Download Leadership and innovations in Artificial Intelligence, Deep Machine

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Leadership and
innovations in
Artificial Intelligence,
Deep Machine
Learning, and other
trends in Linux
environment
Alexey Simanovskiy
IBM EE/A
Source: Gartner, “Architecting the On-Demand Digital Business”; Drue Reeves, Kyle Hilgendorf, Kirk Knoernschild, August 16, 2016
2
AI MOMENTUM
Today
By 2020
By 2020
By 2020
3000
85%
$47B
20%
AI startups
of all customer
service interactions
will be powered by AI
bots
spend on AI
technologies
of companies will
dedicate workers to
monitor and guide
neural networks.
3
Overall Artificial Intelligence (AI) Space
Cognitive /
ML/DL
Human Intelligence Exhibited by Machines
Machine Learning
“Trained” using large amounts of data &
ability to learn how to perform the task
New Data
Sources:
NoSQL,
Hadoop &
Analytics
Deep Learning
Break tasks into Artificial
Neural Networks
New class of applications
Opportunity in ML Training
 Pattern matching
 Image matching (Consumer Photos)
 Real-time decision support
 Complex workflows
 Data Lakes
Extend Enterprise applications
 Finance: Fraud detection /
prevention
 Retail: shopping advisors
 Healthcare: Diagnostics and
treatment
 Supply chain and logistics
Extend Predictive Analytics to
Advance Analytics with AI
Growing across Compute, Middleware, and Storage
© 2017 IBM Corporation
DEEP LEARNING
is the NEW Computing Model for Machine Learning
10s of billions of image, voice, video
queries per day
GPU inference for fast response,
maximize datacenter throughput
DATACENTER INFERENCING
Billions of Trillions of Operations
GPU train larger models, accelerate
time to market
Training
Datacenter
TRAINING
Billions of intelligent devices
& machines
Recognition, reasoning, problem solving
Device
Inference: real-time
accurate response
DEVICE INFERENCING
IBM Systems
DEEP LEARNING IS
SWEEPING ACROSS INDUSTRIES
Internet Services
Media &
Entertainment
Medicine
Security & Defense
Autonomous Machines

Image/Video classification

Cancer cell detection

Video captioning

Face recognition

Pedestrian detection

Speech recognition

Diabetic grading

Content based search

Video surveillance

Lane tracking

Natural language processing

Drug discovery

Real time translation

Cyber security

Recognize traffic signs
IBM Systems
DATA & ANALYTICS USE CASES
$
AUTOMOTIVE
COMMUNICATIONS
CONSUMER PACKAGED GOODS
FINANCIAL SERVICES
EDUCATION & RESEARCH
Auto sensors reporting
location, problems
Location-based advertising
Sentiment analysis of
what’s hot, problems
Risk & portfolio analysis
New products
Experiment sensor analysis
ON-LINE SERVICES /
SOCIAL MEDIA
HEALTH CARE
UTILITIES
LAW ENFORCEMENT
& DEFENSE
HIGH TECHNOLOGY /
INDUSTRIAL MFG.
LIFE SCIENCES
MEDIA/ENTERTAINMENT
Clinical trials
Viewers / advertising
effectiveness
Mfg. quality
Warranty analysis
OIL & GAS
RETAIL
Drilling exploration sensor
analysis
Consumer sentiment
TRAVEL &
TRANSPORTATION
Sensor analysis for
optimal traffic flows
People & career matching
Smart Meter analysis
for network capacity,
Patient sensors,
monitoring, EHRs
Threat analysis - social media
monitoring, photo analysis
IBM Systems
Introducing PowerAI:
Enterprise Deep Learning Distribution
Package of Pre-Compiled
Major Deep Learning
Frameworks
Easy to install & get started
with Deep Learning with
Enterprise-Class Support
Optimized for Performance
To Take Advantage of
NVLink
Enabled by High Performance Computing Infrastructure
IBM Systems
PowerAI Deep Learning Software Distribution
Deep Learning
Frameworks
Supporting
Libraries
Accelerated
Servers and
Infrastructure
for Scaling
Caffe
NVCaffe
IBMCaffe
Torch
TensorFlow
DL4J
Theano
Chainer
OpenBLAS
Bazel
Cluster of NVLink
Servers
Distributed
Frameworks
NCCL
Spectrum Scale:
High-Speed Parallel
File System
DIGITS
Scale to
Cloud
Coming
Soon
IBM Systems
DEEP LEARNING SOFTWARE
developer.nvidia.com/deep-learning
IBM Systems
DEEP LEARNING INSTITUTE
For Everyone
For Developers, Data Scientists, Researchers
INTRO MATERIALS
ON-SITE WORKSHOPS
SELF-PACED LABS
CASE STUDIES
PARTNER COURSES
TECHNICAL BLOGS
12
Weather Forecasting
HPC Touches Our Everyday Lives
HPC is a critical element of product development,
research, discovery and breakthrough science
CarRace
Design
Car Design / Real-Time
Real-Time
AnalyticsAnalytics
Aerospace
Design
Electronics Design
Risk Analytics
Turbine
Design
Reservoir
Simulation
DNA
Sequencing
IBM Systems
Main Components of HPC Systems
Compute Nodes
Storage
Support
Management Nodes Login Nodes
(Administrators)
(End users)
Management Network
High Performance Interconnect
IBM Systems
Power Systems HPC Cluster: Integrated, Modular, Validated
Processors &
Systems
High Speed
Interconnect
•
•
•
High Performance Processors & Systems
Accelerator, networking, storage integration via CAPI & NVLink
Highest memory throughput
•
High speed interconnect / network fabric from Mellanox
Technologies
MPI acceleration in the IB fabric, reducing CPU overhead
Support for GPU Direct, NVMe over fabric
•
•
High
Performance
File System &
Storage
• Highest Performance HPC Storage: Elastic Storage Server
• High Performance Spectrum Scale (GPFS) Parallel File System
• Data centric design
HPC
Software
• Deployment tools, integrated management
• Compilers: gcc, IBM XLC, LLVM OpenMP4, PGI Fortran/C/C++,
Java, OpenACC, OpenMP
• Debuggers, Profilers, Math libraries, MPI & HPC apps
IBM Systems
Fat and Flat Systems for Data - S822LC for HPC
Designed for Programmabilty
Infused with OpenPOWER Ecosystem
InfiniBand Fabric
DDR4
115GB/S
CPU
CPU
115GB/S
DDR4
NVLink
Tesla
P100
•
•
•
•
80GB/S
Tesla
P100
Tesla
P100
80GB/S
Tesla
P100
2.5X the CPU:GPU Interface Bandwidth
Tight coupling: strong CPU: strong GPU performance
Equalizing access to memory - for all kinds of programming
Closer programming to the CPU paradigm
IBM Systems
Why it matters: Stop waiting for Data!
Improve Code Performance for Developers
65% reduction in data transfer time in
for Kinetica GPU-accelerated DB
•
•
•
Less data-induced latency in all applications
Unique to POWER8 with NVLink
Less coding to compensate for slow data
movement!
1.95X of the 2.5X overall performance
improvement attributable to NVLink
•
•
•
100 tick Query Time: Competing System PCI-E x16 3.0
Calculation*
Data Transfer
73 ticks
27 ticks
65% Reduction
Data Transfer
26 ticks
Calculation*
14 ticks
40 tick Query Time: S822LC for HPC, NVLink
* Includes non-overlapping: CPU, GPU, and idle times.
All results are based on running Kinetica “Filter by geographic area” queries on data set of 280 million simulated Tweets with 5 up to 80 simultaneous query streams each with 0 think time.
Power System S822LC for HPC; 20 cores (2 x 10c chips) / 160 threads, POWER8 with NVLink; 2.86 GHz, 1024 GB memory, 2x 6Gb SSDs, 2-port 10 GbEth, 4x Tesla P100 GPU; Ubuntu 16.04.
Competitive stack: 2x Xeon E5-2640 v4; 20 cores (2 x 10c chips) / 40 threads; Intel Xeon E5-2640 v4; 2.4 GHz; 512GB memory 2x 6Gb SSDs, 2-port 10 GbEth, 4xTesla K80 GPU, Ubuntu 16.04.
IBM Systems
POWER as infrastructure accelerator for HPC workloads
High Performance Computing
High Performance Analytics
‘New Workloads’
(Design/Simulation/Modeling)
(Risk Analytics)
(Hadoop, Spark, Containers)
Workload
Engines
Common
Resource
IBM Spectrum Computing
Management
Heterogeneous Infrastructure Support / IBM ESS & Spectrum Scale
X86-64
Flash
Tape
Disk
Power
x86
Linux on z
Sparc
ARM
docker
VM
On-premise, On-cloud, Hybrid Infrastructure
IBM Systems
IBM POWER HPC Platform Strategy
• High-performance computer and high-performance analytics drive common platform design
• Servers will be predominately 2-socket designs
• Developing deeper relationships with technology partners – ref OpenPOWER
• Majority of floating-point performance will come from GPUs
• Utilize Industry-standard compliant 19” racks and electronics enclosures
• Air and water cooling options
• Accelerated platforms will be based on a common enclosure form factor
• Enclosure provides working envelope that we will continue to enhance with the latest
technology from IBM, NVIDIA, Mellanox and other OpenPOWER partners
• Enclosure provides a platform with sufficient power, cooling capability to support these
enhancements
IBM Systems
Processor Technology Roadmap
Continued Investment in POWER
695mm2
659mm2
650mm2
POWER11
POWER10
POWER9
POWER8
22 nm






12 Cores
SMT8
2X DPFP
PCIE Gen 3
Coprocessor (CAPI)
Enhanced Prefetch
2014
POWER8
w/ NVLink
14 nm
22 nm
 NVLink 1.0
 2X CAPI
2016







24 Cores
New µArchitecture
Direct-attach DDR4
Gen4 PCIe
CAPI 2.0
OpenCAPI 3.0
NVLink 2.0
2017





48 Cores
New µArchitecture
Enhanced Memory
OpenCAPI 4.0
Future NVLink
2020+





>48 Cores
New µArchitecture
2x SIMD width
Future NVLINK
Future OpenCAPI
Future
IBM Systems
Performance Leadership Roadmap for HPC / HPDA
Mellanox
Interconnect
Technology
Connect-IB
FDR Infiniband
PCIe Gen3
ConnectX-4
EDR Infiniband
CAPI over PCIe Gen3
ConnectX-5
Next-Gen Infiniband
Enhanced CAPI over PCIe Gen4
NVIDIA GPUs
Kepler
PCIe Gen3
Pascal
NVLink
Volta
NVLink Next Gen
POWER8
with NVLink
POWER9
IBM CPUs
PowerAccel Interfaces:
NVLink, CAPI,
PCIe Gen3
POWER8
OpenPower
CAPI Interface
2015
PowerAccel:
Enhanced CAPI,
NVLink 2.0, PCIe
Gen4
2016
2017
IBM Nodes
| 22
IBM Systems
Current Power Linux Servers – Announced September 2016
S822LC for High Performance
Computing
New POWER8
with NVLink
Processor
New Tesla P100
with NVLink
System Details
 2-socket, 2U
 Up to 20 cores (2.86-3.26Ghz)
 1 TB Memory (32 DIMMs)
 230GB/sec memory bandwidth
 2x SFF (HDD/SSD), SATA
 Up to 4 integrated NVIDIA Tesla P100 GPUs
3 PCIe slots, 3 CAPI enabled, IB Add-in
Air or water cooled
S822LC for Big Data
S821LC
Tesla K80; P100 coming
Tesla K80; P100 coming
System Details
2-socket, 2U
Up to 20 cores (2.9-3.3Ghz)
512 GB Memory (16 DIMMs)
115GB/sec memory bandwidth
12 SFF/LFF (HDD/SSD) 96 TB storage
 5 PCIe slots, 4 CAPI enabled
 2 NVIDIA PCIe GPU capable
System Details
 2 socket, 1U
 Up to 20 cores (2.09-2.32Ghz)
 512 GB Memory (16 DIMMs)
 115 GB/sec memory bandwidth
 4 SFF/LFF (HDD/SSD), 32 TB Storage
 4 PCIe slots, 3 CAPI enabled
 1 NVIDIA PCIe GPU capable
IBM Systems
Built with Collaborative Innovation
OpenPOWER
Open Source Workloads
290+ OpenPOWER members contribute to 87
OpenPOWER ready products and 17 servers delivering
choice to industry
Now hyper-focused on expanding Cognitive/AI industry
applications
Enterprise Support/Subscription model
Close partnership with major AI/accelerator industry leader
Nvidia
OpenCAPI
Open Frameworks
High bandwidth open interconnect to attach to
accelerators and SCM
Highly optimized & accelerated Cognitive/AI frameworks
Cognitive/AI SDK for deployment and deployment tools
Making machine learning and AI more affordable
IBM Systems
OpenCAPI 3.0
What is OpenCAPI?
•
OpenCAPI is an Open Interface Architecture that allows any microprocessor to attach to
• Coherent user-level accelerators and I/O devices
• Advanced memories accessible via read/write or user-level DMA semantics
• Agnostic to processor architecture
Key Attributes of OpenCAPI
•
•
•
High-bandwidth, low latency interface optimized to enable streamlined implementation of attached
devices
• 25Gbit/sec signaling and protocol built to enable very low latency interface on CPU and
attached device
Attached devices operate natively within an application’s user space and coherently with
processors
• Allows attached device to fully participate in application without kernel involvement/overhead
Supports a wide range of use cases and access semantics
• Hardware accelerators
• High-performance I/O devices
• Advanced memories
IBM Systems
IBM Application Development Ecosystem
IBM Tools form a key part of an open source and proprietary software ecosystem
POSIX
Programming Models
Compilers
IBM XL
Libraries
PETSc
Development Tools
IBM Parallel Perf Toolkit
Debuggers
Other Tools
Global Arrays
Threads
ScaLAPACK
IBM ESSL
IBM PowerLinux SDK
TotalView for HPC
HPCToolkit
Open|SpeedShop™
IBM Systems
XL C/C++ & Fortran Compilers for OpenPOWER
Key Features:
• Highly optimized for POWER processors
• XL C/C++ supports C11, C++11, subset of C++14
• XL Fortran supports Fortran 2003 and majority of Fortran 2008 features
GPU exploitation with XL C/C++ V13.1.5 & XL Fortran V13.1.5:
• CUDA C/C++: XL C/C++ works as host compiler for POWER
• CUDA Fortran: XL Fortran supports CUDA Fortran extensions
• OpenMP 4.5: subset features to enable GPU offloading
•
•
http://www-03.ibm.com/software/products/en/xlcpp-linux
http://www-03.ibm.com/software/products/en/xlfortran-linux
IBM Systems
2500+ Linux Applications on POWER
HPC
CHARMM
GROMACS
NAMD
AMBER
RTM
GAMESS
WRF
HYCOM
HOMME
LES
MiniGhost
AMG2013
OpenFOAM
Cloud
Big Data & Machine
Learning
Mobile Enterprise
miniDFT
CTH
BLAST
Bowtie
BWA
FASTA
HMMER
GATK
SOAP3
STAC-A2
SHOC
Graph500
Ilog
100,000+ Packages in Linux Distributions on Power
Major Linux Distros
IBM Systems
IBM HPC Applications Available, in Process, or in Discussion
HPC Applications
EDA
CAE
Crash, NVH,
Structures
Synopsys
CFD
LS-Dyna
Alya Nastin
ANSYS
Mechanical
AMGX
Cadence
+
Genomics
FASTA / S-M
ABySS
FastQC
ALLPATH_LG
Mentor Graphics
AVBP
AVUS
Code_Saturne
FASTX-Toolkit
Analytics
+
BALSA
FreeBayes
bamkit
GATK
BarraCUDA
Galaxy
bcftools
Genome
Analysis Platform
Culises
bedops
Fluent
bedtools
HOOMD_Blue
LBM D2Q37
BioPython
HTSeq
=
OpenFOAM +
MAFIA
CPMD
NEST
CP2k
Neural Tissue
Sim
DL-POLY
PLINK2
htslib
BLAST
IGV
QIME
BLAST-NCB
ISAAC3
Boost
kraken
LoFreq
+
RAPTOR/Raptor
X
QMCPACK
R-EBSeq
HOOMD-blue
+
VMD
+
SOAPDenovo
Churchill
Mothur
Raptor/RaptorX
Spades
cimerascan
MrBayes
RAxML
SplazerS
MUCLE
RSEM
SQLite
conda
MuTect
RNA-star
Sra-tools
Cufflinks 2.1
nose
Sailfish
STAR-fusion
Varient_tools
cutadapt
numpy
salmon
Velvet / Oasis
Databiology
STAR (RMAstar)
NVBIO
samblaster
T-Coffee
Virdatint
DELLY2
NVBowtie
SamTools
Tabix
Zlib
PairHMM
Scalpel
pandas
scikit-bio
ELSA
+
Perl
seqtk
EMBOSS
PHYLIP
setuptools
ESP
Picard
SHRiMP
TopHat
Trimmomatic
peasoup
ACME
bidmach
Aladin
Caffe
AROME
Chainer
Murex MACS
FWI
+
heat3d
CAM-SE
CNTK
Intersect Res
Modeling
Cosmo SVN
cuDNN
Kirchhoff Time
Migration
HYCOM
NWCHEM
IFS
P-Gadget
RTM kernel
=
Quantum Expresso
+
LES
TensorFlow
RTM - Total
Meso-NH
Theano
SeisSol
MG2
Kripke
SPECFEM
MPAS-A
LSDALTON
SPECFEM3D
NEMO
Pluto
RTM – PGS +
heat3d
Hydra
KKRnano
LSMS
Torch
PopPerf
LSQR
BQCD
Lattice QCD
Liblinear
LuaJIT
GYSELA
Octopus
Digits
Jurrassic
GTC
Quantum
ChromoDynamics
R tidyverse
DL_MESO
MOLPRO
VASP
Tassel
Algorithmics
RK - RM
GENE
snpEff
lumpy
diamond
GADGET
LULESH
snpSift
SOAPaligner
=
Machine
Intelligence
LAMMPS
NAMD
SOAP3-dp
FLASH
HACC
Linda
Weather &
Climate
FD - Total
CTH
DIRAC
R cowplot
CLARITY (LIMS)
GAMESS
B-CALM
Gaussian09
+
GROMACS
QMCPACK
Primer3
BioVelocity
BWA
280+ applications
ported or in
process since Fall
2014
DL_MESO
pysam
Bowtie, Bowtie2
CoMD
Virdatint
+
ALE3D
Financial
Modeling
EchelonI
Astrophysics
Chroma
Nextbio
HMMER
bfast
CHARMM
CoreNeuron
Computational
Physics
Casino
-
Amber 14
CNEST
IO Informatics
tranSMART
Computational
Chemistry
Molecular
Dynamics
ADF
Elservier
Pindelg
Bioconductor
Neural &
Other
ACUMI
Genomon Fisher
LatticeBoltzmann
Ludwig
Translational
Med
Geo-sciences
Chemistry & Physics
Biosciences
+
Lulesh
regCM
MCB
WRF
miniDFT
MILC
nekbone
OpenQCD
NUCCOR
QUDA
PIConGPU
In Discussion
QBOX
Plan to Support / In Process
QMCPack
Available on Power
SNAP
Performance Benchmarks Exist (+, =, -)
UMT2013
Trinity
XGC-1
IBM Systems
IBM PowerLinux SDK 1.9 Features
Software development toolkit for application development on Power Systems
Standard Eclipse Integrated Development
Environment
Enhanced PowerLinux tools
 Migration Advisor

Built on Eclipse Mars
 Source Code Advisor

Extensible via plugins
 Trace Analyzer

Common look & feel across tools
 POWER8 CPI Stack Model

Integrated help, accessibility, usability features
 POWER8 Functional Simulator
 Power Linux community message board tool
Additional Eclipse.org plugins
 IBM Advanced toolchain

C/C++ Development Tools (CDT)

LinuxTools

Parallel Tools Platform (for remote service access)
Full support for POWER8

POWER8 CPI

OProfile and perf event profiling
 Feedback Directed Program Restructuring (FDPR)
 Pthread Monitor trace tool
 Build Advisor
 Remote dependencies package
https://developer.ibm.com/linuxonpower/tutorials/sdk_linux_on_power/
https://developer.ibm.com/linuxonpower/sdk/
IBM Systems
Here to help you - Worldwide HPC Centers and Technical Support
•
Power Acceleration & Design Centers (PADC) with IBM, NVIDIA and Mellanox focused on accelerated
applications and technical collaborations
•
IBM Systems Client Centers
UK Science and Technology Facilities Council
(STFC) PADC to optimize Performance of
OpenPOWER systems for Modelling Simulation
and Big Data Analytics
 HPC Briefings
 HPC Workshops
 HPC Benchmarks
IBM Poughkeepsie Client
Center and POWER HPC
Benchmark Center
IBM Austin POWER
HPC Executive
Briefing Center
IBM Boeblingen PADC joint center with
NVIDIA for technical collaborations,
consulting, application enablement, system
external access, …
IBM Montpellier Client Center and PADC
joint center with NVIDIA and Mellanox for
technical collaboration, consulting,
application enablement, benchmarking, . . .
NVIDIA/IBM
Acceleration Lab
http://www-03.ibm.com/systems/services/clientcenters/
New! NIMBIX Cloud
For instantaneous
development and
commercial deployment.
https://power.jarvice.com/landing
[email protected]
IBM Systems
Resources and Support for Linux Developers
IBM PartnerWorldTechnical Support
IBM DeveloperWorks
• IBM Innovation Centers
• Free access to Power Hardware
• Free porting assistance
• Free Eclipse-based development environment
Technical resources, community, blogs,
toolkits, How to articles, beta code
www.ibm.com/developerworks/linux/
www.ibm.com/partnerworld/wps/servlet/ContentHandler/pw_com_pwp_partnerworldprogram
IBM Innovation Centers
IBM Migration Factory
§All 50+ centers worldwide now support Linux on Power
§One-stop for ISVs, developers
§HW access, technical support, demos, toolkits, Hands-on labs
www.ibm.com/systems/power/software/linux/centers
Premier migration services for large applications
http://www03.ibm.com/systems/services/labservices/migrati
onfactory
IBM Power Development Cloud
Site Ox
Provide free access to Power hardware to ISVs for
Porting
On-demand cloud-based development
platform using Linux on POWER8
www.ibm.com/partnerworld/wps/servlet/ContentHandler/stg_com_sys_powerdevelopment-Platform
www.siteox.com
IBM Systems
QUESTIONS
IBM Systems