Slide 1: NVIDIA
@A NVIDIA.
Slide 2: NVIDIA Accelerated Computing Growth
NVIDIA ACCELERATED COMPUTING GROWTH 50% GROWTH OF NVIDIA DEVELOPERS 50% GROWTH IN TOP500 600+ CUDA APPS MORE PERF #1 World, US— ORNL Summit #1 Europe — CSCS Piz Daint #1 Japan — AIST ABC 22 of Top 25 Energy-Efficient 40X 2018 2019 ore Chemistry an ARB Microscopy Genomics Westner
Slide 3: NVIDIA CUDA-X
NVIDIA CUDA-X GPU-ACCELERATED COMPUTING LIBRARIES
Slide 4: NVIDIA CUDA-X Acceleration
Programmable SD, TTS, GID NVIDIA CUDA-X Acceleration TTA, SOP, COI GPU-ACCELERATED COMPUTING Domains COD, Col LIBRARIES Architecture GIA, GID, COl
Slide 5: Programmable Acceleration
PRogrammable Acceleration Domains Architecture NVIDIA CUDA-X GPU-ACCELERATED COMPUTING LIBRARIES
Slide 6: Chapter 1: Graphics
CHAPTER 1: GRAPHICS
Slide 7: Slide 7
Slide 8: Slide 8
Slide 9: Slide 9
Slide 10: Slide 10
Slide 11: Slide 11
Slide 12: NVIDIA Turing RTX
NVIDIA TURING RTX Micrgsott DirectX 18 BILLION TRANSISTORS 32 TOPS SHADER QVatkan. 130 TFLOPS TENSOR CORE
Slide 13: Slide 13
Slide 14: Slide 14
Slide 15: RTX Comes to 9M 3D Creators in 2019
RTX COMES TO 9M 3D CREATORS IN 2019
Slide 16: Slide 16
Slide 17: Pixar ManXPU R&D
R vt ManXPU R&D: #ofNodes TotalRender Time Cost of Power (Syrs) Total Cost CPU - Dual Skylake 25 38 Hours $70k $250k RTX Server - 4x RTX 8000 1 6 Hours $10k $30k Pixar artists already rely on NVIDIA ray tracing, and RTX more than doubles the performance they will see. We're excited to use RTX on our
Slide 18: Complex Rendering Pipeline
{ » ( « 6 4 “t t COMPLEX RENDERING PIPELINE CONCEPT | MODELING | TEXTURE | RIGGING ANIMATION | LIGHTING | RENDER
Slide 19: Global Studio Collaboration
_- OINESES technicolor GLOBAL STUDIO COLLABORATION eee Si avmattosie PIXAR ~ > Bisimat, BSwAln
Slide 20: Announcing NVIDIA Omniverse
_- OINESES ANNOUNCING NVIDIA OMNIVERSE — Si avmattosie OPEN COLLABORATION PLATFORM TO SIMPLIFY STUDIO a WORKFLOWS FOR REAL-TIME GRAPHICS
Slide 21: Sas
We 2 Sas
Slide 22: Early Access at developer.nvidia.com/nvidia-omniverse
Early Access at developer.nvidia.com/nvidia-omniverse Ww 7p) a wi 2 Zz = ro) < a = 2 co) Zz rs) Zz a) ° Zz Zz =
Slide 23: GeForce NOW
lk pi GeForce NOW 500+ Games =e 15 Datacenters 300K Players 1M on Waiting List i*
Slide 24: GeForce NOW
> GeForce NOW 00+ Games 15 Datacenters 300K Players 1M on Waiting List = SoftBank @.LGu* Announcing GEN Alliance
Slide 25: Announcing RTX Server
ANNOUNCING RTX SERVER Datacenter Graphics Server Design 40 Turing GPUs in 8U Virtualize graphics apps up to 320 CCU Optimized end-to-end stack for rendering, remote workstation, and cloud gaming
Slide 26: Announcing RTX Server Pod
ANNOUNCING RTX SERVER POD Modular Designs for Enterprise & Cloud Edge Datacenters Pods scale to 32 RTX servers 1,280 GPUs in 10 racks High-speed storage connected with MLNX IB + Up to 10,000 concurrent users per RTX Pod ve
Slide 27: NVIDIA RTX Server
NVIDIA RTX SERVER Ga” i. be RENDERING OMNIVERSE GEFORCE NOW C — y ie
Slide 28: Slide 28
Slide 29: Chapter 2: AI & HPC
CHAPTER 2: Al & HPC
Slide 30: Data Science - A New Pillar of Discovery
DATA SCIENCE - A NEW PILLAR OF DISCOVERY AL PREDICTIVE FEATURES MODEL DATA -~ a ae ——— __ PREDICTION DATA >
Slide 31: CSV, Parq, HDFS
CSV, PARQ, HDFS ETL, Pandas, TensorFlow, PyTorch TensorFlow Serving Spark, Graph MXNet, Scikit-Learn, XGBoost ONNX, SageMaker NEO
Slide 32: cuIO cuDF cuGraph
cuIO cuDNN TensorRT cuDF cuML TRTIS cuGraph
Slide 33: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 34: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 35: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 36: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 37: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 38: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 39: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 40: Serving Amazon SageMaker Neo
Amazon SageMaker Neo Serving
Slide 41: Audio Visual
Entity Recognition Query Annotation Auto Correct AUDIO Query Search Denoising Speech Recognition Language Model Recommendation VISUAL What are different types of lighting for a living room? JPEG Decode Resize Object Detection Social News Ambient, task and accent Web AUDIO Visual Search Voice Encoder
Slide 42: I'm Listening
I'mlistening HY Microsoft https://bing.app.link/gtc
Slide 43: Announcing World's Leading Tech Companies
ANNOUNCING WORLD'S LEADING TECH COMPANIES ADOPT CUDA-X Al TO ACCELERATE MODEL DEPLOYMENT Voice Search Image Search Recommendations Home Assistant Translation eCommerce 6X TensorRT Downloads a wD Expedia @ vy oo 2017 2018
Slide 44: Announcing Clara AI Toolkit
ANNOUNCING CLARA Al TOOLKIT BUILD, MANAGE AND DEPLOY Al APPLICATIONS FOR RADIOLOGY Al-Assisted Annotation - Hours to Minutes { 10x Less Training Data Needed ‘e)\ a 19 Pre-Trained Models PRE-TRAINED ALASSISTED TRANSFER AL Reference Training and Deployment Pipelines MODELS: ‘ANNOTATION LEARNING DEPLOY
Slide 45: Clara AI - Built for Radiology
CLARA Al - BUILT FOR RADIOLOGY MGH NIH OSU DKFZ 3D Annotation from Hours to Minutes Trained State-of-the-Art Model in Weeks Clinical Model Deployment in Less Than 24 hours Clara Al Annotation Integrated in MITK Open Source Viewer
Slide 46: Data Science Is the New HPC Challenge
DATA SCIENCE IS THE NEW HPC CHALLENGE AL PREDICTIVE FEATURES MODEL DATA -~ a ae ——— __ PREDICTION DATA >
Slide 47: 500K Access Points
500K access points 1TB | day Spectrum> 25 million customers in 40+ states in US DATALOGUE] Omnt-sc1 RAPIDS TRACE3
Slide 48: SEEREHBER HY
SEEREHBER HY 333,772 12437118 9,422,386
Slide 49: A Workstation for Data Scientists
A WORKSTATION FOR DATA SCIENTISTS POWERED BY NVIDIA GPU AND CUDA-X Al Dual Quadro RTX 8000 with 96 GB Memory Pre-installed for CUDAXX Accelerated Data Science — RAPIDS, TensorFlow, PyTorch, Caffe, Anaconda Distribution 10X Faster
Slide 50: Acceleration of FSI Data Science Workflow
Acceleration of FSI Data Science Workflow
Slide 51: Announcing World's Top Computer Makers Offer
ANNOUNCING WORLD'S TOP COMPUTER MAKERS OFFER WORKSTATIONS OPTIMIZED FOR DATA SCIENCE POWERED BY NVIDIA GPU AND CUDA-X Al Dual Quadro RTX 8000 with 96 GB Memory Pre-installed for CUDAX Accelerated Data Science — RAPIDS, TensorFlow, PyTorch, Caffe, Anaconda Distribution 10X Faster
Slide 52: Supercomputer vs. Hyperscale
SUPERCOMPUTER vs. HYPERSCALE “|@ HYPERSCALE
Slide 53: Supercomputer | Capability Machine | Scale-up Architecture
Supercomputer | Capability Machine | Scale-up Architecture A ! SUPERCOMPUTER vs. HYPERSCALE ‘SUPERCOMPUTER L@ 1K 1M 18
Slide 54: Supercomputer vs. Hyperscale
SUPERCOMPUTER vs. HYPERSCALE Supercomputer | Capability Machine | Scale-up Architecture Hyperscale | Capacity Machine | Scale-out Architecture ‘SUPERCOMPUTER He! ®@
Slide 55: Data Science - The New HPC Challenge
DATA SCIENCE - THE NEW HPC CHALLENGE ‘SUPERCOMPUTER @ DATASCIENCE a HYPERSCALE | \ ! | a-E-a-e-e.
Slide 56: Data Science - The New HPC Challenge
DATA SCIENCE - THE NEW HPC CHALLENGE SUPERCOMPUTER LL DATA SCIENCE = ea fe | NVIDIA DGX-2 ae HYPERSCALE Al Supercomputer Appliance = 16x V100 | 2PF | 512GB HBM2 | |_| + |_| + |_| + |_| + |_| 8x MLNX IB
Slide 57: Data Science - The New HPC Challenge
DATA SCIENCE - THE NEW HPC CHALLENGE ‘SUPERCOMPUTER — DATASCIENCE NVIDIA DGX-2 oe eats Al Supercomputer Appliance 16x V100 | 2PF | 512GB HBM2 8x MLNX IB
Slide 58: Data Science - The New HPC Challenge
DATA SCIENCE - THE NEW HPC CHALLENGE ‘SUPERCOMPUTER L) — DATASCIENCE NVIDIA DGX-2 oe eats t Data Science Server Al Supercomputer Appliance X 4x Th | 260 TF FP16 | 64GB GDDR6 16x V100 | 2PF | 512GB HBM2 ar @ MLNX or BRCM EN 8x MLNX IB
Slide 59: Datacenter Becomes a Compute Engine
DATACENTER BECOMES A COMPUTE ENGINE 3 3 2
Slide 60: NVIDIA. Mellanox Technologies
> NVIDIA. Mellanox: TECHNOLOGIES
Slide 61: Eyal Waldman
A \ EYAL WALDMAN nvi D I A. Mellanox: CEO, Mellanox TECHNOLOGIES
Slide 62: NVIDIA DGX Pod - Zero to Data Science in Day
NVIDIA DGX POD — ZERO TO DATA SCIENCE IN 1 DAY “© purestorsce a) v — b ARISTA lutte 34. Mellanox
Slide 63: Announcing World's Most Popular Enterprise Servers
ANNOUNCING WORLD'S MOST POPULAR ENTERPRISE SERVERS OPTIMIZED FOR DATA SCIENCE Powered by NVIDIA T4 and New CUDA-X Al NGC Certified Fy) — cisco D@LLEMC = FUJITSU _Bewlett Packard Enterprise inspur sugon
Slide 64: Announcing World's Most Popular Enterprise Servers
ANNOUNCING WORLD'S MOST POPULAR ENTERPRISE SERVERS OPTIMIZED FOR DATA SCIENCE Powered by NVIDIA T4 and New CUDA-X Al NGC Certified abet peg FUJITSU HewtettPockra cisco LEMC ) Enterprise inspur sugon
Slide 65: Acceleration of Data Science Clusters
Acceleration of Data Science Clusters Deep Learning Scaling 80,000 End to End 38 3 % 60,000 . Ml 10x CPU Nodes 2 40,000 reining I 10X Té Nodes 3” g 20,000 Data Prep [> F 0 0 10 20 30 40 0 20 40 60 Minutes (smaller is better] Number of Servers
Slide 66: Acceleration of Data Science Clusters
End to End Training Data Prep |) Acceleration of Data Science Clusters 20 Minutes (smaller is better] 30 40 lH 10X CPU Nodes I 10X T4 Nodes Images per Second (KesNet-bU) Eo 60,000 40,000 20,000 Deep Learning Scaling With RoCE RDMA 20 40 Number of Servers 60 80
Slide 67: Matt Garman
aWws MATT GARMAN Vice President of Compute Services ~ 9 | Amazon Web Services
Slide 68: The World's Leading Brands Use Amazon EC2 P3 Instances
THE WORLD'S LEADING BRANDS USE AMAZON EC2 P3 INSTANCES WITH NVIDIA V100 GPUS 2 a AWS Q\airbnb Caltech CapitalOhe Go Scerner comcast inTuly lyf 4 <\ Mapillary see SAMSUNG SDS SIEMENS verizon’ WesternDigital yet | 42,
Slide 69: Introducing Amazon EC2 G4 Instances
INTRODUCING AMAZON EC2 G4 INSTANCES WITH NVIDIA T4 TENSOR CORE GPUS MACHINE LEARNING INFERENCE GRAPHICS INTENSIVE APPLICATIONS VIDEO TRANSCODING
Slide 70: Chapter 3: Robotics
CHAPTER 3: ROBOTICS
Slide 71: Jetson Powering Autonomous Machines
JETSON POWERING AUTONOMOUS MACHINES WAREHOUSE DELIVERY AGRICULTURE RETAIL INDUSTRIAL
Slide 72: Announcing Jetson Nano
ANNOUNCING JETSON NANO $99 NVIDIA CUDA-X Al COMPUTER CUDAX acceleration stack High-resolution sensor support Runs all CUDA.X Al models
Slide 73: Announcing Isaac Open SDK
ANNOUNCING ISAAC OPEN SDK KAYA INanol CARTER (Kavier) LINK Mutt Xavier z 8 Isaac Robot Engine ~ Modular robot framework Isaac Sim - Virtual robotics laboratory Isaac Gym - Reinforcement learning simulator Isaac Robot Apps ~ Kaya, Carter and Link Available at developer.nvidia.com/isaacsdk Ti ad S JET
Slide 74: Slide 74
Slide 75: DGX Saturn V
DGX Saturn V Constellation THE DRIVE INITIATIVE Xavier al DRIVE IX
Slide 76: Cars Trucks Mobility Services
CARS TRUCKS MOBILITY SERVICES TIER 1s AI ISVs MAPPING SIMULATION SENSORS
Slide 77: Announcing DRIVE AP2X Release 9.0
ANNOUNCING DRIVE AP2X RELEASE 9.0 HIGH FUNCTION L2+ AUTOPILOT SYSTEM On-ramp to off-ramp Surround perception for auto lane-change Localization to all major HD maps Real-time mapping with camera and radar Integrated AR and VR confidence visualization Integrated DM and voice command
Slide 78: Slide 78
Slide 79: DRIVE AV - Perception
DRIVE AV — PERCEPTION Lanes & Road Markings Distance
Slide 80: DRIVE AV - Perception / Localization & Mapping
DRIVE AV — PERCEPTION DRIVE AV — LOCALIZATION & MAPPING Lanes & Road Markings Distance TomTom — Barcelona, Spain Navinfo — Shanghai, China
Slide 81: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING
Slide 82: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING PREDICTION
Slide 83: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING PREDICTION Emergency Braking | SAFETY FORCE FIELD
Slide 84: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING PREDICTION Emergency Braking Intersection Handling | SAFETY
Slide 85: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING PREDICTION Emergency Braking Intersection Handling Intellig
Slide 86: Announcing DRIVE AV - Path
ANNOUNCING DRIVE AV — PATH PLANNING WITH SAFETY FORCE FIELD Path Planning + Prediction + Safety Force Field Designed for AV safety and drive comfort Mathematically verifiable, validated in simulation Open platform PATH PERCEPTION & PLANNING PREDICTION Intersection Handling Intelligent Steering Conge
Slide 87: Announcing DRIVE Constellation
ANNOUNCING DRIVE CONSTELLATION AVAILABLE NOW Virtual AV Test Fleet Bit-accurate, hardware-in-the-loop simulator Test corner and rare conditions Simulate previous failure scenarios Cloud-based workflow Open platform
Slide 88: Announcing DRIVE Constellation
ANNOUNCING DRIVE CONSTELLATION AVAILABLE NOW Virtual AV Test Fleet Bit-accurate, hardware-in-the-loop simulator Test corner and rare conditions Simulate previous failure scenarios Cloud-based workflow Open platform
Slide 89: Slide 89
Slide 90: Slide 90
Slide 91: Announcing DRIVE Constellation
ANNOUNCING DRIVE CONSTELLATION AVAILABLE NOW Virtual AV Test Fleet Bit-accurate, hardware-in-the-loop simulator Test corner and rare conditions Simulate previous failure scenarios Cloud-based workflow Open platform
Slide 92: Announcing NVIDIA and Tri-Ad Partner
ANNOUNCING NVIDIA AND TRI-AD PARTNER TO CREATE FUTURE OF AUTONOMOUS VEHICLES TOYOTA ee AV Computer \briving Systems Simulation
Slide 93: Slide 93
Slide 94: SoftBank / LG U+
=SoftBank @LGut
Slide 95: Slide 95
Slide 96: Slide 96
Slide 97: Slide 97