460 likes | 602 Views
Kicking off the first in a series of global GPU Technology Conferences, NVIDIA co-founder and CEO Jen-Hsun Huang today at GTC China unveiled technology that will accelerate the deep learning revolution that is sweeping across industries. Huang spoke in front of a crowd of more than 2,500 scientists, engineers, entrepreneurs and press, gathered in Beijing for a day devoted to deep learning and AI. On stage he announced the Tesla P4 and P40 GPU accelerators for inferencing production workloads for AI services and, a small, energy-efficient AI supercomputer for highway driving — the NVIDIA DRIVE PX 2 for AutoCruise.
E N D
THE DEEP LEARNING AI REVOLUTION GTC 2016 — China
GPU DEEP LEARNING BIG BANG ImageNet Classification with Deep Convolutional Neural Networks Ilya Sutskever University of Toronto Geoffrey e. Hinton University of Toronto Alex Krizhevsky University of Toronto NIPS (2012) Deep Learning NVIDIA GPU 2
GPU DEEP LEARNING ACHIEVES “SUPERHUMAN” RESULTS ImageNet — Accuracy % 96% Human Microsoft, Google 3.5% error rate DL 74% Hand-coded CV 2010 2011 2012 2013 2014 2015 2012: Deep Learning researchers worldwide discover GPUs 2015: DNN achieves superhuman image recognition 2015: Deep Speech 2 achieves superhuman voice recognition 3
NVIDIA — “THE AI COMPUTING COMPANY” GPU Computing Computer Graphics Artificial Intelligence 4
ANNOUNCING NEW GRAPHICS SDKS Ansel Volumetric Physical Light Models OptiX 4.0 Mental Ray MDL 1.0 In-game Photography Multi-GPU Ray-Tracing Now GPU-Accelerated! Physically Based Materials Funhouse VR Open Source 360 Video 1.0 Real-Time Panoramic VR Iray VR Photorealistic VR Ray Tracing Remote Rendering Video Compositing GVDB Sparse Volumes for Special Effects 5
GTC — 25X GROWTH IN GPU DL DEVELOPERS • Government 5% • Medical • Finance • Manufacturing 4% • Japan • Korea • United States (Silicon Valley, D.C.) • Australia • China • Europe • India • Higher Ed 35% • Software 19% • Internet 15% • Auto 4% 4% 10% 16,000 400,000 55,000 • Japan • United States 120,000 3,700 2,200 2014 2016 2014 2016 2014 2016 4X Attendees 3X GPU Developers 25x Deep Learning Developers 9
WHY DID AI RESEARCHERS ADOPT GPUs FOR DEEP LEARNING? 10
BRAIN IS LIKE A GPU BRAIN CREATES MENTAL IMAGES WHEN WE THINK 11
GPU DEEP LEARNING IS A NEW COMPUTING MODEL Training Datacenter Device 13
GPU DEEP LEARNING IS A NEW COMPUTING MODEL TRAINING Training Datacenter Billions of Trillions of Operations GPU train larger models, accelerate time to market Device 14
GPU DEEP LEARNING IS A NEW COMPUTING MODEL DATACENTER INFERENCING Training Datacenter 10s of billions of image, voice, video queries per day GPU inference for fast response, maximize datacenter throughput Device 15
GPU DEEP LEARNING IS A NEW COMPUTING MODEL Training Datacenter DEVICE INFERENCING Billions of intelligent devices GPU for real-time accurate response Device 16
AI — THE ULTIMATE COMPUTING CHALLENGE IMAGE RECOGNITION SPEECH RECOGNITION Important Property of Neural Networks 16X Model 10X Training Ops 20 ExaFLOPS Results get better with 152 layers 22.6 GFLOP/image ~3.5% error more data + bigger models + more computation 100M | 12,000 Hours ~5% Error 2 ExaFLOPS 8 layers 1.4 GFLOP/image ~16% Error 25M | 7,000 Hours ~8% Error (Better algorithms, new insights and improved techniques always help, too!) 2012 AlexNet 2015 ResNet 2014 2015 Deep Speech 1 Deep Speech 2 17
PASCAL “5 MIRACLES” BOOST DEEP LEARNING 65X Pascal 70X Pascal 60X 50X 16nm FinFET 40X 30X 20X Maxwell CoWoS HBM2 PaddlePaddle Baidu Deep Learning 10X Kepler X NVLink 2013 2014 2015 2016 cuDNN Pascal — 5 Miracles NVIDIA DGX-1 Supercomputer 65X in 4 yrs Accelerate Every Framework 18 Chart: Relative speed-up of images/sec vs K40 in 2013. AlexNet training throughput based on 20 iterations. CPU: 1x E5-2680v3 12 Core 2.5GHz. 128GB System Memory, Ubuntu 14.04. M40 datapoint: 8x M40 GPUs in a node P100: 8x P100 NVLink-enabled.
ANNOUNCING NEW IBM SERVER POWER8 + NVIDIA TESLA P100 FOR THE AI ENTERPRISE “ Putting NVIDIA’s technology into the IBM system will speed up performance for such emerging workloads as AI, deep learning and data analytics.” — eWeek 19
Training Datacenter Device 21
ANNOUNCING TESLA P4 & P40 INFERENCING ACCELERATORS Pascal Architecture | INT8 P40: 250W | 40X Energy Efficient versus CPU P40: 250W | 40X Performance versus CPU 22
ANNOUNCING TensorRT PERFORMANCE OPTIMIZING INFERENCING ENGINE FP32, FP16, INT8 | Vertical & Horizontal Fusion | Auto-Tuning VGG, GoogLeNet, ResNet, AlexNet & Custom Layers Available Today: developer.nvidia.com/tensorrt 23
NVIDIA GPU DEEP LEARNING EVERYWHERE Alibaba/Aliyun Amazon eBay Flickr Google iFLYTEK Facebook Baidu iQIYI Pinterest Qihoo 360 Netflix JD.com Microsoft Periscope Orange Yandex Tencent Yelp Sogou Skype Yahoo Supermarket Shazam Twitter 26
>1,500 AI STARTUPS AROUND THE WORLD Deep Learning for Cybersecurity Deep Learning for Genomics Deep Learning for Self-Driving Cars Deep Learning for Art 27
AI STARTUPS IN CHINA Weather & Environment Forecast Eye-tracking for Human- machine Interaction Medical Imaging Face Product Recognition, Detection, Search Personal Concierge App Recognition 28
Training Datacenter Device 29
“BILLIONS OF INTELLIGENT DEVICES” “Billions of intelligent devices will take advantage of DNNs to provide personalization and localization as GPUs become faster and faster over the next several years.” — Tractica 30
AI CITY — 1B CAMERAS BY 2020 ~1 billion cameras worldwide by 2020 30 billion inferences/sec Tesla P40: 2,500 inferences/sec @ 720P AI City needs ~10M P40 servers 31 DATA: 1B cameras, IHS “Video Surveillance Intelligence Service, Aug. 2016”
1/20THTHE SPACE, 1/10THTHE POWER ~21 1U Servers 42 CPUs ~4,000 W 1 Hikvision Blade 16 TX1 + 1 CPU >8 1080 streams ~300 W NVIDIA DGX-1 Hikvision Blade 16 Jetson TX1s Traditional Server Hikvision Blade 32
AI TRANSPORTATION — $10T INDUSTRY PERCEPTION AI PERCEPTION AI LOCALIZATION DRIVING AI DEEP LEARNING 34
FREE SPACE DETECTION CAR 3D DETECTION 35
NVIDIA DRIVE PX 2 AutoCruise to Full Autonomy — One Architecture Full Autonomy AutoChauffeur AUTONOMOUS DRIVING Perception, Reasoning, Driving AI Supercomputing, AI Algorithms, Software Scalable Architecture AutoCruise 37
NVIDIA DRIVE PX 2 AUTOCRUISE 10W AI Car Computer | Passive Cooling | Automotive IO AI Highway Driving | Localization & Mapping 38
NVIDIA & BAIDU PARTNER ON AI SELF-DRIVING CARS 39
NVIDIA AI SELF-DRIVING CARS IN DEVELOPMENT Baidu nuTonomy Volvo WEpods NVIDIA 40
NVIDIA END-TO-END DEEP LEARNING PLATFORM PaddlePaddle Baidu Deep Learning TESLA P100 DGX-1 TRAINING 41
NVIDIA END-TO-END DEEP LEARNING PLATFORM PaddlePaddle Baidu Deep Learning ANNOUNCING TensorRT ANNOUNCING TESLA P4 & P40 TESLA P100 DGX-1 TRAINING DATACENTER INFERENCING 42
NVIDIA END-TO-END DEEP LEARNING PLATFORM CUDA PaddlePaddle Baidu Deep Learning ANNOUNCING TensorRT JETPACK DRIVEWORKS ANNOUNCING DRIVE PX 2 AUTOCRUISE ANNOUNCING TESLA P4 & P40 TESLA P100 DGX-1 JETSON TX1 TRAINING DATACENTER INFERENCING INTELLIGENT DEVICES 43
NVIDIA DEEP LEARNING PLATFORM PARTNERS AI ENTERPRISE AI CITY AI CAR 44
AI FOR EVERYONE AI will Revolutionize Transportation AI will Revolutionize Healthcare AI will Revolutionize Society 45