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 q8_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 | 396.0 | 396.2 | 396.4 | 396.9 | 397.9 | 399.9 |
| Q6_K_L | 319.94 GiB | 321.4 | 321.5 | 321.8 | 322.3 | 323.3 | 325.3 |
| Q6_K | 319.66 GiB | 321.1 | 321.3 | 321.5 | 322.0 | 323.0 | 325.0 |
| Q5_K_L | 264.48 GiB | 265.9 | 266.1 | 266.3 | 266.8 | 267.8 | 269.8 |
| Q5_K_M | 264.13 GiB | 265.6 | 265.7 | 266.0 | 266.5 | 267.5 | 269.5 |
| Q5_K_S | 256.01 GiB | 257.4 | 257.6 | 257.9 | 258.4 | 259.4 | 261.3 |
| Q4_1 | 232.94 GiB | 234.4 | 234.5 | 234.8 | 235.3 | 236.3 | 238.3 |
| Q4_K_L | 225.66 GiB | 227.1 | 227.3 | 227.5 | 228.0 | 229.0 | 231.0 |
| Q4_K_M | 225.24 GiB | 226.7 | 226.8 | 227.1 | 227.6 | 228.6 | 230.6 |
| Q4_K_S | 217.21 GiB | 218.6 | 218.8 | 219.1 | 219.6 | 220.5 | 222.5 |
| Q4_0 | 210.73 GiB | 212.2 | 212.3 | 212.6 | 213.1 | 214.1 | 216.1 |
| IQ4_NL | 209.91 GiB | 211.3 | 211.5 | 211.8 | 212.3 | 213.3 | 215.2 |
| IQ4_XS | 198.41 GiB | 199.8 | 200.0 | 200.3 | 200.8 | 201.7 | 203.7 |
| IQ3_M | 176.24 GiB | 177.7 | 177.8 | 178.1 | 178.6 | 179.6 | 181.6 |
| Q3_K_L | 176.09 GiB | 177.5 | 177.7 | 177.9 | 178.4 | 179.4 | 181.4 |
| Q3_K_M | 168.90 GiB | 170.3 | 170.5 | 170.7 | 171.2 | 172.2 | 174.2 |
| IQ3_XS | 168.77 GiB | 170.2 | 170.4 | 170.6 | 171.1 | 172.1 | 174.1 |
| Q3_K_S | 161.09 GiB | 162.5 | 162.7 | 162.9 | 163.4 | 164.4 | 166.4 |
| IQ3_XXS | 154.25 GiB | 155.7 | 155.8 | 156.1 | 156.6 | 157.6 | 159.6 |
| Q2_K_L | 130.22 GiB | 131.6 | 131.8 | 132.1 | 132.6 | 133.6 | 135.6 |
| Q2_K | 129.66 GiB | 131.1 | 131.3 | 131.5 | 132.0 | 133.0 | 135.0 |
| IQ2_M | 123.88 GiB | 125.3 | 125.5 | 125.7 | 126.2 | 127.2 | 129.2 |
| IQ2_S | 112.03 GiB | 113.5 | 113.6 | 113.9 | 114.4 | 115.4 | 117.4 |
| IQ2_XS | 110.12 GiB | 111.5 | 111.7 | 112.0 | 112.5 | 113.5 | 115.5 |
| IQ2_XXS | 98.70 GiB | 100.1 | 100.3 | 100.5 | 101.0 | 102.0 | 104.0 |
| IQ1_M | 84.63 GiB | 86.1 | 86.2 | 86.5 | 87.0 | 88.0 | 90.0 |
| IQ1_S | 75.64 GiB | 77.1 | 77.2 | 77.5 | 78.0 | 79.0 | 81.0 |
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 4,096-token window rather than the full context.