Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

HPC DAY 2017 | HPE Strategy And Portfolio for AI, BigData and HPC

481 views

Published on

HPC DAY 2017 - http://www.hpcday.eu/

HPE Strategy And Portfolio for AI, BigData and HPC

Volodymyr Saviak | CEE HPC & POD Sales Manager at HPE

Published in: Technology
  • Be the first to comment

HPC DAY 2017 | HPE Strategy And Portfolio for AI, BigData and HPC

  1. 1. HPE HPC Strategy Volodymyr Saviak, HPE HPC Sales Manager HPE HPC Business Unit, October the 17th, 2017
  2. 2. Agenda 1. HPC Inspirational Video – 7 minutes 2. HPC BU Introduction - HPE Leadership in HPC – 5 minutes 3. HPC Strategy – 5 minutes 4. HPC Current Portfolio – 5 minutes 5. Memory Driven Computing – 10 minutes 6. AI Inspirational Video – 3 minutes 7. AI... more about AI – 15 minutes 2
  3. 3. HPC Inspirational Video 3
  4. 4. HPE Leadership in HPC 4
  5. 5. Technology features – key points to remember HPC – High Performance Computing High Performance High Density Fast Interconnects Scalable Storage Highly Efficient Infrastructure
  6. 6. HPC Solutions Business Unit HPE HPC BU Solutions Areas HighPerformance Computing - Monte Carlo simulations - Oil & Gas computations - Manufacturing, Intelligence - Life sciences (Bio, Chem,…) AI&BigDataapplications - Deep Learning, AI - HPDA (Hadoop, SPARK) - In memory compute & DB - Rendering, Content Scale-OUTStorage - Scale out Storage - Media assets archives - High Performance Storage - Video surveillance PerformanceOptimized Datacenters - Modular datacenters - Mobile datacenters - Green DC (low PUE) - EMI/EMR protected DC
  7. 7. We design and deliver a complete customer-specified solution, including application software if needed (often we stop with middleware), delivered pre-built and tested to the highest level of quality, ready to plug in and switch on with the shortest time to acceptance. What do deliver?
  8. 8. Hewlett Packard Enterprise Completes Acquisition of SGI 8
  9. 9. HPE is a proven leader in the high end Supercomputing Segment 9 Analysis – Summary – HPE again #1 position - 128 systems (26%) – Lenovo #2, with 84 systems (17%) – Cray, #3, with 60 systems (12%) – SGI, #7 with 25 systems (5%) Vendor Top 25 Top 50 Top 100 Top 500 HPE 0 2 3 127 SGI 4 5 10 26 Cray 11 18 30 60 Lenovo 0 1 6 84 Comparison - # of Systems TOP 500: 47th Edition Top 500 – Vendor Comparison enhanced Top 100 HPE + SGI leadership
  10. 10. HPC Strategy 10
  11. 11. HPE Strategy – Accelerate HPC leadership today and into the future – NRE Efforts – Forward Selling – Early Ships – Time To Market Solutions – Risk Compliant Archive – Trade and Match – Quantitative Finance Library – Next Gen Sequencing – CAE Solution HPC Advanced Technology and Development HPC and AI Compute Solutions HPC and AI Storage Solutions Deep Learning Solutions Optimized Platforms Horizon 2 next 12 to 24 months – Software Stacks – Lustre – Remote Graphics – Cognitive Toolkit Metrics – Domain Expertise – Customer Loyalty – Share Growth – Innovation HPC and AI Market Leadership Horizon 1 6 to 12 months HPC and AI Markets / Industries Financial Services Industry Life Sciences, Health Oil & Gas, Energy ManufacturingEDA / CAE Academia, Research Government Weather We are in the high performance computing solutions business
  12. 12. PathForward Exascale program Ensures US competitiveness in the global market – PathForward is a Department of Energy (DOE) Non-Recurring Engineering (NRE) Initiative – Central element of DOE’s Exascale Compute Program (ECP) Hardware Technology effort – Funding for R&D of technologies to develop the next generation compute infrastructure; includes open architectures and alternative processors – Cornerstone of U.S. scientific progress, technological innovation, economic vitality, and a strong national defense
  13. 13. Solving complex HPC and AI challenges with Hybrid Cluster New Tokyo Institute of Technology Supercomputer Key features − 540 Compute Nodes − Two (2) Intel® Xeon® E5-2680 v4 processors − Four (4) NVIDIA TESLA P100 NVLink GPUs − NVMe-compatible, high-speed 1.08 PB SSDs − Four (4) Intel Omni-Path connectors/node − Rich Fat Tree configuration − 400 Gb/s bandwidth /node TSUBAME 3.0 Supercomputer − Available for outside researchers in private sector through JHPCN1 and HPCI2 − Ranked #1 on Green500 List – most energy efficient supercomputer in the world, running on HPE infrastructure. − Supports significant AI and scientific HPC workloads, providing unprecedented ability to analyze large data sets. − Largest Tesla P100 SXM2 deployment to date with 2,160 NVLink-enabled GPUs “Through our partnership with SGI, and now HPE, the Tokyo Institute of Technology has worked successfully to deliver a converged world-leading HPC and Deep Learning platform….” - Satoshi Matsuoka, Professor and TSUBAME Leader, Tokyo Institute of Technology.. 1, 2 Reference Information provided in speaker notes
  14. 14. World’s largest chemical company creates chemistry with HPC HPE supercomputer enables global digital transformation at BASF Key features − HPE Apollo 6000 Gen10 − > 1 Petaflop using Next Gen platform − Multitude nodes − Work simultaneously on highly complex tasks − Dramatically reduce processing time BASF Supercomputer − Designed to be one of the world’s largest supercomputer − Drive digitalization of BASF's worldwide research − Shorten modeling / simulation times (months to days) − Solve complex problems while decreasing discovery time − Run virtual experiments to reduce time-to-market, lower costs “The new supercomputer will promote the application and development of complex modeling and simulation approaches, opening up completely new avenues for our research at BASF.” − Dr. Martin Brudermueller, Vice Chairman of the Board of Executive Directors and CTO, BASFBASF Cluster - HPE Factory Build in Houston, TX, May 2017
  15. 15. Exascale required to solve the world’s most complex problems Life Sciences Weather Deep Learning, IoT and Artificial Intelligence systems will need Exascale computing Material Sciences Manufacturing Today’s top 500 systems Consume 650MW of power – (> ½ a Gigawatt) Huge CO2 Footprint Aggregated compute power of ~1 ExaFLOPS Accurate regional impact assessment of climate change Accelerate and translate cancer research in RAS pathways, drug responses, and treatment strategies Additive manufacturing process design for qualifiable metal components Efficiency and performance characteristics of materials for batteries, solar cells, and optoelectronics
  16. 16. HPC Current Portfolio 16
  17. 17. − Deliver more choice / flexibility for HPC − ARM processor based system − Proof of concepts with select customers Accelerating HPC innovation for today and tomorrow New HPE SGI 8600 Next gen petaflop scale, liquid cooled supercomputer – Greater performance, scale and efficiency New HPE Apollo 6000 Gen10 Next gen air cooled, purpose built enterprise HPC solution – Best in class performance, rack scale efficiency New HPE Apollo 10 Series – Cost effective platforms for AI and emerging applications A new experience in IT security and protection New HPE Performance Software Suite: Out-of-the- box HPC stack, enhanced cluster system management and acceleration tools New Services and Consumption Model – New Advisory, Professional and Operational Services – HPE Flexible Capacity for HPC DoE PathForward Exascale Program − New Exascale program to create reference designs − Inspired by Memory-Driven Computing and Hewlett Packard Labs technologies New disruptive technology based system architecture – ARM processor based system – Proof of concepts with select customers 1 Substantiation for quantifiable benefits in speaker notes Workload optimized for extreme performance Secure, agile, flexible Compute experience Exascale and advanced technology programs – NEW collaboration for AI application in precision medicine World’s Most Secure Servers1 for HPC and AI – HPE Apollo 6000 Gen10
  18. 18. HPE purpose-built portfolio for HPC HPE Apollo 6500 Gen9 Rack-scale GPU Computing HPE Integrity Superdome X HPE Integrity MC990 X Scale-up, shared memory HPC, UV Technologies HPE Apollo 6000 Gen9 Rack-scale HPC HPE Apollo 2000 Gen9 The bridge to enterprise scale-out architecture HPE SGI 8600 Liquid cooled, delivering industry leading performance, density and efficiency HPE Apollo 6000 Gen10 Extreme Compute Performance in High Density Supercomputing / Enterprise / Commercial HPC Advisory, Professional and Operational Services – HPE Flexible Capacity for HPC, HPE Datacenter Care for Hyperscale HPC Storage Choice of Fabrics HPC Industry Solutions Weather and Climate Research Financial Services Life Sciences, Health Academia, Research, Gov’t Oil and Gas, Energy EDA / CAE Manufacturing HPE Software Open Source Software Commercial HPC Software − HPE Performance Software - Core Stack − HPE Insight Cluster Management Utility − HPE SGI Management Suite − HPE Performance Software – Message Passing Interface* HPE Apollo 4520 Arista Networking – Intel® Omni-Path Architecture – Mellanox InfiniBand – HPE FlexFabric Network HPC Data Management Framework Software Large-scale, storage virtualization & tiered data management platform HPE Performance Software Suite Emerging HPC In-memory HPC Additional Storage Options available * Available in August 2017
  19. 19. MDC Portfolio Optimized Infrastructure CAPACITY 150kW 300kW 500kW 900kW 1,100kW DC8 200kW 368U DC18 500kW 858U ENTERPRISE Availability PERFORMANCE Node Density DC10 290kW 500U DC21-600 580kW 1,050U DC44 1,014kW 2,200U DC4 50kW 200U DC5 50kW 230U DC5-50 50kW 250U DC21-300 290kW 1050U Max. Electrical Load Max. Electrical Load
  20. 20. Memory Driven Computing 20
  21. 21. The New Normal: Compute is not keeping up 21 0,3 0,8 1,2 1,8 4,4 7,9 15,8 31,6 44 0 5 10 15 20 25 30 35 40 45 50 2006 2008 2010 2012 2014 2016 2018 2020 Data (Zettabytes) Data nearly doubles every two years (2013-2020) Data growth Transistors (thousands) Single-thread Performance (SpecINT) Frequency (MHz) Typical Power (Watts) Number of Cores 1975 1980 1985 1990 1995 2000 2005 2010 2015 Microprocessors 107 106 105 104 103 102 101 100
  22. 22. We need new type of compute – Memory Driven Computing Structured data 40 petabytes Walmart’s transaction database (2017) Human interaction data 4 petabytes Per-day posting to Facebook across 1.1 billion active users (May 2016) 4kB per active user Digitization of Analog Reality 40,000 petabytes a day* 10m self-driving cars by 2020 Front camera 20MB / sec Front ultrasonic sensors 10kB / sec Infrared camera 20MB / sec Side ultrasonic sensors 100kB / sec Front, rear and top-view cameras 40MB / sec Rear ultrasonic cameras 100kB / secRear radar sensors 100kB / sec Crash sensors 100kB / sec Front radar sensors 100kB / sec * Driver assistance systems only
  23. 23. Key attributes of Memory-Driven Computing Powerful A quantum leap in performance, beyond what you can imagine Open An open architecture designed to foster a vibrant innovation ecosystem Trusted Always safe, always recoverable All the benefits without asking for sacrifice Simple Structurally simple, manageable and automatic, so that “it just works” 23
  24. 24. GPU ASIC Quantum RISC V Memory 24 Memory Memory Memory Memory SoC SoC SoC SoC Future architecture Memory-Driven Computing Today’s architecture From processor-centric computing
  25. 25. What are core Memory-Driven Computing components? 25 Combining memory and storage in a stable environment to increase processing speed and improve energy efficiency Using photonics where necessary to eliminate distance and create otherwise impossible topologies Optimizing processing from general to specific tasks Radically simplifying programming and enabling new applications that we can’t even begin to build today Fast, persistent memory Fast memory fabric Task-specific processing New and Adapted software
  26. 26. GPU ASIC Quantum RISC V Open architecture Customize the hardware to the workload 26 DRAM DRAM NVRAM Reduced cost Less energy Less space Less complex
  27. 27. Memory-Driven Computing Developer Toolkit Software already available to you ‒ Example Applications ‒ Programming and analytics tools ‒ Operating system support ‒ Emulation/simulation tools Get access to the toolkit: https://www.labs.hpe.com/the- machine/developer-toolkit Open source components Machine (Prototype) hardware Node Operating System Persistent Memory Library (pmem.io) Librarian File System (LFS) Fabric attached memory atomics library Linux for Memory-Driven Computing Example Applications Management Services Librarian Data Management & Programming Frameworks Managed data structures Sparkle Emulation/Simulation Tools Performance emulation for NVM Fabric attached memory emulation X’86 emulation (Superdome X, MC990x, ProLiant) Fault-tolerant programming Fast optimistic engine Image Search Large Scale Graph Inference Persistent memory toolkit
  28. 28. HPE introduces the world’s largest single-memory computer The prototype contains 160 terabytes of memory 28 – 160 TB of shared memory spread across 40 physical nodes, interconnected using a high-performance fabric protocol. – An optimized Linux-based operating system running on ThunderX2, Cavium’s flagship second generation dual socket capable ARMv8-A workload optimized System on a Chip. – Photonics/Optical communication links, including the new X1 photonics module, are online and operational. – Software programming tools designed to take advantage of abundant of persistent memory.
  29. 29. Transform performance with Memory-Driven programming 29 In-memory analytics 15x faster New algorithms Completely rethink Modify existing frameworks Similarity search 40x faster Financial models 10,000x faster Large-scale graph inference 100x faster
  30. 30. AI - Artificial intelligence 30
  31. 31. 31 We have large memory, memory driven computing, its much faster, but there is a problem…
  32. 32. 32 We can’t scale people…easily.
  33. 33. AI Inspirational Video – C:BACKUP_VideoGTC 2017- 'I Am AI' Opening in Keynote.mp4 33
  34. 34. Are we on the brink of a …. 34 Change 1: Moving from gather and hunting to settling down to farms and ports Change 2: Developing the printing press and industrial revolution Latest Change: The greatest change of our lives. Artificial Intelligence
  35. 35. 0 10 20 30 40 50 60 70 Market in billion US dollars 1.38 2016 2.24 2017 4.07 2018 6.63 2019 10.53 2020 16.24 2021 24.16 2022 34.38 2023 46.52 2024 59.75 2025 What is the size of the AI market? 35 1 Source : IDC IT Predictions 2017 Services App Advisory Total AI TAM 2017 TAM 2021 TAM 4yr CAGR $0.7B $2.3B 32% $0.4B $1.6B 40% $2.3B $18.5B 67% Server- ML $7.9B $31.3B 41% $3.5B $0.9BServer- DL $4.6B $4.4B 7% 48% By 2019, 40% of all digital transformation initiatives 100% of all effective IoT efforts will be supported by AI capabilities1 andBy 2018, 75% of developer teams will include AI functionality in one or more applications1
  36. 36. 36
  37. 37. AI vs Brain? AI – HPE, CMU Liberatus Brain: Kim, Les, Chou, MCAulay 10160 Poker - 2017Checkers -1995 AI: UAlberta Chinook: white Brain: Don Lafferty – red 1020 Chess -1997 AI: IBM Deep Blue: white Brain: Garry Kasparov: black 1047 AI: Google AlphaGo - black Brain: Lee Sedol - white 10171 Go - 2016
  38. 38. ‒ Search & information extraction ‒ Security/Video surveillance ‒ Self-driving cars ‒ Medical imaging ‒ Robotics ‒ Interactive voice response (IVR) systems ‒ Voice interfaces (Mobile, Cars, Gaming, Home) ‒ Security (speaker identification) ‒ Health care ‒ People with disabilities ‒ Search and ranking ‒ Sentiment analysis ‒ Machine translation ‒ Question answering ‒ Recommendation engines ‒ Advertising ‒ Fraud detection ‒ AI challenges ‒ Drug discovery ‒ Sensor data analysis ‒ Diagnostic support Where can we use deep learning today? Applications 38 TextVision Speech Other
  39. 39. Applications break down 39 Detection Look for a known object/pattern Classification Assign a label from a predefined set of labels Generation Generate content Anomaly detection Look for abnormal, unknown patterns Images Video Text Sensor Other Speech Video surveillance Speech recognition Sentiment analysis Predictive maintenance Fraud detection Image analysis
  40. 40. Where to start ? Recommend DL stack by vertical application 40 Infrastructure Frameworks Typical layers Data type Data ManufacturingVerticals Oil & gas Connected cars Voice interfaces Social media Speech Images Sensor dataVideo Small Moderate Large Convolutional Fully- connected Recurrent TensorFlow Caffe 2 CNTK … x86 GPUs FPGAs TPU ? … … Torch Neural Network sits here
  41. 41. AI expertise and solutions to “get started” with deep learning models 41 New foundation to “get started” with deep learning models Enhance employee productivity Accelerate app development with New deep learning integrated solution Pre-configured, proven hardware & software solution − Purpose-build platform − Easy to use and install − Simple management − Automated framework updates Train your teams Gain organizational competencies with Enhanced Deep Learning Institute State of the art deep learning training − Latest techniques − Software frameworks − Infrastructure requirements − Hands on, instructor led HPE Fraud Detection Solution with Kinetica − Uses deep learning techniques − Qualified with Kinetica in- memory GPU database − NVIDIA GPU accelerators Leverage “out of the box” solutions Increase security of e-commerce with Enhanced HPE Fraud Detection solution Get Started Select ideal technologies & systems Make Informed technology decisions with New HPE Deep Learning Cookbook Comprehensive technology selection tool − Estimates & refines performance − Characterizes frameworks − Recommends ideal hardware and software stacks IT Expertise Solutions
  42. 42. 42 Conclusions
  43. 43. Where would the AI road take us? 43 Advances in artificial intelligence will transform modern life by reshaping transportation, health, science, finance, and the military. “High-level machine intelligence” (HLMI) is achieved when unaided machines can accomplish every task better and more cheaply than human workers. Grace et al , When Will AI Exceed Human Performance? Evidence from AI Experts Writing a bestseller – 2049 Driving a truck - 2027 Math Research - 2060 Surgeon - 2043 Retail - 2031 Full Automation of labor – 2140
  44. 44. Thank you

×