Can I run DeepSeek-V3.2 on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of DeepSeek-V3.2 fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 138.82 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◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 1250.09 GiB | 1251.1 | 1251.2 | 1251.4 | 1251.7 | 1252.3 | 1253.5 |
| Q8_0 | 664.30 GiB | 665.4 | 665.4 | 665.6 | 665.9 | 666.5 | 667.7 |
| Q6_K | 513.45 GiB | 514.5 | 514.6 | 514.7 | 515.0 | 515.6 | 516.8 |
| Q5_K_M | 443.46 GiB | 444.5 | 444.6 | 444.7 | 445.0 | 445.6 | 446.9 |
| Q5_K_S | 430.86 GiB | 431.9 | 432.0 | 432.1 | 432.4 | 433.0 | 434.3 |
| Q4_1 | 391.87 GiB | 392.9 | 393.0 | 393.2 | 393.5 | 394.1 | 395.3 |
| Q4_K_M | 377.56 GiB | 378.6 | 378.7 | 378.8 | 379.1 | 379.7 | 380.9 |
| Q4_K_S | 354.89 GiB | 355.9 | 356.0 | 356.2 | 356.5 | 357.1 | 358.3 |
| Q4_0 | 353.99 GiB | 355.0 | 355.1 | 355.3 | 355.6 | 356.2 | 357.4 |
| IQ4_NL | 353.10 GiB | 354.2 | 354.2 | 354.4 | 354.7 | 355.3 | 356.5 |
| IQ4_XS | 333.74 GiB | 334.8 | 334.9 | 335.0 | 335.3 | 335.9 | 337.1 |
| Q3_K_M | 298.24 GiB | 299.3 | 299.4 | 299.5 | 299.8 | 300.4 | 301.6 |
| Q3_K_S | 270.50 GiB | 271.6 | 271.6 | 271.8 | 272.1 | 272.7 | 273.9 |
| UD-IQ3_XXS | 254.37 GiB | 255.4 | 255.5 | 255.7 | 256.0 | 256.6 | 257.8 |
| Q2_K_L | 228.72 GiB | 229.8 | 229.8 | 230.0 | 230.3 | 230.9 | 232.1 |
| Q2_K | 228.51 GiB | 229.6 | 229.6 | 229.8 | 230.1 | 230.7 | 231.9 |
| UD-IQ2_M | 212.44 GiB | 213.5 | 213.6 | 213.7 | 214.0 | 214.6 | 215.8 |
| UD-IQ2_XXS | 202.22 GiB | 203.3 | 203.3 | 203.5 | 203.8 | 204.4 | 205.6 |
| UD-IQ1_M | 185.62 GiB | 186.7 | 186.8 | 186.9 | 187.2 | 187.8 | 189.0 |
| UD-IQ1_S | 171.48 GiB | 172.5 | 172.6 | 172.8 | 173.1 | 173.7 | 174.9 |
| IQ2_XXS | 162.59 GiB | 163.6 | 163.7 | 163.9 | 164.2 | 164.8 | 166.0 |
| UD-TQ1_0 | 150.20 GiB | 151.3 | 151.3 | 151.5 | 151.8 | 152.4 | 153.6 |
| IQ1_M | 138.82 GiB | 139.9 | 140.0 | 140.1 | 140.4 | 141.0 | 142.2 |
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, this model uses latent attention and allocates no V cache at all, so any formula reading num_key_value_heads overstates its cache by more than an order of magnitude.