Google has significantly reduced the environmental impact of its AI models, reporting a 30-fold decrease in energy consumption for its Gemini AI over the past year. The company introduced a new methodology for measuring energy, water, and carbon emissions from its AI systems. Key findings revealed a 33x reduction in median energy consumption per text prompt, now averaging around 0.24 watt-hours, while the carbon footprint has dropped by 44x, with each query processing emitting just 0.03 grams of CO2 equivalent and using 0.26 milliliters of water. This success stems from a comprehensive “full stack” strategy that optimizes hardware and software across data centers, including advancements in model architecture and custom-designed Tensor Processing Units (TPUs). Google emphasizes the importance of transparency and thorough measurement methods to accurately assess AI’s environmental footprint. As AI energy demands grow, these innovations highlight Google’s commitment to sustainability and energy efficiency.
Source link
Share
Read more