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 q8_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 | 1907.8 | 1908.6 | 1910.0 | 1912.9 | 1918.7 | 1930.3 |
| Q8_0 | 1012.92 GiB | 1014.6 | 1015.3 | 1016.8 | 1019.7 | 1025.5 | 1037.1 |
| UD-Q6_K | 788.42 GiB | 790.1 | 790.8 | 792.3 | 795.2 | 801.0 | 812.6 |
| UD-Q5_K_M | 705.94 GiB | 707.6 | 708.4 | 709.8 | 712.7 | 718.5 | 730.1 |
| Q5_K_M | 704.85 GiB | 706.5 | 707.3 | 708.7 | 711.6 | 717.4 | 729.1 |
| UD-Q5_K_S | 664.21 GiB | 665.9 | 666.6 | 668.1 | 671.0 | 676.8 | 688.4 |
| Q4_1 | 596.26 GiB | 598.0 | 598.7 | 600.1 | 603.0 | 608.8 | 620.5 |
| UD-Q4_K_M | 586.37 GiB | 588.1 | 588.8 | 590.2 | 593.1 | 599.0 | 610.6 |
| Q4_K_M | 585.99 GiB | 587.7 | 588.4 | 589.9 | 592.8 | 598.6 | 610.2 |
| Q4_K_S | 557.32 GiB | 559.0 | 559.7 | 561.2 | 564.1 | 569.9 | 581.5 |
| UD-Q4_K_S | 548.12 GiB | 549.8 | 550.5 | 552.0 | 554.9 | 560.7 | 572.3 |
| Q4_0 | 539.02 GiB | 540.7 | 541.4 | 542.9 | 545.8 | 551.6 | 563.2 |
| IQ4_NL | 537.62 GiB | 539.3 | 540.0 | 541.5 | 544.4 | 550.2 | 561.8 |
| IQ4_XS | 508.09 GiB | 509.8 | 510.5 | 512.0 | 514.9 | 520.7 | 532.3 |
| UD-IQ4_NL | 466.84 GiB | 468.5 | 469.3 | 470.7 | 473.6 | 479.4 | 491.0 |
| UD-IQ4_XS | 457.00 GiB | 458.7 | 459.4 | 460.9 | 463.8 | 469.6 | 481.2 |
| IQ3_M | 454.54 GiB | 456.2 | 457.0 | 458.4 | 461.3 | 467.1 | 478.7 |
| Q3_K_L | 452.86 GiB | 454.5 | 455.3 | 456.7 | 459.6 | 465.4 | 477.1 |
| Q3_K_M | 434.60 GiB | 436.3 | 437.0 | 438.5 | 441.4 | 447.2 | 458.8 |
| IQ3_XS | 434.58 GiB | 436.3 | 437.0 | 438.4 | 441.4 | 447.2 | 458.8 |
| UD-Q3_K_M | 428.10 GiB | 429.8 | 430.5 | 432.0 | 434.9 | 440.7 | 452.3 |
| Q3_K_S | 414.26 GiB | 415.9 | 416.7 | 418.1 | 421.0 | 426.8 | 438.5 |
| IQ3_XXS | 397.96 GiB | 399.6 | 400.4 | 401.8 | 404.7 | 410.5 | 422.2 |
| UD-IQ3_XXS | 384.32 GiB | 386.0 | 386.7 | 388.2 | 391.1 | 396.9 | 408.5 |
| UD-IQ3_S | 351.94 GiB | 353.6 | 354.4 | 355.8 | 358.7 | 364.5 | 376.1 |
| IQ3_S | 350.82 GiB | 352.5 | 353.2 | 354.7 | 357.6 | 363.4 | 375.0 |
| Q2_K_L | 334.58 GiB | 336.3 | 337.0 | 338.4 | 341.4 | 347.2 | 358.8 |
| Q2_K | 333.73 GiB | 335.4 | 336.1 | 337.6 | 340.5 | 346.3 | 357.9 |
| IQ2_M | 321.02 GiB | 322.7 | 323.4 | 324.9 | 327.8 | 333.6 | 345.2 |
| IQ2_S | 297.46 GiB | 299.1 | 299.9 | 301.3 | 304.2 | 310.0 | 321.7 |
| UD-IQ2_M | 295.48 GiB | 297.2 | 297.9 | 299.3 | 302.3 | 308.1 | 319.7 |
| UD-IQ2_XXS | 295.37 GiB | 297.1 | 297.8 | 299.2 | 302.1 | 308.0 | 319.6 |
| IQ2_XS | 285.38 GiB | 287.1 | 287.8 | 289.2 | 292.2 | 298.0 | 309.6 |
| UD-IQ1_M | 283.21 GiB | 284.9 | 285.6 | 287.1 | 290.0 | 295.8 | 307.4 |
| IQ2_XXS | 256.27 GiB | 258.0 | 258.7 | 260.1 | 263.0 | 268.9 | 280.5 |
| IQ1_M | 220.44 GiB | 222.1 | 222.9 | 224.3 | 227.2 | 233.0 | 244.6 |
| IQ1_S | 197.49 GiB | 199.2 | 199.9 | 201.4 | 204.3 | 210.1 | 221.7 |
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.