3018 - 人工智能赋能的高性能计算：IBM Cloud 的混合多云解决方案.pdf

1、IBM TechXchange 2025 conference AI-Infused HPC:IBM Clouds Hybrid,Multi-Cloud Solution IBM TechXchange|2025 IBM Corporation1#IBMTechXchangeRavi Konuru,Lead Architect,Industry SolutionsKevin Johnson,Principal Cloud Platform Technical SpecialistCore Contributors,Acknowledgements:Derren Dunn,Harsha Kris

2、hna Reddy IBM ResearchPaul Mazzurana,Piyush Chaudary,Weiming Gu IBM CloudAgendavHigh Performance Computing:ChallengesvIBMs Secure,Hybrid Solution OverviewvCase studies:vSecurity/compliance for regulated workloads vQ&AIBM TechXchange|2025 IBM CorporationvCross-site AI Workload in chip designvBursting

3、 Financial Sector workloadvSynthetic Filesystem benchmarkvDemo&Summary3High Performance Computing High Performance Computing ChallengesChallengesCompute:Never enough(Compute:Never enough(e.ge.g 100,000 cores per workload)100,000 cores per workload)Sustained use:Sustained use:Ability to efficiently u

4、se existing computing resourcesElastic use:Elastic use:Ability to creating computing resources on demand and execute jobs when existing compute pool is insufficient:within on-premise or extend to Cloud(bursting)Data Hydration and Sync(TBs in size,globally available,consistent)Data Hydration and Sync

5、(TBs in size,globally available,consistent)Hydration:Hydration:Data needs to be available with low latency where the job is executing for high performanceSynchronization:Synchronization:Data needs to be kept consistent across sites for global collaboration.Security&Compliance(For Regulated Industrie

6、s)Security&Compliance(For Regulated Industries)Core Security:Core Security:Given the type of data and the processing confidentiality,data leakages can be catastrophic,so all security both internal and external are importantCompliance:Compliance:Regulated industries have their own set of compliance r