Best Practices To Build Energy-Efficient AI/ML Systems
InfoQ, Friday, May 9th, 2025
For organizations using AI/ML technologies, it is crucial to systematically track the carbon footprint of ML lifecycle and implement best practices in model development and deployment stages.
Tacking the energy demands has challenges like lack of standardized methods to calculate energy consumptions and the complexity In accurately measuring AI's carbon footprint.
Emissions can be classified into two types: operational emissions which refer to the energy cost of operating training models and inference and the cost of ML hardware support; and lifecycle emissions which include the embedded carbon emitted during the manufacturing of all components involved, ranging from chips to data center buildings.
Best practices to build a sustainable ML lifecycle include prioritizing efficient model selection, model optimizations to reduce complexity, choosing efficient hardware (CPU, GPU, and NPU), and cloud hosting versus on-premise infrastructure.
There are open-source tools like CodeCarbon and MLCarbon to track and reduce energy consumption. Cloud platforms such as Google Cloud Platform (GCP) and Amazon Web Services (AWS) enable sustainability in AI workloads by offering tools to minimize carbon footprints.