Sustainable Data Centers start with Efficient Computing.pdf

1、EMPOWERING OPEN3Data Centers Contribute 3%of Global CO2Share of Global CO2Emissions by Sector2.5-3.7%of Global Emissions attributed toCloud ComputingGlobal Imperative to Increase Data Center EfficiencyDataCentersShippingAviationRiceCultivationTobacco&Food Processing4%3%2%1%0%1 Source:Climatiq:Measur

2、ing GHG in Data CentersData Center Energy Costs Rising Rapidly1 Source:Eurostat Non-household Energy Price Increases.Global Challenges IrelandAmsterdamSingaporeFrankfurtUnited KingdomUS (VA&AZ)Electricity Price Increases across the EU 2021-20221Power cost increases 2-3X from 2021Mainframe ComputersM

3、ainframe Computers(1960s(1960s-1980s)1980s)Client ServerClient Server(1980s(1980s-2010s)2010s)CloudCloud(2010s(2010s-)Cloud Native Processing:The Future of Scale-outSoftware:SpecializedSoftware:Software:MonolithicMicroservicesCloud Native Processors are High Performance ScaleCloud Native Processors

4、are High Performance Scale-out Compute with Revolutionary Energy Efficiency out Compute with Revolutionary Energy Efficiency Traditional Techniques No Longer ScalePower Optimized Consistent PerformanceLinear Core ScalingHigh Performance General-Purpose CoresParadigm ShiftTurbo FrequencyHyperthreadin

5、gScale Up AcceleratorsAmperes Architecture Delivers Performance and Efficiency Cloud Native ProcessorsLegacy x86 ProcessorsAmpere Altra MaxIntel Xeon“Icelake”AMD EPYC“Milan”Twice the PerformanceHalf the Power Lower Is Better POWEREst.SPECrate2017_int_base Performance2Higher Is Better Power Optimized

6、 Consistent PerformanceLinear Core ScalingHigh Performance General-Purpose CoresParadigm ShiftTurbo FrequencyHyperthreadingScale Up AcceleratorsLegacy X86Vendor BLegacy X86Vendor AMetrics that Scale:Performance per RackComposite Web ServiceWeb TierAppPerf vs Vendor APerf vs Vendor BFront EndNGINX3.5