《Leveraging OCP to Create a Purpose Built ARM Processor.pdf》由会员分享,可在线阅读,更多相关《Leveraging OCP to Create a Purpose Built ARM Processor.pdf(16页珍藏版)》请在三个皮匠报告上搜索。
1、As data centers continue to evolve and optimize,the next revolution is through designing Silicon for specific needs.The ARM ecosystem paired with OCP accelerates designs with advanced capabilities.Leveraging OCP to Create a Purpose Built ARM Processor SUSTAINABILITYZachary Bobroff,Sr Director Produc
2、t Management,AMIDong Wei,Lead Standards Architect&Fellow,ArmLeveraging OCP to Create a Purpose Built ARM Processor SUSTAINABLE SCALABLE COMPUTATIONAL INFRASTRUCTUREData Processing Continues to EvolveScaling infrastructure to meet needsScaling on DemandOn-prem and cloud infrastructureSystems from mul
3、tiple manufacturersSolutions using multiple Si vendorsAccelerators and Smart NICs now commonNon-Homogenous Hardware VendorsNon-Homogenous ArchitecturesHybrid EnvironmentsAdditional Compute UnitsOptimizing compute infrastructure continues to drive additional innovationPRIVATE CLOUDPUBLIC CLOUDON PREM
4、ISEWorkload OptimizationInitially,optimizations were done in Software:Today,optimizations are also being done in Hardware:Compiler optimizations HW virtualization optimizations Container optimizations Clustered HW via fabricGPUs,ASICs and FPGAsSmart NICsML is driving the need for more compute,and op
5、timizations may have proprietary IPSuccess of Graviton at AWSCopyright Take a look at Graviton from AWS:Improved performance:Graviton processors are optimized for specific workloads,such as web servers,containerized microservices,and data processing,leading to improved performance and faster process
6、ing speeds.Higher performance per dollar and cost-effective scalingSource:https:/ Hyperscalers have already found success in creating optimized Si for their specific needsPressures on Silicon Development Entails New ApproachAdvanced nodesYieldScalingMAX DIE SIZE MAX DIE SIZE Moores LawHARDER TO ACHI