With increasing use of AI, it is becoming more important to enhance energy efficiency and reduce CO2 emissions in software and AI processes within big data environments. This is particularly true for generative AI solutions, which require far greater computational capacity and energy resources than previous AI systems.
Telefónica aims to design a Python library capable of measuring energy consumption and carbon emissions for executions of AI software in a big data environment, before producing recommendations to reduce emissions. The solution, called Kiri, is independent of company type, cloud infrastructure providers, on-premise machines and hardware manufacturers.
Kiri follows a method that involves identifying the available hardware components for software execution and obtaining information about the computer’s geolocation. The software’s runtime is calculated, followed by a determination of the energy consumption of CPUs, GPUs and RAM, and the amount of data read and written to disk. Leveraging databases containing technical hardware information and CO2 emissions per kWh based on geolocation (by country), the method calculates CO2 emissions. This process provides a detailed assessment of the software’s environmental impact, enabling companies to introduce measures to optimise energy efficiency and reduce their carbon footprint.
Kiri provides recommendations for optimal execution times, suggests cloud locations with a lower environmental impact (due to a better renewable energy mix in each location) and guides resource allocation. Ultimately, Kiri helps organisations make informed decisions to minimise their carbon footprint and contribute to sustainable development.
