Can I run gemma-2-9b on a Radeon RX 6500 XT?
Not at these settings. No indexed quantization of gemma-2-9b fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 3.54 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◐ |
|---|---|---|---|---|---|---|---|
| Q8_0 | 9.15 GiB | 10.5 | 10.7 | 11.0 | 11.8 | 13.3 | 16.2 |
| Q6_K | 7.07 GiB | 8.4 | 8.6 | 9.0 | 9.7 | 11.2 | 14.1 |
| Q5_1 | 6.52 GiB | 7.8 | 8.0 | 8.4 | 9.1 | 10.6 | 13.6 |
| Q5_K_M | 6.19 GiB | 7.5 | 7.7 | 8.1 | 8.8 | 10.3 | 13.2 |
| Q5_K_S | 6.04 GiB | 7.3 | 7.6 | 7.9 | 8.7 | 10.1 | 13.1 |
| Q5_0 | 6.04 GiB | 7.3 | 7.6 | 7.9 | 8.7 | 10.1 | 13.1 |
| Q4_1 | 5.55 GiB | 6.9 | 7.1 | 7.4 | 8.2 | 9.7 | 12.6 |
| Q4_K_M | 5.37 GiB | 6.7 | 6.9 | 7.3 | 8.0 | 9.5 | 12.4 |
| Q4_K_S | 5.10 GiB | 6.4 | 6.6 | 7.0 | 7.7 | 9.2 | 12.2 |
| Q4_0 | 5.07 GiB | 6.4 | 6.6 | 7.0 | 7.7 | 9.2 | 12.1 |
| Q3_K_L | 4.78 GiB | 6.1 | 6.3 | 6.7 | 7.4 | 8.9 | 11.8 |
| Q3_K_M | 4.43 GiB | 5.7 | 6.0 | 6.3 | 7.1 | 8.5 | 11.5 |
| Q3_K_S | 4.04 GiB | 5.3 | 5.6 | 5.9 | 6.7 | 8.1 | 11.1 |
| Q2_K | 3.54 GiB | 4.9 | 5.1 | 5.4 | 6.2 | 7.6 | 10.6 |
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 4,096-token window rather than the full context.