Can I run Laguna-S-2.1 on a AMD Ryzen AI Max 385 (Radeon 8050S)?
Not at these settings. No indexed quantization of Laguna-S-2.1 fits AMD Ryzen AI Max 385 (Radeon 8050S) at any context we compute, with q8_0 KV. The smallest shipped quantization is 23.15 GiB in weights alone, against 22.32 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 219.05 GiB | 219.9 | 220.0 | 220.2 | 220.6 | 221.4 | 223.0 |
| F16 | 219.05 GiB | 219.9 | 220.0 | 220.2 | 220.6 | 221.4 | 223.0 |
| Q8_0 | 119.91 GiB | 120.8 | 120.9 | 121.1 | 121.5 | 122.3 | 123.9 |
| Q6_K_L | 94.83 GiB | 95.7 | 95.8 | 96.0 | 96.4 | 97.2 | 98.8 |
| UD-Q6_K | 91.19 GiB | 92.1 | 92.2 | 92.4 | 92.8 | 93.6 | 95.2 |
| Q6_K | 89.93 GiB | 90.8 | 90.9 | 91.1 | 91.5 | 92.3 | 93.9 |
| Q4_K_M | 89.44 GiB | 90.3 | 90.4 | 90.6 | 91.0 | 91.8 | 93.4 |
| UD-Q5_K_M | 81.83 GiB | 82.7 | 82.8 | 83.0 | 83.4 | 84.2 | 85.8 |
| Q5_K_L | 78.39 GiB | 79.3 | 79.4 | 79.6 | 80.0 | 80.8 | 82.4 |
| Q5_K_M | 78.22 GiB | 79.1 | 79.2 | 79.4 | 79.8 | 80.6 | 82.2 |
| UD-Q5_K_S | 76.98 GiB | 77.9 | 78.0 | 78.2 | 78.6 | 79.4 | 81.0 |
| Q5_K_S | 75.72 GiB | 76.6 | 76.7 | 76.9 | 77.3 | 78.1 | 79.7 |
| Q4_1 | 68.96 GiB | 69.9 | 70.0 | 70.2 | 70.6 | 71.3 | 72.9 |
| UD-Q4_K_M | 68.10 GiB | 69.0 | 69.1 | 69.3 | 69.7 | 70.5 | 72.1 |
| Q4_K_S | 64.36 GiB | 65.3 | 65.4 | 65.6 | 66.0 | 66.8 | 68.3 |
| UD-Q4_K_S | 63.88 GiB | 64.8 | 64.9 | 65.1 | 65.5 | 66.3 | 67.9 |
| Q4_0 | 62.44 GiB | 63.3 | 63.4 | 63.6 | 64.0 | 64.8 | 66.4 |
| IQ4_NL | 62.32 GiB | 63.2 | 63.3 | 63.5 | 63.9 | 64.7 | 66.3 |
| IQ4_XS | 58.98 GiB | 59.9 | 60.0 | 60.2 | 60.6 | 61.4 | 63.0 |
| UD-IQ4_NL | 54.71 GiB | 55.6 | 55.7 | 55.9 | 56.3 | 57.1 | 58.7 |
| UD-IQ4_XS | 53.61 GiB | 54.5 | 54.6 | 54.8 | 55.2 | 56.0 | 57.6 |
| IQ3_M | 52.60 GiB | 53.5 | 53.6 | 53.8 | 54.2 | 55.0 | 56.6 |
| Q3_K_L | 52.33 GiB | 53.2 | 53.3 | 53.5 | 53.9 | 54.7 | 56.3 |
| Q3_K_M | 50.33 GiB | 51.2 | 51.3 | 51.5 | 51.9 | 52.7 | 54.3 |
| UD-Q3_K_M | 50.31 GiB | 51.2 | 51.3 | 51.5 | 51.9 | 52.7 | 54.3 |
| IQ3_XS | 50.29 GiB | 51.2 | 51.3 | 51.5 | 51.9 | 52.7 | 54.3 |
| Q3_K_S | 48.03 GiB | 48.9 | 49.0 | 49.2 | 49.6 | 50.4 | 52.0 |
| IQ3_XXS | 46.07 GiB | 47.0 | 47.1 | 47.3 | 47.7 | 48.5 | 50.1 |
| UD-IQ3_S | 45.10 GiB | 46.0 | 46.1 | 46.3 | 46.7 | 47.5 | 49.1 |
| UD-IQ3_XXS | 41.24 GiB | 42.1 | 42.2 | 42.4 | 42.8 | 43.6 | 45.2 |
| Q2_K_L | 39.20 GiB | 40.1 | 40.2 | 40.4 | 40.8 | 41.6 | 43.2 |
| Q2_K | 38.92 GiB | 39.8 | 39.9 | 40.1 | 40.5 | 41.3 | 42.9 |
| IQ2_M | 37.19 GiB | 38.1 | 38.2 | 38.4 | 38.8 | 39.6 | 41.2 |
| UD-IQ2_M | 34.71 GiB | 35.6 | 35.7 | 35.9 | 36.3 | 37.1 | 38.7 |
| UD-IQ2_XXS | 34.64 GiB | 35.5 | 35.6 | 35.8 | 36.2 | 37.0 | 38.6 |
| IQ2_S | 33.77 GiB | 34.7 | 34.8 | 35.0 | 35.4 | 36.2 | 37.8 |
| UD-IQ1_M | 33.19 GiB | 34.1 | 34.2 | 34.4 | 34.8 | 35.6 | 37.2 |
| IQ2_XS | 33.12 GiB | 34.0 | 34.1 | 34.3 | 34.7 | 35.5 | 37.1 |
| UD-IQ1_S | 31.45 GiB | 32.3 | 32.4 | 32.6 | 33.0 | 33.8 | 35.4 |
| IQ2_XXS | 29.82 GiB | 30.7 | 30.8 | 31.0 | 31.4 | 32.2 | 33.8 |
| IQ1_M | 25.75 GiB | 26.6 | 26.7 | 26.9 | 27.3 | 28.1 | 29.7 |
| IQ1_S | 23.15 GiB | 24.0 | 24.1 | 24.3 | 24.7 | 25.5 | 27.1 |
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 512-token window rather than the full context.