NVIDIA Run_ai 和 Amazon SageMaker HyperPod 集成实现分布式训练（由 NVIDIA 赞助）.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.NVIDIA Run:ai and Amazon SageMakerHyperPodAIM281-S 2025,Amazon Web Services,Inc.or its affiliates.

2、All rights reserved.IntroductionRun:ai OverviewRun:ai+SageMaker HyperPodDemosAgenda 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.NVIDIA Run:ai GPU Orchestration Advanced Workload Scheduling Policy Driven Governa

3、nce Seamless User Experience Open,API First Architecture 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AI Infrastructure accelerated by NVIDIA Run:ai and SageMaker HyperPodImproved ROI and GPU Optimization:Maximi

4、ze ROI with GPU Scheduling,Quota Management,and GPU Fractioning to make the most of your infrastructure,both on-prem and in the cloud.Faster Time to Market:Achieve cloud-like elasticity with Run:ais scheduler,ensuring near-on-demand GPU access to accelerate AI workloads to production.Centralized Con

5、trol&Visibility:Gain real-time and historical insights across jobs,workloads,and teams with a single dashboard.Manage resource access,set compute guarantees,and control oversubscription with built-in policies.Open Architecture and Tool Flexibility:Integrate easily with any MLOPs or data science tool

6、.Empower practitioners with a user-friendly GUI,simplifying access to resources and streamlining AI workloads.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Demos 2025,Amazo