rseng-green-computinglisted

# Environmentally sustainable research computing Research computing has a real carbon footprint - large simulations, ML training runs and always-on services consume energy at a scale funders and institutions increasingly ask about. The GREENER principles for computational science frame the practice: Governance, Responsibility, Estimation, Energy and embodied impacts, New collaborations, Education and Research. For an agent the working core is: measure first, reduce second, schedule smart, report honestly. ## Measure before optimizing Never guess at footprint - estimate it: - CodeCarbon instruments Python code with a few lines (or a CLI wrapper) and estimates energy plus location-adjusted CO2e; add it to representative runs, not every run. - For non-Python or cluster workloads, estimate from job accounting (CPU/GPU hours x hardware power draw x facility PUE x grid carbon intensity) and state the assumptions. - Record estimates alongside results the same way runtimes are recorded, so the cost of a paper's computations is reportable. The Software Carbon Intensity (SCI) specification (an ISO standard) gives a defensible formula when a formal number is needed: operational plus embodied emissions per functional unit. ## Reduce Order interventions by leverage, and quantify the win when possible: 1. Compute less: cache intermediate results, avoid re-running unchanged pipeline stages (rseng-workflows), kill zombie jobs, right-size parameter sweeps before launch