《15-SiFive-Keynote - Jack Kang FINAL.pdf》由會員分享,可在線閱讀,更多相關《15-SiFive-Keynote - Jack Kang FINAL.pdf(21頁珍藏版)》請在三個皮匠報告上搜索。
1、2023 SiFiveIt just keepsgettingbetter:RISC-V PerformanceJack Kang 剛至堅剛至堅SVP Business Development,Customer Experience(CX),Corporate Marketing SiFive Inc.August,20232023 SiFive2Its been 3+yearsSince weve been in ChinaA lot changes in 3+years2023 SiFive3RISC-V Internationalpage 150501001502002503003504
2、00450Q32015Q42015Q12016Q22016Q32016Q42016Q12017Q22017Q32017Q42017Q12018Q22018Q32018Q42018Q12019Q22019Q32019Q42019December 2019More than 435 RISC-V Members across 33 Countries Around the World44 Chip SoC,IP,FPGA9 I/O Memory,network,storage31 Services Fab,design services25 Software Dev tools,firmware,
3、OS24 Industry cloud,mobile,HPC,ML,automotive32 Research Universities,Labs,other alliances200+Individual RISC-V developers and advocates Now3.5 yearsDec 20192023 SiFive4RISC-V in ChinaNow3.5 yearsDec 2019MORE THAN 200 MEMBERS HAVE JOINED THE CRVA AND CRVIC ASSOCIATIONS AS WELL AS 33 MEMBERS IN THE GL
4、OBAL RISC-V FOUNDATION.ACROSS 2019 WEVE LED DISCUSSIONS IN 6 CITIES AS WELL AS HOSTED A DAY OF TALKS AT THE WORLD INTERNET CONFERENCE.IN NOVEMBER,WE HOSTED 500 ATTENDEES AT THE CHINA RISC-V FORUM.CHINApage 7MORE THAN 200 MEMBERS HAVE JOINED THE CRVA AND CRVIC ASSOCIATIONS AS WELL AS 33 MEMBERS IN TH
5、E GLOBAL RISC-V FOUNDATION.ACROSS 2019 WEVE LED DISCUSSIONS IN 6 CITIES AS WELL AS HOSTED A DAY OF TALKS AT THE WORLD INTERNET CONFERENCE.IN NOVEMBER,WE HOSTED 500 ATTENDEES AT THE CHINA RISC-V FORUM.CHINApage 72023 SiFive5RISC-V Performance&Ecosystem3 Years AgoNowMostly Embedded CoresApplication co
6、res outperform ArmComplete feature set32+core systemsMulti-core,multi-clusterAn Open StandardThe fastest growing ecosystemInevitable futureA Promising RoadmapsCustomer silicon shipmentsA Community ProjectA commercial successNow3.5 yearsDec 20192023 SiFive6China is vital to the inevitable RISC-V futu
7、re2023 SiFive7SiFive is expanding in China Shanghai headquarters Local team Local ecosystems&partnership2023 SiFive8Founded by the inventors of RISC-V SiFiveis the RISC-V founder&brand standard Leading the commercialization effort since 2010 Worlds largest technology work with SiFiveto adopt RISC-V
8、Largest team&investment2023 SiFive9 The largest RISC-V CPU IP Portfolio Highest performance RISC-V cores For licensing Silicon proven On our roadmap Most advanced technology Multi-core,Multi-cluster NoC and Uncore Advanced software team(Compiler,Vectors,Performance,Debug,Kernels,BSPs,World-class cus
9、tomer supportWe invest 100%ondeveloping RISC-Vsolutions2023 SiFive10New products in the last 3 years3 completely new families:Automotive,Intelligence,Performance32/64-bitreal-time scalable performancedeeplyembeddedprocessorsPerformanceIntelligence64-bithigh-performance feature-richOS capableapplicat
10、ion processors X200-SeriesAI processor for Edge and Data Center ML applications AI acceleration instructions512b vector length E2-SeriesSmallest,most efficientEssentialS2-Series64-bit,Area optimizedE6-SeriesBalancedperformanceand efficiencyS6-Series64-bit,powerefficiencyE7-Series32-bit,optimized per
11、formance S7-Series64-bit,highperformance,embeddedU6-Series64-bit,highperformanceU7-Series64-bit,superscalar performanceAutomotiveX280-AAI acceleration instructions512b vector lengthASIL BE6-A32-bit,balanced performance andefficiency ASIL B,DS7-A64-bit,highperformanceembedded ASIL DP600-Series13.1 Sp
12、ecINT2k6/GHz4-wide OoO core 128b vector lengthHypervisor ext.Vector cryptoIOMMU&AIAWorldGuardRVA22P400-Series8.6 SpecINT2k6/GHz3-wide OoO core 128b vector lengthHypervisor ext.Vector cryptoIOMMU&AIAWorldGuardRVA22P200-Series4.6 SpecINT2k6/GHz2-wide in-order core256b vector lengthWorldGuardRVA20P500-
13、Series8.6 SpecINT2k6/GHz3-wide OoO core Hypervisor ext.WorldGuardRVA202023 SiFive11Enabling next-generation Automotive Electric vehicles&Autonomous driving is disrupting the Automotive industry Massive opportunity for new market ecosystem SiFive has complete automotive-certified solutions:Applicatio
14、n Cores,Safety Island,Real-time Processor,and AI Intelligence Processors2023 SiFive12Highest performance RISC-V for Infrastructure SiFive delivered highest performance CPUs SiFive solutions are scalable to many cores Complete solutions for IP,SoC,and Chiplets RISC-V Compute+RISC-V AI2023 SiFive13AI
15、from the edgeto the cloud Generative AI and LLM workloads prove need for programmability Low power and low latency to the edge,and new programming capabilities to the cloud SiFive Intelligence and Vector solutions enable AI across all markets2023 SiFive14Power efficiency matters in Consumer and Mobi
16、le 30-40%better power efficiency than Arm Google porting Android,as leading mobile suppliers adopting RISC-V Mix and Match Big,Medium,Little size cores2023 SiFive15SiFive Performance family relentless innovationMore performanceHigher core countLeading RISC-V feature deploymentP550P650/P670P450/P470P
17、870P870-A3rdgeneration OoO coreAutomotive specific features8.6 Specint 2k6/GHz13.1 Specint 2k6/GHz8.6 Specint 2k6/GHz18 Specint 2k6/GHz2023 SiFive16SiFive Performance P400-Series and P600-SeriesHighestPerformanceOptimized performance within constrained area and power envelopeBest-in-class performanc
18、eAcceleration for media,crypto and data processingVirtualization,IOMMU,AIA,debug&trace,securityCompliant with RVA22/RVA23 profile,with support for Vector and Vector crypto extensionsBalancedPPAVectorExtensionsFeatureRichRISC-VCompliant2023 SiFive17P470 Performance and efficiencyCortex-A55:Performanc
19、e 1.88GHz measured on Acer Spin 513 Chromebook with Qualcomm Snapdragon 7c.Area:7nm TechInsights “Die photos show Cortex-A78 shortfall”,SiFive P470:Performance 2.97GHz,0.95V 32KB L1 I$and D$,2MB L2$.Area:measured in 7nm13.7654.79Compute Density(SpecINT2k6/mm)Cortex-A55P4704xCortex-A55P470Lower is be
20、tter0.50.35Area with private caches(mm)Higher is better-35%6.8819Peak Single-Thread Performance(SpecINT2k6)Cortex-A55P470Higher is better2.75x2023 SiFive18P670 Performance and efficiency18.1837Compute Density(SpecINT2k6/mm)Cortex-A78P6702xCortex-A78P670Lower is better2.141Area with private caches(mm
21、)Higher is better-50%38.937Peak Single-Thread Performance(SpecINT2k6)Cortex-A78P670Higher is betterCortex-A78:Performance 3GHz in 5nm with 32KB L1 I$and D$,512KB Private L2,4MB L3 Anandtech.Area;7nm TechInsights “Die photos show Cortex-A78 shortfall”,article SiFive P670:Performance 3.4GHz,0.95V in 5
22、nm with 32KB L1 I$and D$,256KB Private L2$,4MB L3$.Area:measured in 7nm.2023 SiFive19SiFive Provides Complete&Scalable SolutionsP870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$Multi Cluster InterconnectShared L3 CacheQoSUp to 8 memory ports(CHI,AXI)Non-Coherent I/O(AXI)SiFive IP ComplexC
23、PU ClustersScalable Coherent InterconnectAdvanced Interrupt ControllerSiFive Insight Debug&TraceIOMMUSiFive WorldGuard SecurityScalable High-Performance&High-Efficiency Cores:P870,P670,&P470(with selected Mix+Match)System IP to Enable Complete RISC-V SoC solutionsAdvanced Power ManagementP870-AcoreP
24、870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$P870-AcoreP870-AcoreP870-AcoreP870-AcoreUp to 4MB shared L2$SiFive CoresShared Cluster L2 Cache2023 SiFive20RISC-V is here,time is now 一萬年太久,只爭朝夕SiFive is here to helpThe future of Empowering