Local Models & Power
Configure your local machine's power and electricity rates to accurately track local model costs.
Local & Power Insights
Efficiency and cost savings from running models locally.
Local Model Throughput
| Model | Sessions | Total Tokens |
|---|---|---|
How to measure local power
Follow these steps to accurately measure how much electricity your local AI models use:
TokenTelemetry can automatically measure your machine's real power draw. Depending on your hardware and OS, this reads your GPU directly (e.g. nvidia-smi) or uses your system's battery discharge rate.
- If on a Mac/laptop: Unplug it from the wall charger (if plugged in, the battery isn't draining, so power can't be measured without admin privileges).
- Start a heavy prompt in Ollama or your local AI tool to put your machine under load.
- Click "Measure" below while the model is actively generating text.
TokenTelemetry will sample your power draw for 5 seconds and lock in the wattage. This ensures your local model costs are based on actual electricity usage rather than cloud API rates!
Power configuration
Set your wattage and local electricity rate. These numbers are used to price sessions on local models (Ollama, vLLM, etc.) and calculate your carbon footprint.