Can I run UI-TARS-1.5-7B on a Radeon Pro W6400?
Not at these settings. No indexed quantization of UI-TARS-1.5-7B fits Radeon Pro W6400 at any context we compute, with q4_0 KV. The smallest shipped quantization is 2.81 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◐ |
|---|---|---|---|---|---|---|---|
| F16 | 14.19 GiB | 15.2 | 15.3 | 15.4 | 15.6 | 16.1 | 17.1 |
| Q8_0 | 7.54 GiB | 8.6 | 8.6 | 8.7 | 9.0 | 9.5 | 10.5 |
| Q6_K | 5.82 GiB | 6.8 | 6.9 | 7.0 | 7.3 | 7.8 | 8.7 |
| Q5_K_M | 5.07 GiB | 6.1 | 6.1 | 6.3 | 6.5 | 7.0 | 8.0 |
| Q5_K_S | 4.95 GiB | 6.0 | 6.0 | 6.2 | 6.4 | 6.9 | 7.9 |
| Q4_K_M | 4.36 GiB | 5.4 | 5.4 | 5.6 | 5.8 | 6.3 | 7.3 |
| Q4_K_S | 4.15 GiB | 5.2 | 5.2 | 5.4 | 5.6 | 6.1 | 7.1 |
| IQ4_XS | 3.96 GiB | 5.0 | 5.0 | 5.2 | 5.4 | 5.9 | 6.9 |
| Q3_K_L | 3.81 GiB | 4.8 | 4.9 | 5.0 | 5.3 | 5.7 | 6.7 |
| Q3_K_M | 3.55 GiB | 4.6 | 4.6 | 4.7 | 5.0 | 5.5 | 6.5 |
| Q3_K_S | 3.25 GiB | 4.3 | 4.3 | 4.5 | 4.7 | 5.2 | 6.2 |
| Q2_K | 2.81 GiB | 3.8 | 3.9 | 4.0 | 4.3 | 4.7 | 5.7 |
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 32,768-token window rather than the full context.