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.

GTC Taiwan 2017 主題演說

2,291 views

Published on

NVIDIA 創辦人暨執行長 黃仁勳

Published in: Technology
  • Be the first to comment

GTC Taiwan 2017 主題演說

  1. 1. Jensen Huang, Founder & CEO, GTC Taiwan 2017 A NEW COMPUTING ERA
  2. 2. 2 THE ERA OF AI PC MOBILE CLOUD AI
  3. 3. 3 TWO FORCES DRIVING THE FUTURE OF COMPUTING Deep Learning Starts AI Revolution 1980 1990 2000 2010 2020 Original data up to the year 2010 collected and plotted by M. Horowitz, F. Labonte, O. Shacham, K. Olukotun, L. Hammond, and C. Batten New plot and data collected for 2010-2015 by K. Rupp 103 105 107 1.5X per year 1.1X per year CPU Performance Stalled Transistors (thousands) Single-threaded perf
  4. 4. 4 1980 1990 2000 2010 2020 Original data up to the year 2010 collected and plotted by M. Horowitz, F. Labonte, O. Shacham, K. Olukotun, L. Hammond, and C. Batten New plot and data collected for 2010-2015 by K. Rupp 103 105 107 1.5X per year GPU Extends Performance Post-Moore’s Law Single-threaded perf RISE OF NVIDIA GPU COMPUTING GPU-Accelerated Computing 1.1X per year CUDA GPU Parallel Domain-Specialized Accelerator High Compute & Bandwidth High-Throughput GPU Plus Low-Latency CPU
  5. 5. 5 RISE OF NVIDIA GPU COMPUTING CUDA Downloads 5X in 5 Years Global GTC Attendees 10X in 5 Years GPU Developers 15X in 5 Years 20172012 2012 1.8M645,000 201720172012 22,000
  6. 6. 6 RISE OF NVIDIA GPU COMPUTING Performance Reach 10-50X Roadmap CUDA Kepler Maxwell Pascal Volta Applications 500+ TOOLS & APPS
  7. 7. 7 NVIDIA GPU ACCELERATES 2017 NOBEL PRIZES IN CHEMISTRY AND PHYSICS Cryogenic Electron Microscopy Jacques Dubochet, Joachim Frank, Richard Henderson Detection of Gravitational Waves Rainer Weiss, Barry Barish, Kip Thorne Resolution before 2013 Resolution at present
  8. 8. 8 NEW NVIDIA HOLODECK THE DESIGN LAB OF THE FUTURE Photorealistic Models Physically Simulated Interaction Virtual Team Collaboration GPU-accelerated AI
  9. 9. 9
  10. 10. 10 CREATE AND COLLABORATE IN NVIDIA HOLODECK CATIA / Siemens NX Creo / Alias Maya / 3dsMAX
  11. 11. 11 NEW NVIDIA HOLODECK THE DESIGN LAB OF THE FUTURE Photorealistic Models Physically Simulated Interaction Virtual Team Collaboration GPU-accelerated AI Early Access NOW nvidia.com/holodeck
  12. 12. 12 AI — CUDA GPU’S NEXT KILLER APP NIPS RegistrationAI Startup Funding 10X in 5 Years Deep Learning Papers Published 10X in 3 Years 2017201420172012 $6.6B 3,000 -50 -25 0 25 1500 3000 4500 6000 2017 2016 2002
  13. 13. 13 SOLVING THE UNSOLVABLE NVIDIA Interactive Ray Tracing
  14. 14. 14 SOLVING THE UNSOLVABLE NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation
  15. 15. 15 SOLVING THE UNSOLVABLE NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation WRNCH Pose Estimation
  16. 16. 16 SOLVING THE UNSOLVABLE NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation WRNCH Pose Estimation University of Edinburgh Character Animation
  17. 17. 17 SOLVING THE UNSOLVABLE NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation WRNCH Pose Estimation University of Edinburgh Character Animation UC Berkeley / OpenAI One-shot Imitation Learning
  18. 18. 18 THE WORLD’S AI PLATFORM IT Services Automotive Healthcare Smart City Manufacturing Drones Fin. Services Other NVIDIA Inception: 2,000 DL Startups Every Cloud and Data CenterEvery Framework NVIDIA AI PLATFORM
  19. 19. 19 ANNOUNCING TAIWAN MINISTRY OF SCIENCE AND TECHNOLOGY ADOPTS NVIDIA FOR AI GRAND PLAN Top 25 Fastest Supercomputer in the World Training 3,000 Developers in AI for Manufacturing, IoT, Smart Cities, Healthcare Startup Innovation Center
  20. 20. 20 NVIDIA GPU CLOUD GPU-ACCELERATED CLOUD PLATFORM OPTIMIZED FOR DEEP LEARNING Containerized in NVDocker Optimization Across the Full Stack Always Up-to-Date Fully Tested and Maintained by NVIDIA
  21. 21. 21
  22. 22. 22 NVIDIA GPU CLOUD GPU-ACCELERATED CLOUD PLATFORM OPTIMIZED FOR DEEP LEARNING Containerized in NVDocker Optimization Across the Full Stack Always Up-to-Date Fully Tested and Maintained by NVIDIA Coming this Month Sign up now: www.nvidia.com/gpu-cloud
  23. 23. 23 AI INFERENCE IS THE NEXT GREAT CHALLENGE Training InferenceDNN Model
  24. 24. 24 EXPLOSION OF INTELLIGENT MACHINES 20M Inference Servers Trillions of IoT Devices100s of Millions of Autonomous Machines
  25. 25. 25 EXPLOSION OF NETWORK DESIGN Recurrent Networks Generative Adversarial NetworksConvolution Networks Reinforcement Learning GRU HighwayLSTM Embedding BiDirectionalProjection ReLuPRelu Dropout PoolingConcat BatchNorm A3C Dueling DQNDQNConditional GAN Latent space GAN 3D-GAN Coupled GAN Rank GAN Speech Enhancement GAN
  26. 26. 26 2014 2015 2016 2017 20182013 2014 2015 2016 2017 20182011 2013 2015 2017 EXPLOSION OF NETWORK COMPLEXITY Translation Network Complexity GOPS * Bandwidth Image Network Complexity GOPS * Bandwidth Speech Network Complexity GOPS * Bandwidth 2012 2014 2016 ResNet-50 Inception-v2 Inception-v4 AlexNet GoogLeNet 350X 30X DeepSpeech 3 DeepSpeech 2 DeepSpeech 10X GNMT OpenNMT MoE
  27. 27. 27 NEW NVIDIA TENSORRT 3 DRIVE PX 2 JETSON TX2 NVIDIA DLA TESLA P4 TensorRT TESLA V100 Programmable Inference Accelerator Compile and Optimize Neural Networks | Support for Every Framework Optimize for Each Target Platform
  28. 28. 28 NEW NVIDIA TENSORRT 3 Programmable Inference Accelerator Weight & Activation Precision Calibration | Layer & Tensor Fusion Kernel Auto-Tuning | Multi-Stream Execution concat batch nm batch nm batch nm batch nm max pool input relu relu relu relu 1x1 conv 3x3 conv 5x5 conv 1x1 conv relu batch nm 1x1 conv relu batch nm 1x1 conv next input next input max pool input copy 3x3 CR 5x5 CR 1x1 CR 1x1 CR
  29. 29. 29 NEW NVIDIA TENSORRT 3 Programmable Inference Accelerator 40x Speed-up on ResNet-50 | 140x Speed-up on OpenNMT 4 550 0 150 300 450 600 CPU + Torch V100 + TensorRT 140 5,700 - 1,500 3,000 4,500 6,000 CPU + TensorFlow V100 + TensorRTCPU + TensorFlow V100 + TensorRT Images/Sec (ResNet-50) Sentences/Sec (OpenNMT)
  30. 30. 30 NVIDIA TENSORRT 10X BETTER DATA CENTER TCO 160 CPU Servers 45,000 Images / Second 65 KWatts
  31. 31. 31 NVIDIA TENSORRT 10X BETTER DATA CENTER TCO 1 NVIDIA HGX with 8 Tesla V100 GPUs 45,000 Images / Second 3 KWatts 1/6 the Cost | 1/20 the Power 4 Racks in a Box
  32. 32. 32 INFERENCE ON IMAGES
  33. 33. 33
  34. 34. 34 NVIDIA VOLTA IN EVERY CLOUD, EVERY DATACENTER
  35. 35. 35 NVIDIA TESLA DATACENTER MARKETS HPC CSP TRAINING CSP INFERENCE PUBLIC CLOUD INDUSTRIES ENTERPRISE $25B Market20M Inference Servers 600M Amazon Packages / Yr$12B Market 80% of Apps by 2020 $3T IT Industry
  36. 36. 36 NVIDIA TESLA DATACENTER PLATFORM HPC CSP TRAINING CSP INFERENCE PUBLIC CLOUD INDUSTRIES ENTERPRISE CUDA ComputeWorks NVIDIA AI SDK cuDNN, NCCL, TensorRT, DIGITS Every Framework NGC GRID vPC Quadro vWS
  37. 37. 37 NVIDIA TESLA PLATFORM PARTNERS INVENTEC QUANTA GIGABYTE TYAN WISTRONSUPERMICRO GIGABYTEASUS GIGABYTE QUANTA TYAN FOXCONN 2-GPU 8-GPU4-GPU SUPERMICRO
  38. 38. 38 THE AUTONOMOUS VEHICLE REVOLUTION
  39. 39. 39 NVIDIA DRIVE AV COMPUTING PLATFORM Sensor Fusion: RADAR, LIDAR, Camera | Deep Learning, CV, Parallel Computing Diversity of Algorithms | ASIL-D Functional Safety | Fully Integrated into NVIDIA BB8 DRIVE PX — AI CAR COMPUTER DRIVEWORKS SDK DRIVE AV DRIVE OS RADAR Fusion LIDAR Point-Cloud Processing Camera Deep Learning (Detection) Camera Deep Learning (Freespace) Camera Computer Vision (SLAM) HD Map (Localizing to Map) Path Planning
  40. 40. 40
  41. 41. 41 — Goldman Sachs Uber Lyft Waymo GM / Cruise Ford / Argo.ai Baidu Zoox nuTonomy Yandex “RIDE-HAILING INDUSTRY EXPECTED TO GROW EIGHTFOLD TO $285B BY 2030”
  42. 42. 42 STATE-OF-THE-ART DRIVERLESS VEHICLES
  43. 43. 43 NEW “PEGASUS” ROBOTAXI DRIVE PX 320 TOPS CUDA TensorCore | 16x GMSL | 4x 10G | 8x 1G | 16x 100M | Auto-grade | ASIL D 500W | Late Q1 Early Access Partners Supercomputing Data Center in Your Trunk Size of a License Plate
  44. 44. 44 NEW “PEGASUS” ROBOTAXI DRIVE PX 320 TOPS CUDA TensorCore | 16x GMSL | 4x 10G | 8x 1G | 16x 100M | Auto-grade | ASIL D 500W | Late Q1 Early Access Partners Supercomputing Data Center in Your Trunk Size of a License Plate
  45. 45. 45 STATE-OF-THE-ART DRIVERLESS VEHICLES
  46. 46. 46 NEW “PEGASUS” ROBOTAXI DRIVE PX 320 TOPS CUDA TensorCore | 16x GMSL | 4x 10G | 8x 1G | 16x 100M | Auto-grade | ASIL D 500W | Late Q1 Early Access Partners Supercomputing Data Center in Your Trunk Size of a License Plate
  47. 47. 47 THE ERA OF AUTONOMOUS MACHINES
  48. 48. 48 A WORLD OF AUTONOMOUS MACHINES 10% of Manufacturing Tasks Are Automated 1M Pizzas Delivered Per Day by Domino’s 100M People 80+ Years Old Ag Tech: 70% Increase in Farm Yields by 2050 600K Bridges to Inspect in the U.S. 300M Operations per Year WW
  49. 49. 49 NVIDIA JETSON AUTONOMOUS MACHINE PLATFORM Jetson TX2 JetPack SDK DIGITS Isaac Robot Simulator Deep Learning Institute
  50. 50. 50 ADAPTING TO NEW USE CASES Pre-Trained Network Optimized Network Optimized NetworkNVIDIA DGX Station NVIDIA Jetson NVIDIA DIGITS Pre-Trained Network 20’ Camera Indoor Camera 360° Camera Improved Improved TensorRT Improved
  51. 51. 51
  52. 52. 52 A NEW COMPUTING ERA Taiwan’s MOST Adopts NVIDIA for AI Grand Plan NVIDIA AUTONOMOUS MACHINES Jetson TX2 JetPack | DIGITS | Isaac Robot Simulator | Deep Learning Institute NVIDIA AI Volta in Every Cloud | NVIDIA GPU Cloud Registry TensorRT Programmable Inference Accelerator
  53. 53. 53

×