AMI公司推出的开源固件创新方案旨在满足机架级AI基础设施的可扩展性需求.pdf

1、Sanjoy MaityChief Executive Officer,AMIObjectivesScalabilitySecurityStandardization Implementation Complexity CommoditizationSustainabilityOCP Mission:To design,use,and enable mainstream delivery of the most efficient designs for scalable computing.Key Challenges to address:Disaggregated AI Infrastr

2、ucture poses huge Manageability-Firmware complexityAMI driving the disaggregated AI infrastructure solutionServerBoard(CPU x n)ServerBoard(CPU x n)Highspeed InterconnectHighspeed InterconnectBack End NetworkAcceleratorBoard(GPU x n)Front End NetworkLiquid CoolManagerRack ManagementHigh-Speed TOR Swi

3、tchAI SERVERINFRASTRUCTUREPower ShelfA/C PowerDistributionAcceleratorBoard(GPU x n)Customer Workloads:Training,Inference,Storage,etc.SERVER:ACCELERATOR:NETWORK INTERCONNECT:TOR SWITCH:RACK/POWER:THERMAL/COOLING MANAGER:(BMC)(AMC)(SMC)(SMC)(Rack-SMC)(SMC)Platform Manageability FirmwareInfrastructure

4、Manageability FirmwareInnovating&contributing to open-source platform firmware,manageability and securityLeading“Chiplet”enablement firmware Standardizing Cooling solutions Driving OpenRMC&Rack Scale management specificationProviding manageability in every increasing IT white space Scalability Chall

5、enges in the AI Datacenters with heterogeneous disaggregated infrastructure BUILD SECURE SCALE DELIVERFOUNDATIONAL GOALSPLATFORM ECOSYSTEMX86ARMRISC-VDC-SCMBMCROTBIOSModular PlatformCompute Board(HPM)PLATFORMFIRMWARECPMPLEXITIESBOOTMANAGESECUREProcessorArchitecturesBMCVarietyPROTVarietyInteroperabil

6、ity,Heterogeneity,SecurityMotivation for Open-Source&Open-ArchitectureSpeed of InnovationUniform Security PracticesIndustryStandardization1231Use an upstream version for experimentation,POC,and innovationsCreate&maintain a long-term supported downstream codebase for actual product deliveryTest&valid