釋放AI伺服器與機器學習潛力李永輝IBM大中華區硬體系統部首席技術官暨傑出工程師 勢不可擋的AI市場“By 2020, 80% of Big Data and Analytics deployments will need distributed micro analytics and 40% of all business analytics software will incorporate prescriptive analytics built on cognitive computing functionality. Both of these trends require a dramatic increase in processing power that could be enabled by GPUs.” —IDC3 DataComputeAlgorithms組成AI的三個元件 各行各業的數據變換INCREASING DATA VARIETYSearch MarketingBehavioral TargetingDynamic FunnelsUser Generated ContentMobile WebSMS/MMSSentimentHD VideoSpeech To Te x tProduct/Service LogsSocial NetworkBusiness Data FeedsUser Click StreamSensorsInfotainment SystemsWearable DevicesCyberSecurity LogsConnectedVehiclesMachine DataIoTDataDynamic PricingPayment RecordPurchase DetailPurchase RecordSupport ContactsSegmentationOffer DetailsWeb LogsOffer HistoryA/B Te s t i n gBUSINESS PROCESSPETABYTESTERABYTESGIGABYTESEXABYTESZETTABYTESStreaming VideoNatural Language ProcessingWEBDIGITALAI 釋放Data的價值Language UnderstandingLanguage TranslationSpeech TranscriptionFace RecognitionMachine ReasoningObject Detectionknowledge 科技案例進展 AI訓練的數據處理Define training taskPrepare training DataData Pre-processingDNN Model selectionConfigure the training hyper-parameterDNN Model TrainingStartPackage the new DNN model together with preprocessing into inference proc.DL training framework preparationApplication development with inference APITo build a team with deep learning expertise : 2 months ~ 1 year To prepare massive training data : ~ 10 man month(s)To train a new model : 1 hour ~ weekTo give a new inference result : < 1 secondCognitive System:Provide optimized SW + HW design, and tool chains to significantly enhance•Productivity•Performance•Time to market 具備URLI的智能系統LearningDeep Learning / MachineLearningImage/ Video/ Voice RecognitionReasoningHighPerformanceDataAnalytics KnowledgeRepresentationandReasoningUnderstandNatural Language Processing (NLP) UnstructuredInformationManagementInteractiveTextToSpeech (TTS) / Speech to Text (STT) Question Answering Technology DataComputeAlgorithms組成AI的三個元件 邁向AI之路Z14 伺服器LinuxONE伺服器Power Systems UNIX 伺服器向外擴展伺服器（面向NoSQL/Hadoop）CPU + GPU 伺服器（面向AI/HPC）核心基礎架構下一代AI 工作負載企業AI 工作負載大資料工作負載任務關鍵工作負載IBM Systems 的解決方案專為顛覆當今最高級的資料應用而設計，其中不僅包括您目前所運行的任務關鍵應用，還包括下一代AI 工作負載。彈性存儲伺服器(ESS) 全新POWER9 處理器專為認知業務而設計I/O Throughput Increases 61%@ 366GB/sEnergy Efficiency Improves 45%* & Workload Optimized FrequencyImproved Thread Performance with SMT8/4, Shorter PipelineAccelerator : PCIeGen4, CAPI 2.0, OpenCAPI, NVLINK 2.0 & on-Chip SMP Interconnect & Off-Chip Accelerator EnablementOn-Chip AccelDDR4 InterfaceAccelerator SignalingAccelerator SignalingDDR4 InterfacePCIe SignalingSMP SignalingCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCorePCIeCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCoreCoreL2L3 RegionCore•14nm finFET半導體處理器•80 億個電晶體、17 層金屬堆疊•120 MB eDRAM、12 x 20 路關聯區域•芯片上頻寬達7 TB/s POWER9 的創新加速技術CAPI 2.0EnhancementOpenCAPICollaboration•一致性加速處理器介面•計算加速（機器學習、視頻、生物資訊）•存儲加速（記憶體中資料庫、存儲演算法）•網絡加速（壓縮、加密）PCIeGen 4Industry First•2 倍輸送量•後向相容•I/O 週邊設備連接•實際I/O 標準NVLINK 2.0 CPU-GPU 連接•最大輸送量高達300 GB/秒•CPU-GPU / GPU-GPU •統一記憶體訪問•DL、HPC、GPU DB 的理想之選 IBM與客戶、合作夥伴的POWER9創新案例BART SANO谷歌平臺副總裁谷歌對IBM 能夠在最新POWER 技術研發上取得進展感到非常高興。POWER9 OpenCAPI匯流排及大記憶體功能為穀歌資料中心的創新帶來了新的機會。IAN BUCKNVIDIA 總經理、副總裁加速計算IBM 致力於為客戶交付高性能的解決方案，例如適於NVIDIA® Tesla V100 GPU 和NVLink的POWER9 解決方案，旨在説明客戶加速深度學習工作負載。OpenPOWER2018 高峰會文章LimeLight是一家致力於為客戶（如BBC 和Marvel Comics）提供各種各樣的高效工具，説明他們提升數位內容流處理水準的公司；該公司表示，OpenPOWER（以及基於POWER9 的PCIeGen4）幫助他們克服了PCIeGen3 與其他伺服器一同使用時常常會出現的瓶頸，不僅加快了流處理速度，還縮短了緩衝時間。https://www.forbes.com/sites/patrickmoorhead/2018/03/19/headed-into-its-fifth-year-openpower-has-momentum-into-the-power9-generation/2/#74af40ef795a IBM POWER SYSTEMSAC922The best server of enterprise AIThe IBM Power Systems AC922 offers the fastest way to deploy deep learning frameworks and accelerated workloads –with enterprise class support.Up to 3.8X reduction in AI model training for deep learning frameworks1.8Xbetter performance of accelerated databases 新一代的NVLINK GPU 加速NVLinkDDR4NVL100 GB/sNVL100 GB/s100 GB/s100 GB/s170 GB/sV100GPUV100GPUV100 GPUIBM POWER9 CPUIBM POWER9™ 和NVLink 2.0 有助於解決您在代碼方面的PCI-E 瓶頸；相比參與測試的x86 平臺的CUDA 主機設備頻寬，可將資料傳輸速度提升5.6 倍。POWER9 是市場上唯一一款面向NVLink 2.0 而推出的處理器，從CPU 到GPU 均是如此。Source -https://developer.ibm.com/linuxonpower/perfcol/perfcol-technical/ 世界上最快的超級電腦IBM Power AC922•POWER9 CORAL 系統–Summit：橡樹嶺國家實驗室(ORNL) •計算速度超過150 Peta FLOPS，將會成為全球速度最快的超級電腦之一•相比Titan，僅需四分之一的節點，便可實現5-10 倍的應用性能提升•~3,500 x IBM Power AC922 (POWER9 + NVIDIA V100 GPU + Mellanox InfiniBand) + IBM ESS 存儲+ IBM Spectrum Computing HPC Software Stack17 最新業企版1.5PowerAI: Enterprise Software DistributionBinary Package of Major Deep Learning Frameworks with Enterprise SupportAI Vision*/ DL Impact: Tools for Ease of Dev.Graphical tools to Enhance Data Scientist Developer ExperienceDL Impact / DDL/LMS: Faster Traini