Toward Cross-Layer Energy Optimizations in AI Systems
This work tackles the energy consumption problem for AI developers and operators, but it is incremental as it outlines challenges rather than presenting a novel solution.
The paper addresses the critical energy inefficiency of AI and ML systems, particularly large models like LLMs, highlighting that training a 200-billion parameter model consumed 11.9 GWh, and proposes research directions for cross-layer optimizations to improve energy efficiency.
The "AI for Science, Energy, and Security" report from DOE outlines a significant focus on developing and optimizing artificial intelligence workflows for a foundational impact on a broad range of DOE missions. With the pervasive usage of artificial intelligence (AI) and machine learning (ML) tools and techniques, their energy efficiency is likely to become the gating factor toward adoption. This is because generative AI (GenAI) models are massive energy hogs: for instance, training a 200-billion parameter large language model (LLM) at Amazon is estimated to have taken 11.9 GWh, which is enough to power more than a thousand average U.S. households for a year. Inference consumes even more energy, because a model trained once serve millions. Given this scale, high energy efficiency is key to addressing the power delivery problem of constructing and operating new supercomputers and datacenters specialized for AI workloads. In that regard, we outline software- and architecture-level research challenges and opportunities, setting the stage for creating cross-layer energy optimizations in AI systems.