For quite some time, we have handled energy and power as physical quantities spent by running computation, and over time we developed mental models on how they are consumed, measured, and composed across the computing stack. This gave us the urge to write them down in precise language.

We're excited to announce the first version of Energy Calculus: https://ml.energy/calculus

We spent a lot of time thinking about the right way to capture small chunks of measurable computation as a mathematical object, define how they compose across the stack as computation structures get more complex, how time, energy, and uncertainty propagate, and how this leads to the characterization and composition of the time--energy Pareto frontier.

We want to maintain Energy Calculus as a continuously evolving and expanding framework. We'd love to chat with you about what you think about Energy Calculus, and we invite feedback to [email protected] as well.