LLM VRAM Calculator
Estimate the GPU memory a model needs from its parameter count and precision.
Runs entirely in your browser. Nothing you type is uploaded.
About this tool
Enter a parameter count and a weight precision to get the weight memory. Supply the architecture — layers, KV heads, head dimension, context length and batch size — and it adds the KV cache for a detailed estimate. Without those numbers it reports weights only rather than guessing them.
Questions & answers
- How is weight memory calculated?
- Parameters multiplied by bits per weight, divided by eight. A 7B model at FP16 is 7,000,000,000 × 16 / 8 = 14 GB.
- Why is the KV cache sometimes not shown?
- Because it depends on the model's architecture, not just its size. Without the layer count, KV head count and head dimension there is no honest way to compute it, so the tool says weights only instead of inventing a number.
- Does fitting in my GPU mean it will run?
- Not necessarily. The estimate covers weights, the KV cache when known, and the allowance you set. Runtimes also need workspace for activations, and fragmentation and the driver take their share.