VRAM usage for all quantization and cache format combinations. Base overhead: 0.54 GB (CUDA context + activations).
| Quantization | Cache Format | Model Weights | 8K Context | 16K Context | 32K Context | 40K Context |
|---|---|---|---|---|---|---|
| FP16 16.0 bpw | FP32 | 8.82 GB | 11.61 GB (+2.25 KV) | 13.86 GB (+4.5 KV) | 18.36 GB (+9.0 KV) | 20.61 GB (+11.25 KV) |
| FP16 16.0 bpw | FP16 | 8.82 GB | 10.49 GB (+1.12 KV) | 11.61 GB (+2.25 KV) | 13.86 GB (+4.5 KV) | 14.99 GB (+5.62 KV) |
| FP16 16.0 bpw | Q8_0 | 8.82 GB | 9.98 GB (+0.62 KV) | 10.6 GB (+1.24 KV) | 11.84 GB (+2.48 KV) | 12.46 GB (+3.09 KV) |
| FP16 16.0 bpw | FP8 (Exp) | 8.82 GB | 9.92 GB (+0.56 KV) | 10.49 GB (+1.12 KV) | 11.61 GB (+2.25 KV) | 12.17 GB (+2.81 KV) |
| FP16 16.0 bpw | Q4_0 (Exp) | 8.82 GB | 9.7 GB (+0.34 KV) | 10.04 GB (+0.67 KV) | 10.71 GB (+1.35 KV) | 11.05 GB (+1.69 KV) |
| Q8_0 8.0 bpw | FP32 | 4.41 GB | 7.2 GB (+2.25 KV) | 9.45 GB (+4.5 KV) | 13.95 GB (+9.0 KV) | 16.2 GB (+11.25 KV) |
| Q8_0 8.0 bpw | FP16 | 4.41 GB | 6.08 GB (+1.12 KV) | 7.2 GB (+2.25 KV) | 9.45 GB (+4.5 KV) | 10.58 GB (+5.62 KV) |
| Q8_0 8.0 bpw | Q8_0 | 4.41 GB | 5.57 GB (+0.62 KV) | 6.19 GB (+1.24 KV) | 7.43 GB (+2.48 KV) | 8.05 GB (+3.09 KV) |
| Q8_0 8.0 bpw | FP8 (Exp) | 4.41 GB | 5.51 GB (+0.56 KV) | 6.08 GB (+1.12 KV) | 7.2 GB (+2.25 KV) | 7.76 GB (+2.81 KV) |
| Q8_0 8.0 bpw | Q4_0 (Exp) | 4.41 GB | 5.29 GB (+0.34 KV) | 5.63 GB (+0.67 KV) | 6.3 GB (+1.35 KV) | 6.64 GB (+1.69 KV) |
| Q4_K_M 4.65 bpw | FP32 | 2.56 GB | 5.36 GB (+2.25 KV) | 7.61 GB (+4.5 KV) | 12.11 GB (+9.0 KV) | 14.36 GB (+11.25 KV) |
| Q4_K_M 4.65 bpw | FP16 | 2.56 GB | 4.23 GB (+1.12 KV) | 5.36 GB (+2.25 KV) | 7.61 GB (+4.5 KV) | 8.73 GB (+5.62 KV) |
| Q4_K_M 4.65 bpw | Q8_0 | 2.56 GB | 3.72 GB (+0.62 KV) | 4.34 GB (+1.24 KV) | 5.58 GB (+2.48 KV) | 6.2 GB (+3.09 KV) |
| Q4_K_M 4.65 bpw | FP8 (Exp) | 2.56 GB | 3.67 GB (+0.56 KV) | 4.23 GB (+1.12 KV) | 5.36 GB (+2.25 KV) | 5.92 GB (+2.81 KV) |
| Q4_K_M 4.65 bpw | Q4_0 (Exp) | 2.56 GB | 3.44 GB (+0.34 KV) | 3.78 GB (+0.67 KV) | 4.46 GB (+1.35 KV) | 4.79 GB (+1.69 KV) |
| Q4_K_S 4.58 bpw | FP32 | 2.52 GB | 5.32 GB (+2.25 KV) | 7.57 GB (+4.5 KV) | 12.07 GB (+9.0 KV) | 14.32 GB (+11.25 KV) |
| Q4_K_S 4.58 bpw | FP16 | 2.52 GB | 4.19 GB (+1.12 KV) | 5.32 GB (+2.25 KV) | 7.57 GB (+4.5 KV) | 8.69 GB (+5.62 KV) |
| Q4_K_S 4.58 bpw | Q8_0 | 2.52 GB | 3.69 GB (+0.62 KV) | 4.3 GB (+1.24 KV) | 5.54 GB (+2.48 KV) | 6.16 GB (+3.09 KV) |
| Q4_K_S 4.58 bpw | FP8 (Exp) | 2.52 GB | 3.63 GB (+0.56 KV) | 4.19 GB (+1.12 KV) | 5.32 GB (+2.25 KV) | 5.88 GB (+2.81 KV) |
| Q4_K_S 4.58 bpw | Q4_0 (Exp) | 2.52 GB | 3.4 GB (+0.34 KV) | 3.74 GB (+0.67 KV) | 4.42 GB (+1.35 KV) | 4.75 GB (+1.69 KV) |
| Q3_K_M 3.91 bpw | FP32 | 2.16 GB | 4.95 GB (+2.25 KV) | 7.2 GB (+4.5 KV) | 11.7 GB (+9.0 KV) | 13.95 GB (+11.25 KV) |
| Q3_K_M 3.91 bpw | FP16 | 2.16 GB | 3.82 GB (+1.12 KV) | 4.95 GB (+2.25 KV) | 7.2 GB (+4.5 KV) | 8.32 GB (+5.62 KV) |
| Q3_K_M 3.91 bpw | Q8_0 | 2.16 GB | 3.32 GB (+0.62 KV) | 3.93 GB (+1.24 KV) | 5.17 GB (+2.48 KV) | 5.79 GB (+3.09 KV) |
| Q3_K_M 3.91 bpw | FP8 (Exp) | 2.16 GB | 3.26 GB (+0.56 KV) | 3.82 GB (+1.12 KV) | 4.95 GB (+2.25 KV) | 5.51 GB (+2.81 KV) |
| Q3_K_M 3.91 bpw | Q4_0 (Exp) | 2.16 GB | 3.03 GB (+0.34 KV) | 3.37 GB (+0.67 KV) | 4.05 GB (+1.35 KV) | 4.38 GB (+1.69 KV) |
| Q2_K 2.63 bpw | FP32 | 1.45 GB | 4.24 GB (+2.25 KV) | 6.49 GB (+4.5 KV) | 10.99 GB (+9.0 KV) | 13.24 GB (+11.25 KV) |
| Q2_K 2.63 bpw | FP16 | 1.45 GB | 3.12 GB (+1.12 KV) | 4.24 GB (+2.25 KV) | 6.49 GB (+4.5 KV) | 7.62 GB (+5.62 KV) |
| Q2_K 2.63 bpw | Q8_0 | 1.45 GB | 2.61 GB (+0.62 KV) | 3.23 GB (+1.24 KV) | 4.47 GB (+2.48 KV) | 5.09 GB (+3.09 KV) |
| Q2_K 2.63 bpw | FP8 (Exp) | 1.45 GB | 2.55 GB (+0.56 KV) | 3.12 GB (+1.12 KV) | 4.24 GB (+2.25 KV) | 4.8 GB (+2.81 KV) |
| Q2_K 2.63 bpw | Q4_0 (Exp) | 1.45 GB | 2.33 GB (+0.34 KV) | 2.67 GB (+0.67 KV) | 3.34 GB (+1.35 KV) | 3.68 GB (+1.69 KV) |
Total VRAM = Model Weights + KV Cache + 0.54 GB overhead. Actual usage may vary ±5% based on inference engine and optimizations.
Use our calculator to see if this model fits your specific hardware configuration.