MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand

Page created by Gail Barnes
 
CONTINUE READING
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
MARC HAMILTON
VP Solutions Architecture and Engineering
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
NVIDIA
                “THE AI COMPUTING COMPANY”

GPU COMPUTING          COMPUTER GRAPHICS     ARTIFICIAL INTELLIGENCE
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
THE PROMISE OF AI
                                                                                                                                                                                                                                   16T
                                                                                                                                                                                                                                Global GDP Boost
                                                                                                                                                                                                                                  by AI by 2030
Increased access to healthcare
Improved patient outcomes
Safer cities
Safer & more efficient transportation
Intelligent manufacturing                                                                                                                                                                                                         58M
                                                                                                                                                                                                                                New Jobs Created
                                                                                                                                                                                                                                  by AI by 2022

                                                                                                                                                                                                                                   54%
                                                                                                                                                                                                                             Jobs Requiring Reskilling
                                                                                                                                                                                                                                     by 2022

                                        SOURCE: AI could contribute up to $15.7 trillion to the global economy in 2030, PwC, “Sizing the prize: What’s the real value of AI for your business and how can you capitalise?”
                                                                 58 million new jobs created by AI by 2022 and 54% of jobs requiring reskilling, World Economic Forum, “The Future of Jobs.”
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
AI IS FUELING GLOBAL INDUSTRIES
                               Multi-Trillion Dollar Global Industries Turning to AI

SMART CITIES   PUBLIC SAFETY       HEALTHCARE                       STARTUPS           INDUSTRIAL   TRANSPORTATION
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
GPU TECHNOLOGY UPDATE
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
RISE OF GPU COMPUTING

                107                                                                       GPU-Computing perf
                                                                                                                                                                 1000X
                                                                                          1.5X per year
 APPLICATIONS                                                                                                                                                    by 2025
                106
 ALGORITHMS                                                                                                                   1.1X per year
                105

   SYSTEMS      104

    CUDA        103
                                                                                         1.5X per year
                102

ARCHITECTURE                             Single-threaded perf

                      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
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
A YEAR OF RAPID GROWTH

25% MORE TOP500 SUPERCOMPUTERS        50% GROWTH OF NVIDIA DEVELOPERS                   600+ CUDA APPS               MORE PERF SAME GPU

                              125                                                                                                      40X
                                                        1.2M
       98                                  800K
                                                      DEVELOPERS
                                                         +50%
                                                                                                                     25X
                                                                          CRYOSPARC         FUN3D        GROMACS
                                                                            Cryo             CFD         Chemistry
                                            2018         2019
                                                                                                                             AMBER
                                                                                                                            CHROMA
      #1 World, US — ORNL Summit
                                                                                                                            LAMMPS
       #1 Europe — CSCS Piz Daint
          #1 Japan — AIST ABCI                           13M                                                                  MILC
                                                        CUDA                                                                 NAMD
      22 of Top 25 Energy-Efficient
                                                      DOWNLOADS                                                         Quantum Espresso
                                            8M                                                                                SF3D
                                                         +60%
                                                                        MICROVOLUTION     PARABRICKS       WRF
                                                                          Microscopy       Genomics      Weather
       2018                  2019           2018         2019                                                        2018                  2019
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
DATA SCIENCE – A NEW PILLAR OF DISCOVERY

                                                           AI
                                                                              PREDICTIVE
                                    FEATURES    NLU                  CV
                                                                                MODEL
                    DATA
   DATA                                                                                       INFERENCE          PREDICTION
                  ANALYTICS
CSV, PARQ, HDFS
                                                           ML

                                                           DL

                  ETL, Pandas,                    TensorFlow, PyTorch                       TensorFlow Serving
                  Spark, Graph                 MXNet, Scikit-Learn, XGBoost                ONNX, SageMaker NEO
MARC HAMILTON VP Solutions Architecture and Engineering - Search | NVIDIA On-Demand
DATA SCIENCE – A NEW PILLAR OF DISCOVERY

                                        AI
                                                    PREDICTIVE
                      FEATURES   NLU           CV
                                                      MODEL
         DATA
DATA                                                             INFERENCE   PREDICTION
       ANALYTICS

                                        ML

                                        DL

          cuIO                         cuDNN                      TensorRT
          cuDF                          cuML                        TRTIS
        cuGraph
DATA        DEEP     MACHINE       HYPERSCALE   RENDERING
                                                 SCIENCE
                                                           ANALYTICS   LEARNING   LEARNING       INFERENCE     & VIZ

NGC APPLICATION
ACCELERATION STACKS
WEALTH OF ACCELERATED APPS MAXIMIZE DATACENTER                                           CUDA
THROUGHPUT, UTILIZATION, EFFICIENCY
                                                                                         GPU
NVIDIA CUDA-X AI ECOSYSTEM

 FRAMEWORKS                     CLOUD ML SERVICES                                         DEPLOYMENT
                                                    Amazon      Google                Amazon
                                                    SageMaker   Cloud ML              SageMaker Neo

                           Azure Machine Learning                                                       Serving

DA                 GRAPH                            ML                     DL TRAIN                   DL INFERENCE

                                          CUDA-X AI

                                               CUDA

     Workstation                              Server                                            Cloud
ANNOUNCING WORLD’S LEADING TECH COMPANIES
                                ADOPT CUDA-X AI TO ACCELERATE MODEL DEPLOYMENT

       6X TensorRT Downloads
                                                                                 Voice Search
                                                                                 Image Search
                                                                                 Recommendations
                                                                                 Home Assistant
                        300K
                                                                                 News Feed
                                                                                 Translation
50K
                                                                                 eCommerce

2017                     2018
ANNOUNCING WORLD’S TOP COMPUTER MAKERS
                                                                   OFFER WORKSTATIONS OPTIMIZED FOR DATA SCIENCE

                 Acceleration of FSI Data Science Workflow                                                         POWERED BY NVIDIA GPU AND CUDA-X AI
End to end                                                                                                         Dual Quadro RTX 8000 with 96 GB Memory
                                                                                                                   Pre-installed for CUDA-X Accelerated Data Science —
  Training                                                                                                         RAPIDS, TensorFlow, PyTorch, Caffe, Anaconda
                                                                                                                   Distribution
Data Prep                                                                                                          10X Faster
             0          2           4              6     8    10

                                         Minutes

                            2x RTX8000     1x RTX8000   CPU
SUPERCOMPUTER vs. HYPERSCALE

                         1T PF

   COMPUTE REQUIREMENT
                                 SUPERCOMPUTER

                         1B PF

                         1M PF

                         1K PF

                          1PF
                                                  HYPERSCALE

                          1TF

                                   1K        1M    1B
                                                  #CCU
SUPERCOMPUTER vs. HYPERSCALE

Supercomputer | Capability Machine | Scale-up Architecture

                                                                                      1T PF

                                                                COMPUTE REQUIREMENT
                                                                                              SUPERCOMPUTER

                                                                                      1B PF

                                                                                      1M PF

                                                                                      1K PF

                                                                                       1PF
                                                                                                               HYPERSCALE

                                                                                       1TF

                                                                                                1K        1M    1B
                                                                                                               #CCU
ANNOUNCING 3X PERFORMANCE ON SUMMIT FOR HPL-AI
                                     HPL-AI: A New Approach to Benchmarking AI Supercomputing

  FUSION OF HPC & AI                              HPL-AI & ITERATIVE REFINEMENT SOLVERS          3X MORE PERF ON SUMMIT w/ TENSOR CORE GPUs

                                                                                                                        436 PF

     HPC (Simulation) – FP64

                                                                                                          149 PF

AI (Machine Learning) – FP16, FP32                       Proposed by Prof Jack Dongarra, et al              FP64       Mixed Precision
                                                                                                            (HPL)         (HPL-AI)
SUPERCOMPUTER vs. HYPERSCALE

Supercomputer | Capability Machine | Scale-up Architecture                                                                  Hyperscale | Capacity Machine | Scale-out Architecture

                                                                                      1T PF

                                                                COMPUTE REQUIREMENT
                                                                                              SUPERCOMPUTER

                                                                                      1B PF

                                                                                      1M PF

                                                                                      1K PF

                                                                                       1PF
                                                                                                               HYPERSCALE

                                                                                       1TF

                                                                                                1K        1M    1B
                                                                                                                                                                                …
                                                                                                               #CCU
DATA SCIENCE – THE NEW HPC CHALLENGE

Supercomputer | Capability Machine | Scale-up Architecture                                                                               Hyperscale | Capacity Machine | Scale-out Architecture

                                                                                          1T PF

                                                                    COMPUTE REQUIREMENT
                                                                                                  SUPERCOMPUTER

                                                                                          1B PF

                                                                                                                   DATA SCIENCE
                                                                                          1M PF

                                                                                          1K PF

                                                                                           1PF
                                                                                                                            HYPERSCALE

                                                                                           1TF

                                                                                                    1K        1M             1B
                                                                                                                                                                                             …
                                                                                                                            #CCU
DATA SCIENCE – THE NEW HPC CHALLENGE

                                                                                                           Hyperscale | Capacity Machine | Scale-out Architecture

                                                            1T PF

                                      COMPUTE REQUIREMENT
                                                                    SUPERCOMPUTER

                                                            1B PF

                                                                                     DATA SCIENCE
                                                            1M PF

                                                            1K PF

                                                             1PF
         NVIDIA DGX-2                                                                         HYPERSCALE
  AI Supercomputer Appliance
                                                             1TF
16x V100 | 2 PF | 512GB HBM2
          8x MLNX IB
                                                                      1K        1M             1B
                                                                                                                                                               …
                                                                                              #CCU
DATA SCIENCE – THE NEW HPC CHALLENGE

                                                                                                           Hyperscale | Capacity Machine | Scale-out Architecture

                                                            1T PF

                                      COMPUTE REQUIREMENT
                                                                    SUPERCOMPUTER

                                                            1B PF

                                                                                     DATA SCIENCE
                                                            1M PF

                                                            1K PF

                                                             1PF
         NVIDIA DGX-2                                                                         HYPERSCALE
  AI Supercomputer Appliance
                                                             1TF
16x V100 | 2 PF | 512GB HBM2
          8x MLNX IB
                                                                      1K        1M             1B
                                                                                                                                                               …
                                                                                              #CCU
DATA SCIENCE – THE NEW HPC CHALLENGE

                                                            1T PF

                                      COMPUTE REQUIREMENT
                                                                    SUPERCOMPUTER

                                                            1B PF

                                                                                     DATA SCIENCE
                                                            1M PF

                                                            1K PF

                                                             1PF
         NVIDIA DGX-2                                                                         HYPERSCALE        Data Science Server
  AI Supercomputer Appliance                                                                               4x T4 | 260 TF FP16 | 64GB
                                                             1TF
16x V100 | 2 PF | 512GB HBM2                                                                                          GDDR6
          8x MLNX IB                                                                                             MLNX or BRCM EN
                                                                      1K        1M             1B
                                                                                             #CCU
CUDA TO ARM
    Energy-Efficient Supercomputing

                                      NVIDIA GPU Accelerated Computing Platform On ARM
&                                     Optimized CUDA-X HPC & AI Software Stack
                                      CUDA, Development Tools and Compilers

                                      Available End of 2019
NVIDIA AI INFRASTRUCTURE
DATACENTER BECOMES A COMPUTE ENGINE
AI LEADERSHIP NEEDS AI INFRASTRUCTURE LEADERSHIP

                                  DATA                 AI MODEL
                                            MACHINE
                                            LEARNING
SATURNV
The Worlds Largest Enterprise
AI Infrastructure Buildout

1500 DGX Nodes
12,600 GPUs
1.5 ExaFLOPs
5MW Average Power
NVIDIA DGX SUPERPOD
AI Leadership Requires
AI Infrastructure Leadership

Test Bed for Highest Performance Scale-Up Systems
9.4 PF on HPL | ~200 AI PF | #22 on Top500 list
AI IN INDUSTRY
BUILDING AI & DEPLOYING AI
                                                                                 PLAN

         DATA      MACHINE                                                                          ACTUATORS
       ANALYTICS   LEARNING
DATA                                                       SENSORS
                                   AI MODEL                                     AI MODEL
                                  VALIDATION

                                                                     PERCEIVE              REASON
THIS IS AI
                       PLAN

SENSORS                                    ACTUATORS

          PERCEIVE                REASON
THIS IS AI
                       PLAN

SENSORS                                    ACTUATORS

          PERCEIVE                REASON
JETSON POWERING AUTONOMOUS MACHINES

WAREHOUSE     DELIVERY    AGRICULTURE    RETAIL   INDUSTRIAL
ANNOUNCING JETSON NANO

                         $99 NVIDIA CUDA-X AI COMPUTER
                         CUDA-X acceleration stack
                         High-resolution sensor support
                         Runs all CUDA-X AI models
ANNOUNCING ISAAC OPEN SDK

               KAYA (Nano)                   CARTER (Xavier)           LINK (Multi Xavier)

                                                                                                                                        Isaac Robot Engine – Modular robot framework
  Sensor and
                                                                                                                                        Isaac Sim - Virtual robotics laboratory
Actuator Drivers        Core Libraries            GEMS          Reference DNN                Tools
                                                                                                                                        Isaac Gym – Reinforcement learning simulator
                                           ISAAC OPEN TOOLBOX
                                                                                                                                        Isaac Robot Apps – Kaya, Carter and Link
                                                CUDA-X

                                                                                                                                        Available at developer.nvidia.com/isaac-sdk

              JETSON NANO                      JETSON TX2              JETSON AGX XAVIER

                                         Isaac Robot Engine                                             Isaac Sim           Isaac Gym
NVIDIA DELIVERS MORE THAN 6000X SPEEDUP
ON BENCHMARK ALGORITHM FOR HEDGE FUNDS                                                                   Simulations / Hour

IN PYTHON
                                                                                                                              20,000,000
                                                                                20,000,000
20,000,000 simulations per hour on an industry defined benchmark,
compared to the prior record of 3,200.
Over 6,000 times faster on DGX-2                                                15,000,000
Powered by NVIDIA DGX-2, using all 16x V100 GPUs
Accelerated Python via the RAPIDS packages and Numba –                          10,000,000
replicate this performance without needing in-depth knowledge
of GPU programming
How will you use that power – faster time to market?                             5,000,000
More complex models? Test more scenarios? Some of each?
                                                                                                 3,200
                                                                                        0
                                                                                             20xCloud Nodes                     DGX-2

https://blogs.nvidia.com/blog/2019/05/13/accelerated-backtesting-hedge-funds/
www.STACresearch.com/news/2019/05/13/NVDA190425
REAL-TIME FRAUD DETECTION
Recently, PayPal was looking to deploy a new fraud detection
system. The team working on it set a high bar: this system had to
operate worldwide 24/7, and work in real-time to protect customer
transactions from potential fraud. In spec’ing the system, it became
evident that CPU-only servers couldn’t meet these requirements.

Using NVIDIA T4 GPUs, PayPal delivered a new level of service,
using GPU inference to improve real-time fraud detection by 10%
while lowering server capacity by nearly 8x.
NVIDIA METROPOLIS
Smart City Platform

Billions of IoT sensors                                                                                                                                  500M+
                                                                                                                                                             Sensors
The data lifeblood of a modern city                                                                                                                      Worldwide Today
The fuel for AI software:
      Reducing traffic congestion
      Energy grid management
      Finding lost children
      Other new services                                                                                                                                 $158B
                                                                                                                                                           Smart Cities
                                                                                                                                                         Funding by 2022

                                      SOURCE: $158B, Smart Cities Initiative funding by 2022, IDC, “Worldwide Semiannual Smart Cities Spending Guide.”
THIS IS AI
                       PLAN

SENSORS                                    ACTUATORS

          PERCEIVE                REASON
THE DRIVE INITIATIVE

DGX Saturn V   Constellation        Xavier

  DRIVE AV       DRIVE IX         KITT Resim
NVIDIA SELF-DRIVING CAR
                                                                             PLAN

         DATA      MACHINE                                                                   ACTUATORS
       ANALYTICS   LEARNING
DATA                                                    SENSORS
                                 VALIDATION
                                  AI MODEL
                                 VALIDATION

                                                                  PERCEIVE          REASON
AI FOR TRANSPORTATION

AD is revolutionizing transportation                          1.5B
                                                               Vehicles in
Saving lives                                                   the World
Reducing shipping costs
Reduced insurance costs
Vehicle of future is software defined
NVIDIA DRIVE – an open platform for research and production
                                                              $10T
                                                              Transportation
                                                                 Industry
AI FOR TRANSPORTATION: NVIDIA DRIVE
HEALTHCARE DATA IS ENORMOUS
                      The Perfect Fuel for AI

Genomics Data             Instrument Data       Hospital Data
  2x/7Months                 3+ TB/day            50 PB/Year
MEDICAL IMAGING
Essential tool of early detection and disease management
Demand outpacing supply of world’s radiologists                                                                                                                     70%
                                                                                                                                                                  Medical Imaging
Imaging field enormously complex                                                                                                                                 Research based on
                                                                                                                                                                     DL Today
Perfect application for AI

                                                                                                                                                                 $8.6B
                                                                                                                                                                  Annual Software
                                                                                                                                                                 Revenue for AI Use
                                                                                                                   DL-BASED IMAGE   DL-BASED BRAIN   CINEMATIC     Cases by 2025
                                                                                                                  RECONSTRUCTION    SEGMENTATION     RENDERING

SOURCE: Global software revenue from 22 key healthcare AI use cases will grow to $8.6 billion annually by 2025,
Tractica, “Artificial Intelligence for Healthcare Applications.”
CLARA AI TOOLKIT

PRE-TRAINED MODELS   AI-ASSISTED ANNOTATION            TRANSFER LEARNING   AI DEPLOYMENT
AI-ASSISTED ANNOTATION
15 Datacenters
GeForce NOW     300K Players       Announcing
 500+ Games   1M on Waiting List   GFN Alliance
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
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
NVIDIA RTX SERVER

RENDERING        OMNIVERSE   GEFORCE NOW
PILLARS OF NATIONAL AI INITIATIVES

                                                                                                                  Affordable,
                                                                       University-Industry    Safe, Efficient
        Industry-Standard Platform                   Basic Research                                                Accessible
                                                                         Collaboration        Transportation
                                                                                                                  Healthcare

Rich Software             State-of-the-Art           Vibrant Startup       Reskilling        Hyper-Productive     Smart, Safe
  Ecosystem                AI Computing                Community           Workforce          Manufacturing         Cities

        AI CENTERS OF                                           SKILLING                                   INDUSTRY
         EXCELLENCE                                          AND RESKILLING                               SOLUTIONS
NVIDIA PARTNERSHIP FRAMEWORK

   Every System Maker, Every Cloud                 NVIDIA        NVAIL University   Transportation     Healthcare
                                                  Research        Collaboration      NVIDIA DRIVE     NVIDIA CLARA

                                              NVIDIA Inception      NVIDIA DLI        Robotics            AI City
NVIDIA AI                  NVIDIA DGX         Startup Program        Training       NVIDIA ISAAC     NVIDIA METROPOLIS
Software

     TECHNOLOGY &                                        EXPERTISE &                    INDUSTRY SOLUTION
       ECOSYSTEM                                         INVESTMENT                         PLATFORMS
You can also read