Can I run Medgamma27B on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of Medgamma27B fits Radeon RX 6500 XT at any context we compute, with q8_0 KV. The smallest shipped quantization is 5.83 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◐ |
|---|---|---|---|---|---|---|---|
| I1-Q6_K | 20.64 GiB | 22.1 | 22.3 | 22.6 | 23.3 | 24.6 | 27.3 |
| I1-Q5_K_M | 17.95 GiB | 19.4 | 19.6 | 19.9 | 20.6 | 21.9 | 24.6 |
| I1-Q5_K_S | 17.48 GiB | 18.9 | 19.1 | 19.4 | 20.1 | 21.4 | 24.1 |
| I1-Q4_1 | 15.99 GiB | 17.5 | 17.6 | 18.0 | 18.6 | 19.9 | 22.6 |
| I1-Q4_K_M | 15.41 GiB | 16.9 | 17.0 | 17.4 | 18.0 | 19.4 | 22.0 |
| I1-Q4_K_S | 14.60 GiB | 16.1 | 16.2 | 16.6 | 17.2 | 18.6 | 21.2 |
| I1-Q4_0 | 14.55 GiB | 16.0 | 16.2 | 16.5 | 17.2 | 18.5 | 21.2 |
| I1-IQ4_XS | 13.75 GiB | 15.2 | 15.4 | 15.7 | 16.4 | 17.7 | 20.4 |
| I1-Q3_K_L | 13.54 GiB | 15.0 | 15.2 | 15.5 | 16.2 | 17.5 | 20.2 |
| I1-Q3_K_M | 12.51 GiB | 14.0 | 14.2 | 14.5 | 15.1 | 16.5 | 19.1 |
| I1-IQ3_M | 11.69 GiB | 13.2 | 13.3 | 13.7 | 14.3 | 15.6 | 18.3 |
| I1-IQ3_S | 11.33 GiB | 12.8 | 13.0 | 13.3 | 14.0 | 15.3 | 17.9 |
| I1-Q3_K_S | 11.33 GiB | 12.8 | 13.0 | 13.3 | 14.0 | 15.3 | 17.9 |
| I1-IQ3_XS | 10.77 GiB | 12.2 | 12.4 | 12.7 | 13.4 | 14.7 | 17.4 |
| I1-IQ3_XXS | 9.98 GiB | 11.5 | 11.6 | 11.9 | 12.6 | 13.9 | 16.6 |
| I1-Q2_K | 9.78 GiB | 11.3 | 11.4 | 11.8 | 12.4 | 13.7 | 16.4 |
| I1-Q2_K_S | 9.09 GiB | 10.6 | 10.7 | 11.1 | 11.7 | 13.0 | 15.7 |
| I1-IQ2_M | 8.84 GiB | 10.3 | 10.5 | 10.8 | 11.5 | 12.8 | 15.5 |
| I1-IQ2_S | 8.18 GiB | 9.6 | 9.8 | 10.1 | 10.8 | 12.1 | 14.8 |
| I1-IQ2_XS | 7.86 GiB | 9.3 | 9.5 | 9.8 | 10.5 | 11.8 | 14.5 |
| I1-IQ2_XXS | 7.16 GiB | 8.6 | 8.8 | 9.1 | 9.8 | 11.1 | 13.8 |
| I1-IQ1_M | 6.33 GiB | 7.8 | 8.0 | 8.3 | 9.0 | 10.3 | 12.9 |
| I1-IQ1_S | 5.83 GiB | 7.3 | 7.5 | 7.8 | 8.5 | 9.8 | 12.5 |
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.