Unleashing Data Value with High Performance Computing and Artificial Intelligence
Todd ChurchwardHPC and AI Solution ArchitectHPE
28th August 2019
HPC delivers compelling financial returnsUp to $463 dollars on average in revenue and up to $44 on average of profits (or cost savings) per dollar invested in HPC
2
Avg. of Revenue $ per HPC $
Avg. of Profit or Cost Savings $
per HPC $
$834
$61
$416
$54
$160
$41
$83
$20
Transforming explorationand production
Enhancing patient care and predictive medicine
Automating product lifecycle management
Financial Services Oil and Gas Life Sciences Manufacturing
Hyperion Research Study of HPC ROI, June 2018The reports and data are posted at: www.hpcuserforum.com/ROI
Accelerating trading &portfolio management
Commercial in confidence
The scientific data explosion
3
15 TB/Day1 Exabyte/Day
[1PB of correlated data/Day]
Personal Genomics
10X Increase in Data Volume
Commercial in confidence
The data explosion is occurring everywhere
4Commercial in confidence
5
Convergence of IoT, Big Data and AI is creating a Data Tsunami
TeraByte 1012
PetaByte 1015
ExaByte 1018
ZettaByte 1021
IDC predicts more than 40ZB of data will be generated in 2020, 37% of which will have significant business value!
Commercial in confidence
- Facebook friend tagging- Facial recognition technology- Pinterest – find similar pins
- All voice assistants use AI techniques- Understands intent and context to respond
accurately- Virtual chat assistants for automated
customer service
- Search and optimization are two big use cases of AI
- Shows shortest path to reach destination using AI techniques
- Dynamic pricing during peak/low demand- Best match for ride share to minimize detour
- Smart email filters- Email categorization- Gmail successfully filters 99%
of spam using AI
- Instant credit decisions- Fraud detection and prevention- For 25 yrs., FICO uses AI for credit
decisions, threat detection and cybersecurity
Data driven AI experiences are already an integral part of our daily lives
6Commercial in confidence
AI and advanced analytics are moving into the enterprise mainstreamUnlocking data and enabling digital transformation
7
HealthPersonalized medicine, image analytics
ManufacturingPredictive and prescriptive maintenance
Consumer techNatural language interfaces
Financial servicesFraud detection, ID verification
GovernmentCyber-security, smart cities and utilities
EnergySeismic and reservoir modeling
Service providersMedia delivery
RetailVideo surveillance, shopping patterns
Commercial in confidence
HPE is making AI enterprise-grade
8
Sensor data analyzed during testing and racing, informs changes that should be
made to the car or driving strategy
Converging AI, IT and OTto boost output and quality
Accelerating Alzheimer’s research100x using in-memory analytics
Image Based Quality Control
9
Scalable and flexible edge-to-core AI solution
Real-time detection, classification and location of defects
AI models trained and evaluated at the datacenter on HPE Apollo
CI/CD Pipeline to publish models as micro-services
Real-time inference at the edge on HPE Edgeline
Central operational dashboard
Visualization at the edge with labelling interface and local dashboard
Problem: Cost of poor quality (CoPQ) is estimated to account for 5-30% of gross sales. Traditional image analysis techniques based on hard coded rules or pixel-wise comparisons are rigid, complex and inefficient
Improve defect detection rate by 90%
Increase productivity by 50%
Improve flexibility of processes
Automate and reduce deployment times
Use Case Description Business Impact
Why HPE: Agile approach from AI Transformation Workshop, to Proof of Value and end to end deployment to productionCustomer reference:
Solution Approach:Converging AI, IT and OT to automate quality control processesApplying deep learning to surface quality monitoring, from raw materials to finished products
Functionalities
Watch the video
Supercomputing and HPC contributes invaluable and proven techniques for dealing with the data tsunami
Commercial in confidence
XFS All-Flash
11
HPC & AI | Data Management
11
HPC – AI – Machine Learning Cluster
High-Performance Storage
Lustre
HPE Data Management Framework• Metadata management & data provenance• Policy-based data migration with job scheduler integration• Data protection, repair and disaster recovery
TapeZero Watt Storage
Object Storage & Cloud
Software
Defined
Storage
Tier ZeroGPFS
Commercial in confidence
Dynamic or Static Namespaces
Policy Engine Data Movers
File
Object
Metadata
Cassandra
Changelogs
Open big data technologies with proven scalability
Modern open source architecture• Kafka for Changelog processing
• Cassandra for Scalable Metadata
• Mesos for Task Scheduling• Spark for Query Engine
− Scalable for capacity and performance
− Flexible new ways to manage data with the ability to create/delete/recover namespaces
12Commercial in confidence
Flexibility and extreme scale for Linux HPC workloads
13
*
5U, 4-socket chassis
Unparalleled Scale– Modular scale-up architecture– Scales seamlessly from 4s to 32s as a single system – Intel Cascade Lake processors– 768GB-48TB of shared memory– High bandwidth (13.3GB/sec)/low latency (<400ns) shared memory interconnect
Unbounded I/0– Up to 128 PCIe standup cards, LP/FH PCIe
Optimum Flexibility– 4-socket chassis building blocks, low entry cost; HPE nPARs– Nvidia GPUs, Intel SDVis– 1/10/25 Gbe, 16GbFC, IB EDR/Ethernet 100gb, Omni-Path– SAS, Multi-Rail LNet for Lustre; NVMe SSD– MPI, OpenMP
Extreme Availability– Advanced memory resilience, Firmware First, diagnostic engine, self-healing– HPE Serviceguard for Linux
Simplified User Experience– HPE OneVew management*, IRS, OpenStack– HPE Proactive Care
Scales up to 32 sockets and 48TB of memory in a single system
Commercial in confidence
Enterprises are embarking on a Hybrid IT journey
- 74% of customers surveyed will be operating in multi-cloud environments in 2 years
- 59% of those will rely on Hybrid IT interoperability
- 70% of data will be generated at the edge in 3 years
Deliver new productsand services faster
Keep theorganization running
Apps and data The Right Mix
Edge
Managedcloud
Privatecloud
Publiccloud
Owned data centers
Centers of Data
SaaS applications
Multi-public clouds
IoT
451 Research. Voice of the Enterprise Quarterly Advisory Report. 2015 – 2017.M2M Global Forecast & Analysis 2011-22
Commercial in confidence
Containers enable the Hybrid IT vision
Commercial in confidence
HPE AI Infrastructure Software Portfolio
16
Workload Management
AI Productivity Software
PBS Professional
Bare MetalContainers
AI Productivity Software
Docker & Singularity container repository Driverless AI
AI Frameworks
BlueData EPIC
Bright Data ScienceDownload, compile and install software from various sources (HPE white papers available)
Deep Learning Cookbook: - Deep Learning Benchmarking Suite- Deep Learning Performance Guide
Infrastructure Software
HPE Performance Cluster Manager for HPE Apollo, ProLiant, SGI 8600
Operating System
Bright Cluster Manager
Containers Singularity Community Edition Singularity Pro
InfiniBand100Gb Ethernet
FabricSoftware
System Management
Deep Learning Cookbook: - Deep Learning Benchmarking Suite- Deep Learning Performance Guide
Commercial in confidence
BlueData Recently acquired by Hewlett Packard Enterprise
17
• A leading provider of infrastructure software for deploying AI / ML and big data analytics
• Deliver innovative container-based solutions and services that help customers speed deployments, increase agility, and reduce costs
• Experts at helping customers accelerate their AI and data-driven transformation
• Solutions providing the performance, agility and efficiency benefits of Docker containers
• AI / ML and Big Data Analytics applications on-premises, in the public cloud or in a hybrid architecture
• Deployment of AI / ML and Big Data environments within minutes
Who are BlueData? What does BlueData deliver?
Commercial in confidence
BlueData App Store
− A library of unmodified Docker-based application images
− Big Data Applications in a single click
− Ability to add or modify applications / tools / services
− Capability to create clustered environments (e.g. Hadoop, Spark, Kafka, TensorFlow, etc.)
Commercial in confidence
The data analytics, AI and HPC pipelines
Commercial in confidence
Each autonomous car will generate 4TB of data per hour, from many sensors- 16+ 4K video cameras- 12 radar sensors- 6 ultrasonic sensors- 5 LIDAR sensors
It is estimated that each car would require 300 TFlops of computing power or a network pipe of 25-40GB/s to enable data to be moved from edge to core
https://a16z.com/2016/12/16/the-end-of-cloud-computing/
The intelligent edge
Commercial in confidence
HPE enables real time analytics - from intelligent edge to the core
Intelligent Edge(Track Side )
High Speed Storage Core data centre(deep learning)
TRAINING DATAEDGEDATA
HPE Apollo 6500 Gen10
WekaIO
NVIDIA®
Tesla® GPU accelerators
HPE System Management
Software
IOTDevices
HPE Aruba
HPE Edgeline
HPE Networking
HPE Pointnext
HPE OneView
HPE DMF
HPE Apollo
HPE ProLiant
HPE Apollo
HPE ProLiant
HPE Synergy
InfiniBand
HPE Infosight
HPE Introspect
HPE Apollo 21Commercial in confidence
Thank [email protected]
22