How to track the carbon footprint of your computational pipeline
The neuroimaging pipeline WG has developed a few tools to track compute costs and consequent carbon footprint of neuroimaging pipelines. These leverage existing utilities (Code Carbon, EIT) to monitor CPU and GPU power draws during processing.
- Custom BIDS-app tracker: The first tool uses a wrapper module that tracks any containerized BIDS-app pipeline. See this tutorial for tracking mriqc compute costs. Give this a try to monitor the carbon footprint of your own containerized code!
- Built-in fMRIPrep tracker: The second tool integrates the trackers within the popular fmriprep pipeline. Users can enable the tracking by simply setting a command line a option*. See these guidelines to try tracking your fmriprep processing!
- Carbon Footprint Predictor: Lastly, there is also an existing Carbon Tracker to predict the carbon footprint of deep learning models. You can use this to energy-optimize model training and selection procedures.
* This is currently part of a dev branch which is being reviewed for a PR.
For more information on underlying trackers used in these tools: