 | 131.9 | 132.9 |
| IQ2_M | 123.88 GiB | 125.1 | 125.2 | 125.3 | 125.6 | 126.1 | 127.1 |
| IQ2_S | 112.03 GiB | 113.2 | 113.3 | 113.4 | 113.7 | 114.2 | 115.3 |
| IQ2_XS | 110.12 GiB | 111.3 | 111.4 | 111.5 | 111.8 | 112.3 | 113.4 |
| IQ2_XXS | 98.70 GiB | 99.9 | 100.0 | 100.1 | 100.4 | 100.9 | 102.0 |
| IQ1_M | 84.63 GiB | 85.8 | 85.9 | 86.0 | 86.3 | 86.8 | 87.9 |
| IQ1_S | 75.64 GiB | 76.8 | 76.9 | 77.1 | 77.3 | 77.8 | 78.9 |
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.