Runs locally on your NVIDIA GPU

GPU Watt WatcherRun on your GPU

A small app that shows how much energy your GPU uses for basic compute, memory, and AI-style workloads.

bash

git clone https://github.com/raffrant/gpu-watt-watcher.git
cd gpu-watt-watcher
python -m venv .venv
source .venv/bin/activate      # .venv\Scripts\Activate.ps1 on Windows
pip install -r requirements.txt
cd gpu_energy_bench
streamlit run streamlit_app.py

Requires an NVIDIA GPU with CUDA and PyTorch installed.

What you can learn

Is your workload compute-bound or memory-bound?
Does changing batch size or precision (FP32 vs FP16) reduce Joules per token?
How does your GPU's energy usage change with different workloads?