Can I run gemma-4-26B-A4B-it on a GeForce RTX 3080?
Not at these settings. No indexed quantization of gemma-4-26B-A4B-it fits GeForce RTX 3080 at any context we compute, with q4_0 KV. The smallest shipped quantization is 8.99 GiB in weights alone, against 9.30 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 48.65 GiB | 49.6 | 49.6 | 49.7 | 49.9 | 50.2 | 50.9 |
| Q8_0 | 25.89 GiB | 26.8 | 26.8 | 26.9 | 27.1 | 27.5 | 28.2 |
| UD-Q6_K | 21.58 GiB | 22.5 | 22.5 | 22.6 | 22.8 | 23.2 | 23.9 |
| Q6_K_L | 21.46 GiB | 22.4 | 22.4 | 22.5 | 22.7 | 23.0 | 23.7 |
| Q6_K | 21.29 GiB | 22.2 | 22.3 | 22.3 | 22.5 | 22.9 | 23.6 |
| UD-Q5_K_M | 19.70 GiB | 20.6 | 20.7 | 20.7 | 20.9 | 21.3 | 22.0 |
| Q5_K_L | 18.16 GiB | 19.1 | 19.1 | 19.2 | 19.4 | 19.7 | 20.4 |
| Q5_K_M | 17.99 GiB | 18.9 | 19.0 | 19.0 | 19.2 | 19.6 | 20.3 |
| UD-Q5_K_S | 17.56 GiB | 18.5 | 18.5 | 18.6 | 18.8 | 19.1 | 19.8 |
| Q5_K_S | 16.88 GiB | 17.8 | 17.8 | 17.9 | 18.1 | 18.5 | 19.2 |
| Q4_K_L | 16.03 GiB | 16.9 | 17.0 | 17.1 | 17.3 | 17.6 | 18.3 |
| Q4_K_M | 15.87 GiB | 16.8 | 16.8 | 16.9 | 17.1 | 17.4 | 18.1 |
| UD-Q4_K_M | 15.78 GiB | 16.7 | 16.7 | 16.8 | 17.0 | 17.4 | 18.1 |
| UD-Q4_K_S | 15.36 GiB | 16.3 | 16.3 | 16.4 | 16.6 | 16.9 | 17.6 |
| Q4_1 | 15.04 GiB | 15.9 | 16.0 | 16.1 | 16.3 | 16.6 | 17.3 |
| Q4_K_S | 14.76 GiB | 15.7 | 15.7 | 15.8 | 16.0 | 16.3 | 17.0 |
| Q4_0 | 14.04 GiB | 15.0 | 15.0 | 15.1 | 15.3 | 15.6 | 16.3 |
| IQ4_NL | 13.69 GiB | 14.6 | 14.6 | 14.7 | 14.9 | 15.3 | 16.0 |
| IQ4_XS | 13.23 GiB | 14.1 | 14.2 | 14.3 | 14.5 | 14.8 | 15.5 |
| UD-IQ4_NL | 12.68 GiB | 13.6 | 13.6 | 13.7 | 13.9 | 14.3 | 15.0 |
| UD-IQ4_XS | 12.66 GiB | 13.6 | 13.6 | 13.7 | 13.9 | 14.2 | 14.9 |
| IQ3_M | 12.37 GiB | 13.3 | 13.3 | 13.4 | 13.6 | 13.9 | 14.6 |
| Q3_K_L | 12.29 GiB | 13.2 | 13.3 | 13.3 | 13.5 | 13.9 | 14.6 |
| Q3_K_M | 12.13 GiB | 13.0 | 13.1 | 13.2 | 13.3 | 13.7 | 14.4 |
| UD-Q3_K_M | 11.85 GiB | 12.8 | 12.8 | 12.9 | 13.1 | 13.4 | 14.1 |
| Q3_K_S | 11.69 GiB | 12.6 | 12.6 | 12.7 | 12.9 | 13.3 | 14.0 |
| IQ3_XS | 11.58 GiB | 12.5 | 12.5 | 12.6 | 12.8 | 13.2 | 13.9 |
| IQ3_XXS | 11.33 GiB | 12.2 | 12.3 | 12.4 | 12.5 | 12.9 | 13.6 |
| UD-IQ3_XXS | 10.63 GiB | 11.5 | 11.6 | 11.7 | 11.9 | 12.2 | 12.9 |
| UD-IQ3_S | 10.51 GiB | 11.4 | 11.5 | 11.6 | 11.7 | 12.1 | 12.8 |
| Q2_K_L | 10.37 GiB | 11.3 | 11.3 | 11.4 | 11.6 | 11.9 | 12.6 |
| Q2_K | 10.20 GiB | 11.1 | 11.2 | 11.2 | 11.4 | 11.8 | 12.5 |
| IQ2_M | 9.97 GiB | 10.9 | 10.9 | 11.0 | 11.2 | 11.5 | 12.2 |
| IQ2_S | 9.53 GiB | 10.4 | 10.5 | 10.6 | 10.7 | 11.1 | 11.8 |
| IQ2_XS | 9.43 GiB | 10.3 | 10.4 | 10.5 | 10.7 | 11.0 | 11.7 |
| UD-IQ2_M | 9.33 GiB | 10.2 | 10.3 | 10.4 | 10.5 | 10.9 | 11.6 |
| UD-IQ2_XXS | 9.24 GiB | 10.2 | 10.2 | 10.3 | 10.5 | 10.8 | 11.5 |
| IQ2_XXS | 8.99 GiB | 9.9 | 10.0 | 10.0 | 10.2 | 10.6 | 11.3 |
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.