Can I run Kimi-Dev-72B on a GeForce RTX 2080 Ti?
Not at these settings. No indexed quantization of Kimi-Dev-72B fits GeForce RTX 2080 Ti at any context we compute, with q8_0 KV. The smallest shipped quantization is 21.47 GiB in weights alone, against 10.23 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 | 137.0 | 137.7 | 139.0 | 141.7 | 147.0 | 157.6 |
| Q8_0 | 71.96 GiB | 73.6 | 74.2 | 75.5 | 78.2 | 83.5 | 94.1 |
| Q6_K | 59.93 GiB | 61.5 | 62.2 | 63.5 | 66.2 | 71.5 | 82.1 |
| Q4_1 | 42.56 GiB | 44.2 | 44.8 | 46.1 | 48.8 | 54.1 | 64.7 |
| Q4_0 | 38.54 GiB | 40.1 | 40.8 | 42.1 | 44.8 | 50.1 | 60.7 |
| IQ4_NL | 38.48 GiB | 40.1 | 40.7 | 42.1 | 44.7 | 50.0 | 60.7 |
| IQ4_XS | 37.02 GiB | 38.6 | 39.3 | 40.6 | 43.3 | 48.6 | 59.2 |
| Q3_K_M | 35.11 GiB | 36.7 | 37.4 | 38.7 | 41.4 | 46.7 | 57.3 |
| Q3_K_S | 32.12 GiB | 33.7 | 34.4 | 35.7 | 38.4 | 43.7 | 54.3 |
| UD-IQ3_XXS | 29.67 GiB | 31.3 | 31.9 | 33.3 | 35.9 | 41.2 | 51.8 |
| UD-IQ2_M | 27.56 GiB | 29.2 | 29.8 | 31.2 | 33.8 | 39.1 | 49.7 |
| UD-IQ2_XXS | 23.94 GiB | 25.5 | 26.2 | 27.5 | 30.2 | 35.5 | 46.1 |
| UD-IQ1_M | 22.35 GiB | 23.9 | 24.6 | 25.9 | 28.6 | 33.9 | 44.5 |
| UD-IQ1_S | 21.47 GiB | 23.1 | 23.7 | 25.1 | 27.7 | 33.0 | 43.6 |
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.