Can I run Pantheon-Reasoning-31B-1.1 on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of Pantheon-Reasoning-31B-1.1 fits Radeon RX 6500 XT at any context we compute, with q8_0 KV. The smallest shipped quantization is 7.10 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 | 31.79 GiB | 33.7 | 34.1 | 34.7 | 36.0 | 38.7 | 44.0 |
| I1-Q6_K | 24.55 GiB | 26.5 | 26.8 | 27.5 | 28.8 | 31.5 | 36.8 |
| Q6_K | 24.55 GiB | 26.5 | 26.8 | 27.5 | 28.8 | 31.5 | 36.8 |
| I1-Q5_K_M | 21.25 GiB | 23.2 | 23.5 | 24.2 | 25.5 | 28.2 | 33.5 |
| Q5_K_M | 21.25 GiB | 23.2 | 23.5 | 24.2 | 25.5 | 28.2 | 33.5 |
| I1-Q5_K_S | 20.75 GiB | 22.7 | 23.0 | 23.7 | 25.0 | 27.7 | 33.0 |
| Q5_K_S | 20.75 GiB | 22.7 | 23.0 | 23.7 | 25.0 | 27.7 | 33.0 |
| I1-Q4_1 | 18.96 GiB | 20.9 | 21.2 | 21.9 | 23.2 | 25.9 | 31.2 |
| I1-Q4_K_M | 18.14 GiB | 20.1 | 20.4 | 21.1 | 22.4 | 25.1 | 30.4 |
| Q4_K_M | 18.14 GiB | 20.1 | 20.4 | 21.1 | 22.4 | 25.1 | 30.4 |
| I1-Q4_K_S | 17.28 GiB | 19.2 | 19.5 | 20.2 | 21.5 | 24.2 | 29.5 |
| Q4_K_S | 17.28 GiB | 19.2 | 19.5 | 20.2 | 21.5 | 24.2 | 29.5 |
| I1-Q4_0 | 17.22 GiB | 19.2 | 19.5 | 20.2 | 21.5 | 24.1 | 29.5 |
| IQ4_XS | 16.40 GiB | 18.3 | 18.7 | 19.3 | 20.7 | 23.3 | 28.6 |
| I1-IQ4_XS | 16.28 GiB | 18.2 | 18.6 | 19.2 | 20.5 | 23.2 | 28.5 |
| I1-Q3_K_L | 16.05 GiB | 18.0 | 18.3 | 19.0 | 20.3 | 23.0 | 28.3 |
| Q3_K_L | 16.05 GiB | 18.0 | 18.3 | 19.0 | 20.3 | 23.0 | 28.3 |
| I1-Q3_K_M | 14.80 GiB | 16.7 | 17.1 | 17.7 | 19.1 | 21.7 | 27.0 |
| Q3_K_M | 14.80 GiB | 16.7 | 17.1 | 17.7 | 19.1 | 21.7 | 27.0 |
| I1-IQ3_M | 14.00 GiB | 15.9 | 16.3 | 16.9 | 18.3 | 20.9 | 26.2 |
| I1-IQ3_S | 13.38 GiB | 15.3 | 15.6 | 16.3 | 17.6 | 20.3 | 25.6 |
| I1-Q3_K_S | 13.38 GiB | 15.3 | 15.6 | 16.3 | 17.6 | 20.3 | 25.6 |
| Q3_K_S | 13.38 GiB | 15.3 | 15.6 | 16.3 | 17.6 | 20.3 | 25.6 |
| I1-IQ3_XS | 12.74 GiB | 14.7 | 15.0 | 15.7 | 17.0 | 19.7 | 25.0 |
| I1-IQ3_XXS | 11.81 GiB | 13.7 | 14.1 | 14.7 | 16.1 | 18.7 | 24.0 |
| I1-Q2_K | 11.53 GiB | 13.5 | 13.8 | 14.5 | 15.8 | 18.4 | 23.8 |
| Q2_K | 11.53 GiB | 13.5 | 13.8 | 14.5 | 15.8 | 18.4 | 23.8 |
| I1-IQ2_M | 10.73 GiB | 12.7 | 13.0 | 13.7 | 15.0 | 17.6 | 23.0 |
| I1-Q2_K_S | 10.65 GiB | 12.6 | 12.9 | 13.6 | 14.9 | 17.6 | 22.9 |
| I1-IQ2_S | 10.02 GiB | 12.0 | 12.3 | 13.0 | 14.3 | 16.9 | 22.3 |
| I1-IQ2_XS | 9.31 GiB | 11.2 | 11.6 | 12.2 | 13.6 | 16.2 | 21.5 |
| I1-IQ2_XXS | 8.51 GiB | 10.4 | 10.8 | 11.4 | 12.8 | 15.4 | 20.7 |
| I1-IQ1_M | 7.63 GiB | 9.6 | 9.9 | 10.6 | 11.9 | 14.5 | 19.9 |
| I1-IQ1_S | 7.10 GiB | 9.0 | 9.4 | 10.0 | 11.4 | 14.0 | 19.3 |
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 1,024-token window rather than the full context.