《优化生成式人工智能工作负载以实现可持续性和成本效益.pdf》由会员分享,可在线阅读,更多相关《优化生成式人工智能工作负载以实现可持续性和成本效益.pdf(15页珍藏版)》请在三个皮匠报告上搜索。
1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 2 5 3Optimizing Generative AI workloads for sustainability and costIsha Dua Sr Solutions Architect She/Her 2025,Amazon Web Services,Inc.or its affiliates.All r
2、ights reserved.AI models are getting biggerand resources consumption is increasingEstimated total data centre electricity demand due to AI,2020-2030Sources:IEA analysis based on data from Deloitte(2024),Gartner(2024),Goldman Sachs(2024),Schneider Electric(2024),SemiAnalysis(2024),and Shehabi,et al.,
3、(2024).Relative Training Compute(FLOPs)Since 2014 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.An analysis from Goldman Sachs Research forecasts that about 60%of the increasing electricity demands from data centers will be met by burning fossil fuels,which could increase global
4、 carbon emissions by approximately 220 million tons.Goldman Sachs Research forecasts that about 60%of the increasing electricity demands from data centers will be met by burning fossil fuels,which could increase global carbon emissions by approximately 220 million tons.A 2024 article from the World
5、Economic Forum notes that the computational power required for AI is doubling approximately every 100 days.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Carbon Reduction OpportunityOptimizing workloads on AWS can lower customers associated carbon footprint by up to 99%Find the r
6、eport at https:/ 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Managed Services 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Which data is used for pre-training?Is it fine-tunable?Multilingual capabilities?Which size(number of parameters and corpus size)?Op