NVIDIA and H2O


Accelerate ML on GPUs
Joshua Patterson — NVIDIA
Arno Candel — H2O.ai
 NVIDIA
Leader in AI Computing

Gaming Pro Visualization Data Center Self-Driving Cars

GPU Computing

2
 AMAZING ACHIEVEMENTS IN AI

Play Go Play Doom Learn Paint Style Synthesize Voice

Write Captions Learn Motor Skills Learn to Walk Drive

3
 LIFE AFTER MOORE’S LAW
40 Years of Microprocessor Trend Data
107

106
Transistors

1.1X per year
105 (thousands)

104

103
1.5X per year
102
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

4
 RISE OF GPU COMPUTING

GPU-Computing perf 1000X

107 by
1.5X per year
APPLICATIONS 2025
106
ALGORITHMS 1.1X per year
105

SYSTEMS 104

103
CUDA 1.5X per year
102
Single-threaded perf
ARCHITECTURE 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

5
 NVIDIA GPU COMPUTING MODEL 

EVERYWHERE, ANYWHERE

ALL GPU DGX SYSTEMS CLOUD

Servers in Every Shape and Size The Essential AI Tools for Instant Productivity Everywhere

6
END-TO-END SOLUTIONS FOR DATA SCIENCE

EMBEDDED DESKTOP DATA CENTER ENTERPRISE

Jetson TX1, Drive PX2 DGX Station, Titan Xp Tesla V100 DGX-1

Inference at the Edge Accelerators for PCs Most advanced data center Fully integrated deep learning
GPU solution

7
 At a Glance
NVIDIA DGX GPUs 4x NVIDIA® Tesla® V100

STATION TFLOPS (GPU FP16)

GPU Memory
480
16 GB per GPU
SPECIFICATIONS NVIDIA Tensor Cores 2,560 (total)
NVIDIA CUDA Cores 20,480 (total)
Intel Xeon E5-2698 v4 2.2 GHz (20-
CPU
core)
System Memory 256 GB LRDIMM DDR4
Data: 3 x 1.92 TB SSD RAID 0

Storage Go
OS: 1 x 1.92 TB SSD
Network Dual 10 Gb LAN
Display 3x DisplayPort, 4K Resolution
Acoustics < 35 dB
Maximum Power Requirements 1500 W
Operating Temperature Range 10 - 30 oC
Ubuntu Desktop Linux OS
Software DGX Recommended GPU Driver
CUDA Toolkit

Learn more: www.nvidia.com/station

8
 DGX STATION
The Personal AI Supercomputer
VOLTA-POWERED DESIGNED FOR 

PERFORMANCE THE OFFICE EFFORTLESS PRODUCTIVITY

400 x86 CPU’s – 
 Desk-friendly 
 Experiment on Station


in a workstation Whisper-quiet Scale on DGX-1 / Cloud

9
 OUR DATA CENTER STRATEGY: 
 At a Glance
NVIDIA DGX-1
Highest Performance,
Fully Integrated System
8 TB SSD 8 x Tesla V100 16GB
960 TFLOPS

300 GB/s NVLink Hybrid

Cube Mesh

8x Tesla V100 16GB

2x Xeon | 8 TB RAID 0
Quad IB 100Gbps, Dual 10GbE
3U — 3200W
Learn more: www.nvidia.com/station

10
 NVIDIA GPU CLOUD
GPU-accelerated Cloud Platform Optimized for Deep Learning

Registry of
NVIDIA Containers, Datasets,
GPU CLOUD and Pre-trained models

CSPs

Containerized in NVDocker | Optimization across the full stack

Always up-to-date | Fully tested and maintained by NVIDIA | Beta in July

11
 How GPU Acceleration Works
Application Code

Compute-Intensive Functions
Rest of Sequential
CPU Code
GPU CPU

12
+
 GPU Acceleration In Action

Deep learning researcher & educator.

Founder: fast.ai; Faculty: USF & Singularity
University; // Previously - CEO: Enlitic;
President: Kaggle; CEO Fastmail

Rewrote @scikit_learn
PolynomialFeatures in
@ContinuumIO Numba. Got a 40x
speedup (would be bigger with more
data!) 12 lines of code

13
 GPU Acceleration In Action

14
 What’s machine learning?

15
 Bringing machine learning to data

DATABASES ETL SQL VISUALIZATION MACHINE LEARNING

DATA

GPU ACCELERATED

Reference blog: https://www.nextplatform.com/2017/05/08/crunching-machine-learning-

databases-together-gpus/
16
 Bringing machine learning to data

DATABASES ETL SQL VISUALIZATION MACHINE LEARNING

DATA

GPU ACCELERATED

Reference blog: https://www.nextplatform.com/2017/05/08/crunching-machine-learning-

databases-together-gpus/
17
 This is a team effort!

18
 Who is H2O.ai?

20
 10,000+ Companiesusing
10,000 Companies use H2O
H2O -— World
World Wide
Wide Community
Community Adoption
Adoption
Companies Using H2O.ai H2O.ai Users
A.C. Nielsen
 Bell Canada
 14,000 Case Western Reserve University

Delft University of Technology Network
 Enbridge Pipelines
 140,000
A1 Telekom Austria
 Delhi Technical University(Dce)
 Ency For Science Technology and Research

Beltelecom
 Catalina Marketing

AAPT
 Deloitte
 End-User Numericable

Beyond The Network America
 Catalina Marketing Oration

Abovenet Communications
 Deloitte Services
 Energy Sciences Network

Bezeq International-
 Cect-Chinacomm Communications

Academic Administrative and Research Network
 Deloitte Touche Tohmatsu Services
 Enom Orporated

Bh - Tec
 Cedars-Sinai Health Systems

Academic Computer Centre Cyfronet H
 Deloitte and Touch Regional Consulting Services
 Ensync Business Solutions Pty

Bharti Airtel
 Celgene Oration

Accelerated Data Works
 Delphon Industries
 Entanet International

Bibliotheque Nationale De France
 Center For Governmental Research

Accenture
 Delta Dental Plan of Michigan
 Enterprise Teaming

Big Fish Games
 Centerbeam

Accenture Services
 Delta Leasedline Network
 Enzu

Bigleaf Networks
 Central Telegraph Public Joint-Stock

Ace Ina Holdings
 Deluxe Oration
 Eotvos Lorand University of Sciences

Biglobe
 Centre De Calcul El-Khawarizmi - Cck
 Den Networks
 Epam Systems

Ace International

Ace Telecom

10,281
Bilink
 Centre For Advanced Computing
 Dena
 97,620
Epm Telecomunicaciones E.S.P.

Bimeh Dormitory Sharif University of Technology
 Centro De Tecnologia Da Informa O Renato Archer

Acton
 Deutsche Telekom
 Epsilon Data Manement Dba

Bio-Rad Laboratories
 Ceom Israel

Acxiom Oration
 Deutsches Reisebuero
 Equant

Biocontrol
 Cerfnet

Adamo Telecom Iberia
 Develon
 Equinox Consulting

Bisiness Network Jv
 Cerner Oration

Administracion Nacional De Telecomunicaciones
 Dhirubhai Ambani Institute of Information
 Erasmus Mc

Bite Communications
 Certara USA

Admiral Objekt Waesche & Arbeitskleidung
 Dialog Axiata
 Erasmus University Rotterdam

Biznet
 Ceu

Adobe Systems
 Digi Tavkozlesi Es Szolgaltato
 Ericsson Business Communications

Biznet Metronet
 Cgi Group

Adobe Systems India
 Digia
 Ericsson Network Systems

Blekinge Institute of Technology
 Champaign Telephone

Adsl Maroc Telecom
 Digital Entertainment
 Escout Consulting

Blue Line Infotech
 Charles University

Advanced Cable Communications
 Digital Hosting Technology
 Espn

Charlesbrauer

Advanced Computer Solutions
 6,427 Blueconnect

Boingo Wireless
 Charter Communications

Digital Network Associates - Franchisee
 Estate Valuations and Pricing Systems

Affecto
 Digital Ocean
 Etapa Ep

Afrihost-Dynamic

Bol.Com Bv

Boots UK Retail

Chegg

Chengdu West Dimension Digital Technology

Digital Realm
 54,163 Etex Communications

Ainet Telekommunikations-Netzwerk Betriebs
 Digital River
 Etheric Networks

Boranet
 Cheonanjeonhwakukjang

Air Bank A.S.
 Digital-Entertainment-Industry-Development-Co--Zhongshan Zho
 Ethio Telecom

Borlange Energi
 Chico Board of Trade

Air Liquide Sa
 Digitalocean Cloud
 Ethz Swiss Federal Institute of Technology Zurich

Boston Scientific Oration
 China Digital Kingdom Technology

Airess Cesko
 Bouygues Telecom Division Mobile
 China Education and Research Network

Direct Supply
 38,257 Etisalat Lanka (Private)

Akamai Technologies
 3,810 Bouygues Telecom Sa
 Chinatelecom Group Beijing Co

Discoveries In Sight

Dishnet Wireless

European Bioinformatics Institute

Evergy

Aktia Saastopankki Oy
 Chongqing Times Newper Office

Brain Telecommunication
 Distributel Communications
 Excell Media

Aktiv-I Szolgaltato
 Chs - Bna Lan

Bright House Networks
 Disy Informationssysteme
 Exe2 Newton Abbot

Al-Shahad Information Technology
 Chunghwa Telecom Data Communication Business Group

Brighthouse Networks Cfl Division
 Diverge Consulting
 Exetel Act Dsl

Albert Einstein College of Medicine of Yeshiva University
 Cik Telecom

Brighthouse Networks Indianapolis
 Dna Oy
 Exponential-E
Albert-Ludwigs-Universitaet Freiburg
 Cisco

Bristish Petroleum
 Doclernet
 FPL Fibernet

Alexander & Alexander Information Technology
 Cisco Systems

British Sky Broadcasting
 Dongbeicaijingdaxue-Dl-Ln
 Facebook

Algar Telecom
 Cisco Systems Ironport Division

Broadriver Communication
 Doorway As
 Fachhochschule Dortmund

Aliyun Computing
 Citadel Investment Group L.L.C.

Broadstripe
 Dotomi
 Fachhochschule Nordwestschweiz

Allbusiness.Com
 Citrix Systems

Brutele Sc
 Drivetime
 Faculty of Sciences University of Lisbon
Allianz Maned Operations & Services Se
 City University
Bryant University

16

ow

l
15

oa
20

20
15

16

ow

N
oa

G
20

20

17
N

20
17
20

21
H2O.ai Select Paying Customers

Retail Healthcare Marketing Financial Advisory & Insurance Telecom

Accounting

“Overall customer satisfaction is very high.” - Gartner

22
 AI in Financial Services
Wholesale / Commercial Banking IT Infrastructure
• Know Your Customers (KYC) • Security Cyberlake
• Anti-Money Laundering • DoS Detection and Protection
(AML) • Master Data Management

Retail Banking Card/Payments Business

• Deposit Fraud • Transaction Frauds
• Customer Churn Prediction • Real-time Targeting
• Auto-Loan • Credit Risk Scoring
• In-Context Promotion

23
 AI in Healthcare
Medical Claim Fraud Detection Early Cancer Detection / Oncology

Flu Season Prediction

Medical Imaging and Diagnostics

Drug Discovery

Personalized Drug Matching

Emergency Room and Hospital Management

Product Recommendation

Remote Patient Monitoring

24
H2O.ai Strongly Positioned in Key Analyst Reports
H2O.ai is a Visionary 
 H2O.ai is a Strong Performer

in the Gartner Magic Quadrant
 in the Forrester Predictive H2O.ai Deep Water Included in
for Data Science Platforms Gartner Deep Learning Report
Analytics & Machine Learning

Publish: January 2017

“H2O had the highest reference “H2O.ai has significant adoption by
customer analytics support score large enterprises such as Macy’s,
H2O.ai named alongside Caffe, Facebook
of all the vendors.” Comcast, and Capital One.” Torch, Google TensorFlow, and Intel
Nervana, as a platform that assists users in
“H2O is especially suited to IoT “H2O.ai is best known for developing creating their own deep-learning and AI
open source, cluster-distributed ML solutions.
edge and device scenarios.” algorithms at a time (2011) when big
data demanded them, but no one else
“Overall customer had them.”
satisfaction is very high.”
25
 The Road Ahead

26
 H2O AI Platform Timeline
Visual
Interpretation
Analysts
Auto ML

App Developers
Deep Learning
H2O
Dev Ops Steam AI Edition
Q3 2017

Developers/ Data.table
Engineers
Sparkling
Advanced Data Water H2O GPU
Scientists Edition
H2O Core
GPU ASIC
Users 2012 2014 2016 2017 2018 2019

Roadmap
27
 Accuracy, Speed
and Interpretability

28
 https://www.youtube.com/watch?v=LrC3mBNG7WU
29
https://www.youtube.com/watch?v=4RKSXNfreLE

30
 171 with latest solver

87

51

https://www.youtube.com/watch?v=NkeSDrifJdg

31
32
 This performance based on
NVIDIA’s technology will
lead to…

33
 Driverless AI for the Digital Brain — Enabled by Fast Model Training

Da
H2O Customers

ta
Business Leaders
Visual Model Interpretation Model Fitness

Pipeline Driverless AI

Feature
Engine Auto ML Deploy
Deep Learning
Algorithms
Data Prep

Distributed Multi-CPU Multi-GPU

H2O Kaggle
Grandmasters
Model Repository
H2O PhDs &
Professors
H2O Systems Engineers
Accuracy Speed Interpretability
34
 Driverless AI on GPUs

https://www.youtube.com/watch?v=KkvWX3FD7yI
35
 Driverless AI — Competitive with Kagglers!

Top 8 position in Kaggle with zero manual labor!

(ranked above multiple Kaggle Grandmasters)

https://www.kaggle.com/c/mercedes-
benz-greener-manufacturing/leaderboard

36
 Model Interpretability — Insights Through Computing

37
38
 GPU OPEN ANALYTICS INITIATIVE
github.com/gpuopenanalytics

Exploratory ML/DL
Scoring
Analysis Algorithms
Ingest/

Parse
Feature Model

Grid Search
Engineering Export

GPU Data Frame (GDF)

39
Thank You