Can I run Kimi-Dev-72B on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of Kimi-Dev-72B fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 21.47 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◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 135.44 GiB | 136.8 | 137.2 | 137.9 | 139.3 | 142.1 | 147.7 |
| Q8_0 | 71.96 GiB | 73.3 | 73.7 | 74.4 | 75.8 | 78.6 | 84.2 |
| Q6_K | 59.93 GiB | 61.3 | 61.7 | 62.4 | 63.8 | 66.6 | 72.2 |
| Q4_1 | 42.56 GiB | 43.9 | 44.3 | 45.0 | 46.4 | 49.2 | 54.8 |
| Q4_0 | 38.54 GiB | 39.9 | 40.3 | 41.0 | 42.4 | 45.2 | 50.8 |
| IQ4_NL | 38.48 GiB | 39.9 | 40.2 | 40.9 | 42.3 | 45.1 | 50.8 |
| IQ4_XS | 37.02 GiB | 38.4 | 38.8 | 39.5 | 40.9 | 43.7 | 49.3 |
| Q3_K_M | 35.11 GiB | 36.5 | 36.8 | 37.5 | 39.0 | 41.8 | 47.4 |
| Q3_K_S | 32.12 GiB | 33.5 | 33.9 | 34.6 | 36.0 | 38.8 | 44.4 |
| UD-IQ3_XXS | 29.67 GiB | 31.1 | 31.4 | 32.1 | 33.5 | 36.3 | 41.9 |
| UD-IQ2_M | 27.56 GiB | 28.9 | 29.3 | 30.0 | 31.4 | 34.2 | 39.8 |
| UD-IQ2_XXS | 23.94 GiB | 25.3 | 25.7 | 26.4 | 27.8 | 30.6 | 36.2 |
| UD-IQ1_M | 22.35 GiB | 23.7 | 24.1 | 24.8 | 26.2 | 29.0 | 34.6 |
| UD-IQ1_S | 21.47 GiB | 22.8 | 23.2 | 23.9 | 25.3 | 28.1 | 33.7 |
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 131,072-token window rather than the full context.