Can I run step-3.5-flash on a GeForce RTX 5060 Ti?
Not at these settings. No indexed quantization of step-3.5-flash fits GeForce RTX 5060 Ti at any context we compute, with q4_0 KV. The smallest shipped quantization is 38.47 GiB in weights alone, against 14.88 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| Q8_0 | 195.04 GiB | 196.6 | 197.0 | 197.8 | 199.5 | 202.9 | 209.7 |
| Q6_K | 150.81 GiB | 152.4 | 152.8 | 153.6 | 155.3 | 158.7 | 165.4 |
| Q5_K_L | 130.88 GiB | 132.4 | 132.8 | 133.7 | 135.4 | 138.8 | 145.5 |
| Q5_K_M | 130.58 GiB | 132.1 | 132.5 | 133.4 | 135.1 | 138.4 | 145.2 |
| Q5_K_S | 126.55 GiB | 128.1 | 128.5 | 129.4 | 131.0 | 134.4 | 141.2 |
| Q4_1 | 115.15 GiB | 116.7 | 117.1 | 118.0 | 119.6 | 123.0 | 129.8 |
| Q4_K_L | 112.07 GiB | 113.6 | 114.0 | 114.9 | 116.6 | 119.9 | 126.7 |
| Q4_K_M | 111.70 GiB | 113.2 | 113.7 | 114.5 | 116.2 | 119.6 | 126.3 |
| Q4_K | 110.56 GiB | 112.1 | 112.5 | 113.4 | 115.1 | 118.4 | 125.2 |
| Q4_K_S | 107.70 GiB | 109.2 | 109.7 | 110.5 | 112.2 | 115.6 | 122.3 |
| Q4_0 | 105.60 GiB | 107.1 | 107.6 | 108.4 | 110.1 | 113.5 | 120.2 |
| IQ4_NL | 103.97 GiB | 105.5 | 105.9 | 106.8 | 108.5 | 111.8 | 118.6 |
| IQ4_XS | 98.34 GiB | 99.9 | 100.3 | 101.1 | 102.8 | 106.2 | 113.0 |
| Q3_K_L | 87.26 GiB | 88.8 | 89.2 | 90.1 | 91.8 | 95.1 | 101.9 |
| Q3_K_M | 84.04 GiB | 85.6 | 86.0 | 86.8 | 88.5 | 91.9 | 98.7 |
| IQ3_M | 83.98 GiB | 85.5 | 85.9 | 86.8 | 88.5 | 91.8 | 98.6 |
| Q3_K_S | 80.06 GiB | 81.6 | 82.0 | 82.9 | 84.6 | 87.9 | 94.7 |
| IQ3_XS | 75.46 GiB | 77.0 | 77.4 | 78.3 | 80.0 | 83.3 | 90.1 |
| IQ3_XXS | 72.82 GiB | 74.4 | 74.8 | 75.6 | 77.3 | 80.7 | 87.4 |
| Q2_K_L | 65.26 GiB | 66.8 | 67.2 | 68.1 | 69.8 | 73.1 | 79.9 |
| Q2_K | 64.78 GiB | 66.3 | 66.7 | 67.6 | 69.3 | 72.6 | 79.4 |
| IQ2_M | 59.59 GiB | 61.1 | 61.5 | 62.4 | 64.1 | 67.5 | 74.2 |
| IQ2_S | 51.70 GiB | 53.2 | 53.7 | 54.5 | 56.2 | 59.6 | 66.3 |
| IQ2_XS | 51.40 GiB | 52.9 | 53.4 | 54.2 | 55.9 | 59.3 | 66.0 |
| IQ2_XXS | 44.74 GiB | 46.3 | 46.7 | 47.5 | 49.2 | 52.6 | 59.4 |
| IQ1_M | 40.17 GiB | 41.7 | 42.1 | 43.0 | 44.7 | 48.0 | 54.8 |
| IQ1_S | 38.47 GiB | 40.0 | 40.4 | 41.3 | 43.0 | 46.3 | 53.1 |
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 512-token window rather than the full context.