Can I run gpt-oss-120b-Uncensored-xCloud on a GeForce RTX 2080 Ti?
Not at these settings. No indexed quantization of gpt-oss-120b-Uncensored-xCloud fits GeForce RTX 2080 Ti at any context we compute, with q4_0 KV. The smallest shipped quantization is 61.54 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◐ |
|---|---|---|---|---|---|---|---|
| I1-Q5_K_M | 87.49 GiB | 88.3 | 88.4 | 88.4 | 88.6 | 88.9 | 89.6 |
| Q5_K_M | 87.49 GiB | 88.3 | 88.4 | 88.4 | 88.6 | 88.9 | 89.6 |
| I1-Q5_K_S | 81.92 GiB | 82.8 | 82.8 | 82.9 | 83.0 | 83.4 | 84.0 |
| Q5_K_S | 81.92 GiB | 82.8 | 82.8 | 82.9 | 83.0 | 83.4 | 84.0 |
| I1-Q4_K_M | 81.82 GiB | 82.7 | 82.7 | 82.8 | 82.9 | 83.2 | 83.9 |
| Q4_K_M | 81.82 GiB | 82.7 | 82.7 | 82.8 | 82.9 | 83.2 | 83.9 |
| I1-Q4_K_S | 75.38 GiB | 76.2 | 76.3 | 76.3 | 76.5 | 76.8 | 77.4 |
| Q4_K_S | 75.38 GiB | 76.2 | 76.3 | 76.3 | 76.5 | 76.8 | 77.4 |
| I1-Q4_1 | 68.42 GiB | 69.3 | 69.3 | 69.4 | 69.5 | 69.8 | 70.5 |
| I1-Q3_K_L | 68.39 GiB | 69.2 | 69.3 | 69.3 | 69.5 | 69.8 | 70.5 |
| Q3_K_L | 68.39 GiB | 69.2 | 69.3 | 69.3 | 69.5 | 69.8 | 70.5 |
| I1-Q3_K_M | 66.24 GiB | 67.1 | 67.1 | 67.2 | 67.3 | 67.7 | 68.3 |
| Q3_K_M | 66.24 GiB | 67.1 | 67.1 | 67.2 | 67.3 | 67.7 | 68.3 |
| IQ4_XS | 62.40 GiB | 63.2 | 63.3 | 63.3 | 63.5 | 63.8 | 64.5 |
| I1-IQ3_M | 62.16 GiB | 63.0 | 63.0 | 63.1 | 63.3 | 63.6 | 64.2 |
| I1-Q2_K_S | 62.06 GiB | 62.9 | 62.9 | 63.0 | 63.2 | 63.5 | 64.1 |
| I1-Q4_0 | 61.90 GiB | 62.7 | 62.8 | 62.9 | 63.0 | 63.3 | 64.0 |
| I1-IQ4_XS | 61.65 GiB | 62.5 | 62.5 | 62.6 | 62.8 | 63.1 | 63.7 |
| I1-IQ2_M | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-IQ2_S | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-IQ3_S | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-IQ3_XS | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-IQ3_XXS | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-Q2_K | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| Q2_K | 61.61 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-Q3_K_S | 61.60 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| Q3_K_S | 61.60 GiB | 62.4 | 62.5 | 62.6 | 62.7 | 63.0 | 63.7 |
| I1-IQ2_XS | 61.55 GiB | 62.4 | 62.4 | 62.5 | 62.7 | 63.0 | 63.6 |
| I1-IQ1_M | 61.54 GiB | 62.4 | 62.4 | 62.5 | 62.7 | 63.0 | 63.6 |
| I1-IQ1_S | 61.54 GiB | 62.4 | 62.4 | 62.5 | 62.7 | 63.0 | 63.6 |
| I1-IQ2_XXS | 61.54 GiB | 62.4 | 62.4 | 62.5 | 62.7 | 63.0 | 63.6 |
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 128-token window rather than the full context.