Tommy Tran will explore the intersection of AI and energy efficiency, focusing on how to optimize GPUs and computational resources to minimize energy consumption in AI workloads. As AI models grow increasingly complex and resource-intensive, the energy demands of training and inference are becoming a significant challenge, both financially and environmentally. His talk will begin with an overview of the energy consumption landscape in AI, supported by recent data on the carbon footprint of large-scale models.

Tommy will then dive into practical strategies for optimizing GPUs, such as power-efficient tuning, precision reduction techniques like mixed precision training, and advanced scheduling to maximize utilization while minimizing idle power draw. He’ll also discuss innovations in hardware-aware neural architecture search (NAS) and how it can identify model designs that achieve high accuracy with lower energy costs.

Finally, Tommy will showcase case studies where these techniques were applied in real-world AI systems, achieving measurable reductions in energy consumption without sacrificing model performance. This session is aimed at researchers, engineers, and decision-makers who are passionate about building AI systems that are not only powerful but also environmentally and economically sustainable