LLM VRAM Estimator
Processing locally
Estimate weight memory for a parameter count and data type. INT4 uses 0.5 bytes per parameter.
- 1Enter input
- 2Set options
- 3Get results
This tool runs in your browser. Nothing is sent to a server.
Results
Waiting for inputProcess your input to preview the results here.
How it works
- Enter parameters in billions as a number.
- Choose a weight type. INT4 is 0.5 bytes per parameter.
- Process to see decimal GB and binary GiB.
No files. Settings may be shared; nothing is uploaded.
Weight storage only. KV cache and activations are omitted.
Supported inputs and outputs
No file formats. Numeric parameter count and weight type.
Example
8 billion INT4 parameters → 4.000 GB = 3.725 GiB.
Questions and answers
Does this include KV cache?
No. The estimate is weight storage only.