live Market average today

What does running AI on your own GPU cost?

Electricity cost per million tokens on your graphics card, and how it compares with an API price.

Your numbers

Shown by llama.cpp, Ollama, LM Studio and similar tools.

More options
75%

Token generation often does not use the full power budget.

Your result

€1.11 of electricity per million tokens. Enter an API price to compare.

€1.11per million tokens
Power while generating442 W
Time for a million tokens6.9 h
Energy per million tokens3.07 kWh
Tokens per €1.00901,155

How it works

  1. Measure your speed

    Run your model and read tokens per second from your tool. Throughput drives the cost far more than model size.

  2. Estimate power

    We take the card’s board power × the load you set, plus 60 W for the rest of the PC, divided by power supply efficiency.

  3. Cost per million

    Hours to generate a million tokens × power × your price per kWh.

Tips

Questions people ask

Is running AI locally cheaper than using an API?

On electricity alone it often is for small, fast models, and often is not for large, slow ones. The hardware itself is usually the bigger cost, and API prices keep falling, so compare honestly for your own volume.

Why does my GPU draw less than its rated power while generating text?

Generating tokens is mostly limited by memory bandwidth, not raw compute, so the card often runs below its power limit. Prompt processing and image generation can push it higher.

Where do I find tokens per second?

Most local AI tools print it after each response, including llama.cpp, Ollama (with verbose output) and LM Studio.

Related tools

Popular graphics cards

Prices: Eurostat, nrg_pc_204, EIA, Ofgem, ElCom. Updated 31 October 2026. How we calculate.