AI IS NOT ONLY DRIVING CARS - brainandheart.de · NVIDIA DGX-1 AI supercomputer to analyze...

An overview of use-cases and introduction to DGX

AI IS NOT ONLY DRIVING CARS

Ralph Hinsche( rhinsche@nvidia.com )

2

“Find where I parked my car”

AI IS EVERYWHERE

“Find the bag I just saw in this magazine”

“What movie should I watch next?”

3

Bringing grandmother closer to family by bridging language barrier

TOUCHING OUR LIVES

Predicting sick baby’s vitals like heart rate, blood pressure, survival rate

Enabling the blind to “see” their surrounding, read emotions on faces

4

Increasing public safety with smart video surveillance at airports & malls

AI FOR PUBLIC GOOD

Providing intelligent services in hotels, banks and stores

Separating weeds as it harvests, reduces chemical usage by 90%

5

“ Mobile computing, inexpensive sensors collecting terabytes of data, rise of machine learning that can use that data will fundamentally change the way the global economy is organized.”

Fortune, CEOs: The Revolution is Coming March 8, 2016

6

PERSONALIZATION

2 TrillionMessages Per Day On

LinkedIn

AI INFERENCING IS EXPLODING

SPEECH TRANSLATION VIDEO

60 BillionVideo frames/day uploaded on

Youtube

140 BillionWords Per Day Translated by

Google

500MDaily active users of

iFlyTek

7

AI INFERENCE IS THE NEXT GREAT CHALLENGE

Inferencing

DNN Model

Training

8

2016 – Baidu Deep Speech 2Superhuman Voice Recognition

2015 – Microsoft ResNetSuperhuman Image Recognition

2017 – Google Neural Machine TranslationNear Human Language Translation

100 ExaFLOPS8700 Million Parameters

20 ExaFLOPS300 Million Parameters

7 ExaFLOPS60 Million Parameters

To Tackle Increasingly Complex ChallengesNEURAL NETWORK COMPLEXITY IS EXPLODING

9

GPU-ACCELERATED INFERENCE POWERS CHINA CSPS

SPEECH RECOGNITION70% of China Market

SPEECH TO CONTEXT~1B Users

INTELLIGENT VIDEO ANALYSIS1K Channels

LANGUAGE TRANSLATION~8B Queries/Day

2.5X Throughput

+20% Accuracy

10XConcurrent Requests Per Server

20X Per-Server Efficiency

3.5X Latency Reduction

3XRequests Serviced

10

JD.COM SUBSIDIARY JDX SELECTS NVIDIA FOR AUTONOMOUS MACHINESFrom Warehouse to Door Delivery

jROVER jDRONE1 Million Drones by 2022

11

INTERNET SERVICES

DEEP LEARNING IS SWEEPING ACROSS INDUSTRIES

MEDICINE MEDIA & ENTERTAINMENT SECURITY & DEFENSE AUTONOMOUS MACHINES

Cancer cell detectionDiabetic gradingDrug discovery

Pedestrian detectionLane trackingRecognize traffic signs

Face recognitionVideo surveillanceCyber security

Video captioningContent based searchReal time translation

Image/Video classificationSpeech recognitionNatural language processing

12

THE NEW SCIENCE OF SPORTSPredictive analytics, commonly used in business to identify risks and opportunities, is increasingly used by the sports industry to tap into its massive troves of data. Scientists at NYU are applying deep learning and the NVIDIA DGX-1 AI supercomputer to analyze unprecedented amounts of Major League Baseball’s data — four years-worth of every player’s every move — to ask bigger and better questions to help improve the game.

13

TEACHING A ROBOT TO STAND UP FOR ITSELFNew approaches to AI promise to help scientists build machines with greater autonomy. Researchers at UC Berkeley are tapping into the processing power and integrated software of NVIDIA’s DGX-1 to advance robotics using reinforcement learning. DGX-1 will allow them to iterate faster and ultimately build robots that are able to understand and navigate a diverse and changing world on their own.

14

ACCELERATING DISCOVERIESWITH AINew drugs typically take 12-14 years and $2.6 billion to bring to market. BenevolentAI is using GPU deep learning for NLP to bring new therapies to market quickly and more affordably. They’ve automated the process of identifying patterns within large amounts of research literature, enabling scientists to form hypotheses and draw conclusions quicker than any human researcher could. And using the NVIDIA DGX-1 AI supercomputer, they identified two potential drug targets for Alzheimer’s in less than one month.

15

Our daily life, economic vitality, and national security depend on a stable, safe and resilient cyberspace. But attacks on IT systems are becoming more complex and relentless, resulting in loss of information and money and disruptions to essential services. Accenture’s dedicated cyber security lab uses NVIDIA GPUs, CUDAlibraries, and machine learning to accelerate the analysis and visualization of 200M-300M alerts daily so analysts can take timely action.

AI-ACCELERATED CYBER DEFENSE

16

AI IMPROVESTHE CUSTOMER EXPERIENCEAI is dramatically changing the online shopping experience with tangible improvements to retailers and consumers. In 2016 online British grocery giant Ocado improved customer service with their AI-enhanced contact center, and is applying machine learning and NVIDIA GPUs to develop humanoid robotics to assist maintenance technicians, and advanced computer vision for image classification and recognition to replace barcode systems. Computer vision will expedite the picking process and better ensure orders are filled correctly so customers receive exactly what they ordered.

17

THE MODERN WAREHOUSE BUILT ON AIWorldwide retail e-commerce sales are expected to reach $2 trillion in 2016, according to eMarketer. With thousands of orders placed every hour, data scientists at Zalando, Europe’s leading online fashion retailer, applied deep learning and GPUs to develop the Optimal Cart Pick algorithm. Applying the algorithm resulted in an 11% decrease in workers’ travel time per item picked. The work is a good example of the efficiencies that AI can discover for e-commerce, manufacturing and other large-systems-based industries.

18

AI PREDICTS AND PREVENTS DISEASEGPU deep learning is giving doctors a life-saving edge by identifying high-risk patients before diseases are diagnosed. Icahn School of Medicine at Mount Sinai built an AI-powered tool, “Deep Patient,” based on NVIDIA GPUs and the CUDA programming model. Deep Patient can analyze a patient’s medical history to predict nearly 80 diseases up to 1 year prior to onset.

19

Weather forecasting involves processing vast amounts of data to derive predictions that can save lives and protect property. Colorful Clouds is using GPU deep learning to speed the processing of data by 30-50x. It’s location-based reporting tool can forecast and communicate weather and air-quality conditions with high-accuracy in real-time.

AI-POWERED WEATHER FORECASTING

20

AN AI MONITOR OF EARTH’S VITALSThe Earth’s climate has changed throughout history, but in recent years there have been record increases in temperature, glacial retreat and rising sea levels. NASA Ames is using satellite imagery to measure the effects of carbon and greenhouse gas emissions on the planet. To do so, they developed DeepSat — a deep learning framework for satellite image classification trained on a GPU-powered supercomputer. The enhanced satellite imagery will help scientists plan to protect ecosystems and farmers improve crop production.

21

DEFENDING THE PLANET WITH AIThe U.S. government’s Asteroid Grand Challenge seeks to identify asteroid threats to human populations. The team at NASA Frontier Development Labs picked up the challenge by employing GPU deep learning to identify threats and their unique characteristics. The resulting “Deflector Selector” achieved a 98% success rate in determining which technology produced the most successful deflection.

22

AI PLATFORM TO ACCELERATE CANCER RESEARCHTo speed advances in the fight against cancer, the Cancer Moonshot initiative unites the Department of Energy, the National Cancer Institute and other agencies with researchers at Oak Ridge, Lawrence Livermore, Argonne, and Los Alamos National Laboratories. NVIDIA is collaborating with the labs to help accelerate their AI framework called CANDLE as a common discovery platform, with the goal of achieving 10X annual increases in productivity for cancer researchers.

23

Fastest AI Supercomputer in TOP5004.9 Petaflops Peak FP6419.6 Petaflops Peak FP1613x DGX-1 to get into Top500

Most Energy Efficient Supercomputer#1 Green5009.46 GFLOPS per Watt

Rocket for Cancer MoonshotCANDLE Development Platform Common platform with DOE labs – ANL, LLNL,

ORNL, LANL

NVIDIA DGX SATURNVGiant Leap Towards Exascale AI

24

TESLA V100THE MOST ADVANCED DATA CENTER GPU EVER BUILT

5,120 CUDA cores640 NEW Tensor cores7.5 FP64 TFLOPS | 15 FP32 TFLOPS120 Tensor TFLOPS20MB SM RF | 16MB Cache | 16GB HBM2 @ 900 GB/s300 GB/s NVLink

25

NEW TENSOR CORE BUILT FOR AIDelivering 120 TFLOPS of DL Performance

TENSOR CORE

ALL MAJOR FRAMEWORKSVOLTA-OPTIMIZED cuDNN

MATRIX DATA OPTIMIZATION:

Dense Matrix of Tensor Compute

TENSOR-OP CONVERSION:

FP32 to Tensor Op Data for

Frameworks

TENSOR CORE

VOLTA TENSOR CORE 4x4 matrix processing array

D[FP32] = A[FP16] * B[FP16] + C[FP32]Optimized For Deep Learning

26

REVOLUTIONARY AI PERFORMANCE3X Faster DL Training Performance

Over 80x DL Training Performance in 3 Years

1x K80cuDNN2

4x M40cuDNN3

8x P100cuDNN6

8x V100cuDNN7

0x

20x

40x

60x

80x

100x

Q115

Q315

Q217

Q216

Googlenet Training Performance(Speedup Vs K80)

Spee

dup

vs K

80

85% Scale-Out EfficiencyScales to 64 GPUs with Microsoft

Cognitive Toolkit

0 5 10 15

64X V100

8X V100

8X P100

Multi-Node Training with NCCL2.0(ResNet-50)

ResNet50 Training for 90 Epochs with 1.28M images dataset | Cognitive Toolkit with NCCL 2.0 | V100 performance measured on pre-production

hardware.

1 Hour

7.4 Hours

18 Hours

3X Reduction in Time to Train Over P100

0 10 20

1X V100

1X P100

2X CPU

LSTM Training(Neural Machine Translation)

Neural Machine Translation Training for 13 Epochs |German ->English, WMT15 subset | CPU = 2x Xeon E5 2699 V4 | V100 performance

measured on pre-production hardware.

15 Days

18 Hours

6 Hours

27

TUNE IN TO THE LATEST AI NEWS

NVIDIA AI Twitter

AI on Nvidia.com

AI Newsletter sign up

AI Podcast

Deep Learning blog

28

29

30

Saveadditional25%withthepromocode: RalphHinscheGTCEU17Specialstudentpromocode: STU1GTCEU17(1day,€75)

STUFGTCEU17(3day,€165)

32

BACK UP

33

A NEW COMPUTING MODEL

TRADITIONAL APPROACHRequires domain expertsTime consuming

Error proneNot scalable to new problems

Algorithms that learn from examples

DEEP LEARNING APPROACHLearn from dataEasily to extend

Speedup with GPUs

Expert Written Computer Program

CarVehicle

Coupe

CarVehicle

Coupe

Deep Neural Network

34

DEEP LEARNINGHow it works

Option 1- Slide is on click animation

35

DEEP LEARNINGHow it works

Option 2- No animation