A NEW COMPUTING ERA JENSEN HUANG, FOUNDER & CEO GTC CHINA 2017

Transcription

1 A NEW COMPUTING ERA JENSEN HUANG, FOUNDER & CEO GTC CHINA 2017

2 TWO FORCES DRIVING THE FUTURE OF COMPUTING 10 7 Transistors (thousands) X per year X per year Single-threaded perf Years of Microprocessor Trend Data 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 by K. Rupp

3 TWO FORCES DRIVING THE FUTURE OF COMPUTING 10 7 Transistors (thousands) X per year X per year Single-threaded perf Years of Microprocessor Trend Data The Big Bang of Deep Learning 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 by K. Rupp

4 RISE OF NVIDIA GPU COMPUTING GPU-Computing perf 1.5X per year 1.1X per year 1000X by X per year Single-threaded perf Years of Microprocessor Trend Data The Big Bang of Deep Learning 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 by K. Rupp

5 RISE OF NVIDIA GPU COMPUTING 22, , M Global GTC Attendees 10X in 5 Years GPU Developers 14X in 5 Years 2017 CUDA Downloads 5X in 5 Years, +800K in 1 Year

6 NVIDIA HOLODECK Photorealistic Models Physically Simulated Interaction Virtual Team Collaboration GPU-accelerated AI

7

8 AI ON THE RISE $6.6B 3, AI Startup Funding 12X in 5 Years Deep Learning Papers Published 13X in 3 Years NIPS Registration

9 NVIDIA Interactive Ray Tracing AMAZING ADVANCES IN AI

10 AMAZING ADVANCES IN AI NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation

11 AMAZING ADVANCES IN AI NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation WRNCH Pose Estimation

12 AMAZING ADVANCES IN AI NVIDIA Interactive Ray Tracing NVIDIA / Remedy Audio-driven Facial Animation WRNCH Pose Estimation University of Edinburgh Character Animation

13 AMAZING ADVANCES IN AI 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

14 NVIDIA IS THE WORLD S LEADING AI PLATFORM ONE ARCHITECTURE CUDA

15 ANNOUNCING CHINA TOP CSPs ADOPT NVIDIA AI COMPUTING 2B USERS

16 ANNOUNCING CHINA ENTERPRISE IT LEADERS ADOPT NVIDIA AI COMPUTING $200B CHINA IT MARKET

17 NVIDIA AI FOR DEVELOPERS EVERYWHERE Every Framework NVIDIA Inception: 1,900 DL Startups Every Cloud and Data Center Other IT Services Financial Services Drones Manufacturing Smart City Automotive Healthcare NVIDIA AI PLATFORM

18 AI INFERENCE IS THE NEXT GREAT CHALLENGE Training DNN Model Inferencing

19 EXPLOSION OF NETWORK DESIGN Convolution Networks Recurrent Networks Generative Adversarial Networks Reinforcement Learning PRelu ReLu BatchNorm LSTM GRU Highway 3D-GAN Rank GAN Conditional GAN DQN Dueling DQN Concat Dropout Pooling Projection Embedding BiDirectional Coupled GAN Speech Enhancement GAN Latent space GAN A3C

20 EXPLOSION OF NETWORK DESIGN Convolution Networks Recurrent Networks Generative Adversarial Networks Reinforcement Learning PRelu ReLu BatchNorm LSTM GRU Highway 3D-GAN Rank GAN Conditional GAN DQN Dueling DQN Concat Dropout Pooling Projection Embedding BiDirectional Coupled GAN Speech Enhancement GAN Latent space GAN A3C

21 EXPLOSION OF NETWORK COMPLEXITY 350X Inception-v4 30X DeepSpeech 3 10X MoE GNMT AlexNet GoogLeNet ResNet-50 Inception-v2 DeepSpeech DeepSpeech 2 OpenNMT Image Network Complexity GOPS * Bandwidth Speech Network Complexity GOPS * Bandwidth Translation Network Complexity GOPS * Bandwidth

22 EXPLOSION OF INTELLIGENT MACHINES 20M Inference Servers 100s of Millions of Autonomous Machines Trillions of IoT Devices

23 ANNOUNCING TESLA P4 NVIDIA TENSORRT 3 PROGRAMMABLE INFERENCE ACCELERATOR TensorRT JETSON TX2 Compile and Optimize Neural Networks Support for Every Framework Optimize for Each Target Platform DRIVE PX 2 NVIDIA DLA TESLA V100

24 ANNOUNCING NVIDIA TENSORRT 3 PROGRAMMABLE INFERENCE ACCELERATOR next input concat relu relu relu relu batch nm batch nm batch nm batch nm 1x1 conv 3x3 conv 5x5 conv 1x1 conv next input copy 3x3 CR 5x5 CR 1x1 CR Weight & Activation Precision Calibration Layer & Tensor Fusion Kernel Auto-Tuning relu batch nm 1x1 conv relu batch nm 1x1 conv max pool 1x1 CR input max pool Multi-Stream Execution input

25 6,000 5, ANNOUNCING 4, NVIDIA TENSORRT 3 PROGRAMMABLE INFERENCE ACCELERATOR 3, x Speed-up on ResNet x Speed-up on OpenNMT 1, CPU + TensorFlow 300 V100 + TensorFlow V100 + TensorRT CPU + Torch 25 V100 + Torch V100 + TensorRT Images/Sec (ResNet-50) Sentences/Sec (OpenNMT)

26 6, ANNOUNCING 4, ,000 14ms ms NVIDIA TENSORRT 3 PROGRAMMABLE INFERENCE ACCELERATOR 40x Speed-up on ResNet x Speed-up on OpenNMT 1,500 7ms 7ms ms 117ms 0- CPU + TensorFlow V100 + TensorFlow V100 + TensorRT 0 CPU + Torch V100 + Torch V100 + TensorRT Images/Sec (ResNet-50) Sentences/Sec (OpenNMT)

27 AI INFERENCE DRIVES DATACENTER TCO 160 CPU servers 45,000 images / second 65 KWatts

28 AI INFERENCE DRIVES DATACENTER TCO 1 NVIDIA HGX with 8 Tesla V100 GPUs 45,000 images / second 3 KWatts 4 Racks in a Box

29 INFERENCE ON IMAGES

30

31 INFERENCE ON SPEECH

32

33 ANNOUNCING CHINA CSPs ADOPT NVIDIA GPU-ACCELERATED INFERENCE PLATFORM

34 AI CITY SMARTER, SAFER CITIES 1B cameras WW by 2020 Finding lost people Improving traffic Enhancing law enforcement

35 NVIDIA TensorRT ANNOUNCING Hikvision Server with NVIDIA Tesla HIKVISION ADOPTS NVIDIA FOR AI CITY Training Inferencing Hikvision NVR with NVIDIA Jetson Hikvision Camera with NVIDIA Jetson

36 NVIDIA AI CITY SMARTER, SAFER CITIES IN CHINA 10X Larger Dataset for Face Recognition 100s of Virtual Security Guards 15% Reduction in Peak Traffic +90% Accuracy for Congestion Alerts

37 ADAPTING TO NEW USE CASES 20 Camera NVIDIA DIGITS NVIDIA TensorRT Improved Indoor Camera Improved Pre-Trained Network Optimized Network NVIDIA Jetson 360 Camera Pre-Trained Network NVIDIA DGX Station Improved Optimized Network

38 THE AUTONOMOUS VEHICLE REVOLUTION

39 NVIDIA DRIVE AV COMPUTING PLATFORM Perception Localization Planning DRIVE AV Level 3 to Level 5 ASIL-D Functional Safety DRIVEWORKS SDK DRIVE OS DRIVE PX

40

41 145 AV STARTUPS ON NVIDIA DRIVE

42 145 AV STARTUPS ON NVIDIA DRIVE

43 145 AV STARTUPS ON NVIDIA DRIVE

44 145 AV STARTUPS ON NVIDIA DRIVE

45 145 AV STARTUPS ON NVIDIA DRIVE

46 145 AV STARTUPS ON NVIDIA DRIVE

47 145 AV STARTUPS ON NVIDIA DRIVE

48 THE ERA OF AUTONOMOUS MACHINES

49 XAVIER WORLD S FIRST AUTONOMOUS MACHINE PROCESSOR Supercomputing Performance Extreme Energy Efficiency Rich High-Speed Sensor IOs 30 TOPS of CV, Deep Learning, and Parallel Computing

50 ANNOUNCING JD X SELECTS NVIDIA FOR AUTONOMOUS MACHINES

51 CREATING AUTONOMOUS MACHINES Project Isaac SIMULATION TRAINING AUTONOMOUS

52

53

54 A NEW COMPUTING ERA NVIDIA TensorRT NVIDIA The World s AI Computing Platform NVIDIA TensorRT 1st Programmable Inference Accelerator NVIDIA TensorRT AI Cities Platform NVIDIA DRIVE Open AV Platform for the Transportation Industry Xavier 1 st Autonomous Machine Processor

55