Can I run Llama-4-Maverick-17B-128E-Instruct on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of Llama-4-Maverick-17B-128E-Instruct fits Radeon RX 6500 XT at any context we compute, with q8_0 KV. The smallest shipped quantization is 92.59 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 | 746.43 GiB | 747.8 | 748.1 | 748.9 | 750.5 | 753.7 | 760.1 |
| Q8_0 | 396.57 GiB | 397.9 | 398.3 | 399.1 | 400.7 | 403.9 | 410.2 |
| Q6_K | 306.19 GiB | 307.5 | 307.9 | 308.7 | 310.3 | 313.5 | 319.9 |
| UD-IQ2_M | 273.65 GiB | 275.0 | 275.4 | 276.2 | 277.8 | 280.9 | 287.3 |
| Q5_K_M | 264.93 GiB | 266.3 | 266.7 | 267.5 | 269.0 | 272.2 | 278.6 |
| Q5_K_S | 256.76 GiB | 258.1 | 258.5 | 259.3 | 260.9 | 264.1 | 270.4 |
| Q4_1 | 233.50 GiB | 234.8 | 235.2 | 236.0 | 237.6 | 240.8 | 247.2 |
| Q4_K_M | 226.09 GiB | 227.4 | 227.8 | 228.6 | 230.2 | 233.4 | 239.8 |
| Q4_K_S | 212.15 GiB | 213.5 | 213.9 | 214.7 | 216.3 | 219.5 | 225.8 |
| Q4_0 | 211.19 GiB | 212.5 | 212.9 | 213.7 | 215.3 | 218.5 | 224.9 |
| IQ4_NL | 210.26 GiB | 211.6 | 212.0 | 212.8 | 214.4 | 217.6 | 223.9 |
| UD-IQ4_XS | 205.52 GiB | 206.8 | 207.2 | 208.0 | 209.6 | 212.8 | 219.2 |
| IQ4_XS | 199.61 GiB | 200.9 | 201.3 | 202.1 | 203.7 | 206.9 | 213.3 |
| Q3_K_M | 177.95 GiB | 179.3 | 179.7 | 180.5 | 182.1 | 185.2 | 191.6 |
| Q3_K_S | 160.80 GiB | 162.1 | 162.5 | 163.3 | 164.9 | 168.1 | 174.5 |
| UD-IQ3_XXS | 157.69 GiB | 159.0 | 159.4 | 160.2 | 161.8 | 165.0 | 171.4 |
| Q2_K_L | 135.87 GiB | 137.2 | 137.6 | 138.4 | 140.0 | 143.2 | 149.5 |
| Q2_K | 135.64 GiB | 137.0 | 137.4 | 138.2 | 139.8 | 142.9 | 149.3 |
| UD-IQ2_XXS | 125.91 GiB | 127.2 | 127.6 | 128.4 | 130.0 | 133.2 | 139.6 |
| UD-IQ1_M | 118.78 GiB | 120.1 | 120.5 | 121.3 | 122.9 | 126.1 | 132.5 |
| UD-IQ1_S | 112.48 GiB | 113.8 | 114.2 | 115.0 | 116.6 | 119.8 | 126.2 |
| UD-TQ1_0 | 98.45 GiB | 99.8 | 100.2 | 101.0 | 102.6 | 105.8 | 112.1 |
| TQ1_0 | 92.59 GiB | 93.9 | 94.3 | 95.1 | 96.7 | 99.9 | 106.3 |
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 8,192-token window rather than the full context.