利用CXL架构提升人工智能和高性能计算应用性能.pdf

1、1|2024 SNIA.All Rights Reserved.Increasing AI and HPC Application Performance with CXL FabricsPresented byKurtis Bowman,AMD Sandeep Dattaprasad,Astera LabsSteve Scargall,MemVerge 2|2024 SNIA.All Rights Reserved.Meet your panelists Kurtis Bowman CXL MWG Co-ChairDirector,Server System Performance at A

2、MD Steve ScargallSenior Product Manager&Software Architect at MemVerge Sandeep DattaprasadSenior Product Manager at Astera Labs3|2024 SNIA.All Rights Reserved.CXL EcosystemGrowth of CXL ecosystem since its inception 4|2024 SNIA.All Rights Reserved.CXL Specification Feature Summary5|2024 SNIA.All Rig

3、hts Reserved.CXL 3.1 Use Cases for AI&HPC6|2024 SNIA.All Rights Reserved.CXL 3.1 Trusted Security Protocol(TSP)Allows for Virtualization-based,Trusted Execution Environments(TEEs)to host Confidential Computing Workloads(CC WL)Key Capabilities:Separation between TVM*&CSPs infrastructure(VMM)Configura

4、tion of CXL deviceEncryption of sensitive data in both Host/Device memoryCryptographically verify correct configuration of trusted computing environmentBenefits:Freedom to migrate sensitive WLs to TSP-enabled CloudsCollaboration with multiple parties for sharing dataConform to Compliance&Data sovere

5、ignty programsStrengthen Application security&Software IP protection*TVM=Trusted VM 7|2024 SNIA.All Rights Reserved.Please take a moment to rate this session.Your feedback is important to us.8|2024 SNIA.All Rights Reserved.Backup9|2024 SNIA.All Rights Reserved.Igniting Innovation:The Expanding CXL S

6、oftware UniverseRecent Highlights:Linux Kernel and QEMU continue to add CXL features-up to 3.1 QEMU has CXL expansion and sharing.DCD and MHD are in development.Kernel 6.8 added NUMA QoS:read/write latency and bandwidth for devices Kernel 6.9 added Weighted NUMA Interleaving Fabr