Can I run diffusiongemma-26B-A4B-it-HERETIC-Uncensored on a GeForce RTX 5050?
Not at these settings. No indexed quantization of diffusiongemma-26B-A4B-it-HERETIC-Uncensored fits GeForce RTX 5050 at any context we compute, with q8_0 KV. The smallest shipped quantization is 7.99 GiB in weights alone, against 7.44 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 47.07 GiB | 48.1 | 48.2 | 48.3 | 48.7 | 49.3 | 50.7 |
| Q8_0 | 25.03 GiB | 26.1 | 26.1 | 26.3 | 26.6 | 27.3 | 28.6 |
| Q6_K | 21.10 GiB | 22.1 | 22.2 | 22.4 | 22.7 | 23.4 | 24.7 |
| Q5_K_M | 17.83 GiB | 18.9 | 18.9 | 19.1 | 19.4 | 20.1 | 21.4 |
| Q4_K_M | 15.65 GiB | 16.7 | 16.8 | 16.9 | 17.3 | 17.9 | 19.3 |
| IQ4_NL | 13.59 GiB | 14.6 | 14.7 | 14.9 | 15.2 | 15.9 | 17.2 |
| IQ4_XS | 13.11 GiB | 14.1 | 14.2 | 14.4 | 14.7 | 15.4 | 16.7 |
| Q3_K_M | 12.38 GiB | 13.4 | 13.5 | 13.7 | 14.0 | 14.7 | 16.0 |
| IQ3_M | 11.55 GiB | 12.6 | 12.7 | 12.8 | 13.2 | 13.8 | 15.1 |
| IQ3_S | 11.39 GiB | 12.4 | 12.5 | 12.7 | 13.0 | 13.7 | 15.0 |
| Q2_K | 9.86 GiB | 10.9 | 11.0 | 11.1 | 11.5 | 12.1 | 13.5 |
| TQ2_0 | 8.72 GiB | 9.7 | 9.8 | 10.0 | 10.3 | 11.0 | 12.3 |
| TQ1_0 | 7.99 GiB | 9.0 | 9.1 | 9.3 | 9.6 | 10.3 | 11.6 |
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 1,024-token window rather than the full context.