All tools

Local LLM Memory Calculator

Estimate whether a local language model fits in available RAM or VRAM at a chosen quantization and context length.

Processing happens in your browser. Your inputs, files and results stay on this device.

Model and hardware

Enter available memory. For unified memory such as Apple Silicon, enter it as RAM and set VRAM to 0.

Memory estimate

See the weights, KV cache and estimated fit in RAM or VRAM.

How to estimate memory for a local LLM

Weight size depends on parameter count and average bits per parameter. Use roughly 4.8 bits for Q4_K_M, 5.7 for Q5_K_M and 8.5 for Q8_0. The KV cache grows with context length.

Choose one of four Qwen2.5 presets. For another model, use a custom configuration and look up layers, KV heads and head dimension in config.json.

Limitations

The estimate adds a runtime reserve of at least 1 GiB. Actual usage varies with the model file, runtime and concurrent requests. RAM and VRAM are evaluated separately.