为什么处理器对人工智能推理和通用计算至关重要（由 AMD 赞助）.pdf

1、1|2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Madhu RangarajanCorp VP,Product ManagementAMDMike ThompsonDirector,Global Cloud ProductAMDKyle McLaughlinLead Director,FinOps EngineeringCVSWhy Your Processor Matte

2、rs for AI Inference and General ComputeM A M 2 1 0-S3|Corp VP,Product ManagementAMDMadhu RangarajanDirector,Global Cloud ProductAMDMike ThompsonLead Director,FinOps EngineeringCVS HealthKyle McLaughlinMarket Insight&Market Insight&Relevance of CPUs in AIRelevance of CPUs in AIFinOps onFinOps onEC2 I

3、nstances EC2 Instances Reinvesting Reinvesting Infrastructure SavingsInfrastructure SavingsMarket Insight&Relevance of CPUs in AIMadhu RangarajanCorp VP,Product ManagementAMD5|GPUsCPUsWorlds Best CPU for Cloud,Enterprise,AIOptimized Solutions based on industry standard technologies delivered through

4、 a broad ecosystem of partnersSoftwareAMD ZenSoftware StudioOpen Source,major frameworks,compilers,librariesCluster Level Reference DesignsSolutionsSee Endnote EPYC-029DAnnual Roadmap,Leadership Memory and PerformanceNetworkingDPUs,UALink andUltra Ethernet NetworkingAMD End-to-End Solutions6|x86 CPU

5、 GPU From Analytics to Generative AI to Agentic AIDataInputDataCleaningPre-ProcessingModelTrainingDeploymentAgentic AIAI Pipeline WorkloadsSmall to Medium AI and ML modelsBatch,offline&small-scale real-time inferenceLLM Training and InferenceMedium to Large Gen AI modelsHeavy large-scale inference a

6、nd advanced reasoningAI Technology Addressing Enterprise Workloads7|Jan 2023Jan 2024May 2023 Sep 2023May 2024 Sep 2024GPT 3.5Gemini 1.5 Flash BB280 x reductionIN 18 MONTHSINFERENCE PRICESThe Cost of Inference is Plummeting$20PER MILLION TOKENS in 2022$0.07PER MILLION TOKENS in 2024Source:https:/hai.