Put a meter on it

Nobody can argue with your own plug meter. This is the lesson that turns a topic you have opinions about into one you have data about.

Measure at the wall

Put a plug meter between your machine and the socket. Record idle draw for five minutes, then run a real workload — a batch of inference, a training run, an hour of your ordinary work — and record the total.

Subtract idle. What remains is the marginal cost of doing the thing, which is the figure that belongs in any comparison. Divide by the number of tasks completed.

Your own watt hours per task

Marginal energy at the wall, divided by tasks completed. Compare it to the published claim you converted last lesson.

Average intensity is the wrong multiplier

To turn watt hours into grams of CO₂ you need a carbon intensity, and there are two.

Average intensity is the whole grid's emissions divided by the whole grid's output. It is what most calculators use and it answers a question nobody asked.

Marginal intensity is the emissions of whatever generation responds to your extra demand. At 2am on a windy night, that can be near zero; on a still evening it is a gas turbine, and your marginal number is several times the average.

For a decision — should I run this now or at 3am — marginal is the only figure that means anything. Several grid operators publish it live and free.

Grams of CO₂ per task

Your measured watt hours multiplied by your grid's marginal intensity at the hour you ran it. Note the hour alongside the number; the same workload at a different time is a different answer, and that variation is the finding.

Explain why the hour matters

Explain to a colleague why running the same job at 3am can be a genuine reduction, and when it is not.