VRAM Calculator
How much GPU memory does a model really need? Choose a model, a quantization level and a context length — and see which GPUs it fits.
Estimated VRAM needed
≈ 6.0 GB
| Weights (Q4 (4-bit)) | 4.5 GB |
|---|---|
| KV cache (8K context) | ≈ 0.5 GB |
| Framework overhead | ≈ 1.0 GB |
8 GB card
Fits — 2.0 GB to spare
12 GB card
Fits — 6.0 GB to spare
16 GB card
Fits — 10.0 GB to spare
24 GB card
Fits — 18.0 GB to spare
Rule of thumb: Q4 is the sweet spot for most GPUs. Estimates are approximate — real usage varies by framework and prompt length.