Can I run MiMo-V2.5-Pro on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of MiMo-V2.5-Pro fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 197.49 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 | 1906.14 GiB | 1907.5 | 1907.9 | 1908.6 | 1910.2 | 1913.3 | 1919.4 |
| Q8_0 | 1012.92 GiB | 1014.3 | 1014.7 | 1015.4 | 1017.0 | 1020.0 | 1026.2 |
| UD-Q6_K | 788.42 GiB | 789.8 | 790.1 | 790.9 | 792.5 | 795.5 | 801.7 |
| UD-Q5_K_M | 705.94 GiB | 707.3 | 707.7 | 708.4 | 710.0 | 713.1 | 719.2 |
| Q5_K_M | 704.85 GiB | 706.2 | 706.6 | 707.4 | 708.9 | 712.0 | 718.1 |
| UD-Q5_K_S | 664.21 GiB | 665.6 | 665.9 | 666.7 | 668.2 | 671.3 | 677.5 |
| Q4_1 | 596.26 GiB | 597.6 | 598.0 | 598.8 | 600.3 | 603.4 | 609.5 |
| UD-Q4_K_M | 586.37 GiB | 587.7 | 588.1 | 588.9 | 590.4 | 593.5 | 599.6 |
| Q4_K_M | 585.99 GiB | 587.3 | 587.7 | 588.5 | 590.0 | 593.1 | 599.3 |
| Q4_K_S | 557.32 GiB | 558.7 | 559.1 | 559.8 | 561.4 | 564.4 | 570.6 |
| UD-Q4_K_S | 548.12 GiB | 549.5 | 549.9 | 550.6 | 552.2 | 555.2 | 561.4 |
| Q4_0 | 539.02 GiB | 540.4 | 540.8 | 541.5 | 543.1 | 546.1 | 552.3 |
| IQ4_NL | 537.62 GiB | 539.0 | 539.4 | 540.1 | 541.7 | 544.7 | 550.9 |
| IQ4_XS | 508.09 GiB | 509.4 | 509.8 | 510.6 | 512.1 | 515.2 | 521.4 |
| UD-IQ4_NL | 466.84 GiB | 468.2 | 468.6 | 469.3 | 470.9 | 474.0 | 480.1 |
| UD-IQ4_XS | 457.00 GiB | 458.3 | 458.7 | 459.5 | 461.0 | 464.1 | 470.3 |
| IQ3_M | 454.54 GiB | 455.9 | 456.3 | 457.0 | 458.6 | 461.7 | 467.8 |
| Q3_K_L | 452.86 GiB | 454.2 | 454.6 | 455.4 | 456.9 | 460.0 | 466.1 |
| Q3_K_M | 434.60 GiB | 436.0 | 436.3 | 437.1 | 438.6 | 441.7 | 447.9 |
| IQ3_XS | 434.58 GiB | 435.9 | 436.3 | 437.1 | 438.6 | 441.7 | 447.8 |
| UD-Q3_K_M | 428.10 GiB | 429.4 | 429.8 | 430.6 | 432.1 | 435.2 | 441.4 |
| Q3_K_S | 414.26 GiB | 415.6 | 416.0 | 416.8 | 418.3 | 421.4 | 427.5 |
| IQ3_XXS | 397.96 GiB | 399.3 | 399.7 | 400.5 | 402.0 | 405.1 | 411.2 |
| UD-IQ3_XXS | 384.32 GiB | 385.7 | 386.1 | 386.8 | 388.4 | 391.4 | 397.6 |
| UD-IQ3_S | 351.94 GiB | 353.3 | 353.7 | 354.4 | 356.0 | 359.1 | 365.2 |
| IQ3_S | 350.82 GiB | 352.2 | 352.6 | 353.3 | 354.9 | 357.9 | 364.1 |
| Q2_K_L | 334.58 GiB | 335.9 | 336.3 | 337.1 | 338.6 | 341.7 | 347.8 |
| Q2_K | 333.73 GiB | 335.1 | 335.5 | 336.2 | 337.8 | 340.8 | 347.0 |
| IQ2_M | 321.02 GiB | 322.4 | 322.8 | 323.5 | 325.1 | 328.1 | 334.3 |
| IQ2_S | 297.46 GiB | 298.8 | 299.2 | 300.0 | 301.5 | 304.6 | 310.7 |
| UD-IQ2_M | 295.48 GiB | 296.8 | 297.2 | 298.0 | 299.5 | 302.6 | 308.7 |
| UD-IQ2_XXS | 295.37 GiB | 296.7 | 297.1 | 297.9 | 299.4 | 302.5 | 308.6 |
| IQ2_XS | 285.38 GiB | 286.7 | 287.1 | 287.9 | 289.4 | 292.5 | 298.6 |
| UD-IQ1_M | 283.21 GiB | 284.6 | 284.9 | 285.7 | 287.2 | 290.3 | 296.5 |
| IQ2_XXS | 256.27 GiB | 257.6 | 258.0 | 258.8 | 260.3 | 263.4 | 269.5 |
| IQ1_M | 220.44 GiB | 221.8 | 222.2 | 222.9 | 224.5 | 227.6 | 233.7 |
| IQ1_S | 197.49 GiB | 198.8 | 199.2 | 200.0 | 201.5 | 204.6 | 210.8 |
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 128-token window rather than the full context.