2. Data
Time
Available
Data
Understood
Data
Enterprise
Amnesia
80 million
wearable health
devices will
be available by
2017.
2.5
quintillion
bytes of data
generated daily
by connected
machines.
There
will be
28 times
more
sensor-
enabled
devices
than
people
by the
year 2020.
25 gigabytes
of data per hour
is generated by a
connected car.
90% of cars will
be connected by 2020.
153 exabytes
of healthcare
data generated by
devices in 2013.
Increasing to 2,314
exabytes in 2020.
1.7 megabytes
of data per
second
generated by
every human
being on the
planet by 2020.
2
5. Assembled the 1.2 billion letter genome
(faster and cheaper than ever before)
to understand its vulnerabilities
Culex quinquefasciatus
The Southern House Mosquito
5
10. As neural networks go deeper,
they provide
a dramatic increase
in accuracy.
Higher accuracy
networks require
way higher
computation
which increases
prediction latency.
10
11. When scale-out is not enough…
Deep Learning model training is not easy to distribute
Training can take hours,
days or weeks with large
data-sets
Real-time analytics possible with:
Unprecedented demand for offloaded computation,
accelerators, and higher memory bandwidth systems
Resulting in….
Moore’s law is dying
11
12. OpenPOWER: Open Hardware for High Performance
1
2
Systems designed for
big data analytics
and superior cloud economics
Upto:
12 cores per cpu
96 hardware threads per cpu
§1 TB RAM
7.6Tb/s combined I/O Bandwidth
GPUs and FPGAs coming…
OpenPOWER
Traditional
Intel x86
http://www.softlayer.com/POWER-SERVERS
https://mc.jarvice.com/
13. Why IBM's shrinking transistors look
like a breakthrough for all of IT…
Faster – Lower Power - Smaller
5 NM 50% more switches than 7nm
1st ever 5nm transistor structure
(nanosheet)
40% more throughout @ fixed power …or
75% power savings at same throughput
13
14. IBM PowerAI
the accelerated
platform for
deep learning
dramatically improved training times
The LARGER the problem …
the BIGGER the NVLink advantage
4Xthreads/core
memory bandwidth
more cache
UNIQUE
CPU ßà NVLink ßà GPU
more
powerful
vs. x86 +
14
16. 16
Large Model SupportDistributed Deep Learning
faster training times
for data scientists
(Competitors)
Limited memory on GPU forces
trade-off in model size / data
resolution
POWER
CPU
DDR4
GPU
NVLink
Graphics
Memory
(PowerAI)
Use system memory and GPU
to support more complex models
and higher resolution data
Traditional Model Support à
CPUDDR4
GPU
PCIe
Graphics
Memory
Performance…
Faster Training
and Inferencing
95% scaling efficiency on the Caffe deep learning
framework over 256 NVIDIA GPUs in 64 systems
IBM Research achieved a new image recognition accuracy of 33.8% for a
neural network trained on a very large data set (7.5M images). The
previous record published by demonstrated 29.8% accuracy.
https://www.ibm.com/blogs/research/2017/08/distributed-deep-learning/
19. Tools for Ease
of Development
rich advisory and building
toolsets to flatten
time to value
AI Vision
rich toolset image
recognition neural
networks
automated deep learning
toolkit data preparation
DL Insight toolkit supports
auto-training runs for
hyper parameter tuning
+++
19
20. In my dreams
I’m coding in
an open data science
framework,
running on Spark and
Power
…in minutes
IBM Data Science Experience
Learn
Create
Collaborate
Tools for Ease
of Development
20
21. Accelerators/GPUs in a Cloud Stack
21
Containers
and images
Accelerators
Clustering frameworks
Workload
Aware
Scheduling
Shared
Resource
Management
Emerging
Workloads
Dev Ops & Micro Services
High Performance
Computing
Design / Simulation / Modeling
‘New-gen
Workloads’
Hadoop, Spark, Containers
with Spark
IBM
Cloud
private
New
High Performance
Analytics
Trade / Risk Analytics
IBM Data
Science
Experience
Deep Learning Training & Inference
23. Build Deep Learning Docker Images Using PowerAI Software
23
Dockerfile extending Nvidia base images for POWER:
FROM nvidia/cuda-ppc64le:8.0-cudnn6-devel-ubuntu16.04
ENV POWERAI_REPO mldl-repo-local_4.0.0_ppc64el.deb
RUN apt-get update && apt-get install -y git wget ssh vim curl &&
apt-get clean
# import PowerAI repo
RUN cd /tmp && wget
https://public.dhe.ibm.com/software/server/POWER/Linux/mldl/ubu
ntu/${POWERAI_REPO} && dpkg -i ${POWERAI_REPO} && rm
${POWERAI_REPO}
# install PowerAI
RUN apt-get update && apt-get install -y power-mldl && apt-get
clean
IBM
Cloud
private
See example: https://github.com/knm3000/nvidia-powerai/blob/master/Dockerfile
24. Run PowerAI software with NVIDIA Docker
24
A Docker wrapper and tool
to package and GPU based
apps
Enhance portability of
images by using drivers on
the host
No need to include drivers
in Docker image
See Blog:
https://developer.ibm.com/linuxonpo
wer/tutorials/powerai-docker-images/
https://github.com/NVIDIA/nvidia-docker/tree/ppc64le
25. Manage GPU clusters with Kubernetes and IBM Cloud private
• Open source orchestration system
for Docker containers on multiple
hosts: https://kubernetes.io/
• GPU scheduling features getting
upstream
• IBM Cloud private, :
Download free Community Edition,
Ask on slack
• Download RPM from
https://www.rpmfind.net/linux/rpm2html/search.php?q
uery=kubernetes&arch=ppc64le
| 25
http://on-demand.gputechconf.com/gtc/2017/presentation/s7258-seetharami-seelam-
speed-up-deep-learning-service.pdf
26. Show >100x speedup for Caffe inferencing with GPUs in PowerAI in
NIMBIX in less than 5 minutes
PowerAI Trial Configurations in a public cloud:
• Docker container builds and comes up in minutes
• Single P100 GPUs
• 30 days with 60 hrs standard (120 for Sales referral)
• 128GB RAM, 32 CPU threads, 1TB shared storage
• Quad P100 GPUs
• 30 days with 120hrs standard (more by request)
• 512GB RAM, 128 CPU threads, 1TB shared storage
Nimbix Cloud Advantages
• Easier to use
• Highest Performance
• Ultra Fast Launch Times
• Lower Cost
• Faster time to Value
• Bare-Metal Acceleration
• Enterprise Accounting
• Application Marketplace
• Private Apps
https://www.slideshare.net/IndrajitPoddar/fast-scalable-easy-machine-
learning-with-openpower-gpus-and-docker
Experience performance
with productivity
A superior integrated stack and
adequate hardware resources
for deep learning insights
https://www.nimbix.net/ibm-power-nimbix-cloud
26
27. 9Days
Acceleration training …. days become hours
4Hours
Recognition
Shape
Attenuation
Boundary
Recognition
Shape
Attenuation
Boundary
54x
Learning
runs with
Power 8
4Hours
4Hours
4Hours
4Hours
. . . . . . . .
. . . . . . .
4Hours
What will you do?
Iterate more and create more accurate models?
Create more models?
Both?
27
28. Developer Resources for POWER systems
• Linux on POWER Developer Portal
https://developer.ibm.com/linuxonpower/
• Find open source Linux packages in
popular OS distros
https://developer.ibm.com/linuxonpower/open-source-
pkgs/
• Request free VMs from Oregon State
University Open Source Lab:
http://osuosl.org/services/powerdev/
• Get answers to Linux specific questions
in Stack Overflow
https://developer.ibm.com/answers/smartspace/linuxo
npower/index.html
• See Blogs on Deep Learning and
PowerAI topics
https://developer.ibm.com/linuxonpower/blog/
| 28
30. Notices and Disclaimers Con’t.
30
Information concerning non-IBM products was obtained from the suppliers of those products, their published announcements or other publicly available sources.
IBM has not tested those products in connection with this publication and cannot confirm the accuracy of performance, compatibility or any other claims related to
non-IBM products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of those products. IBM does not warrant the quality of
any third-party products, or the ability of any such third-party products to interoperate with IBM’s products. IBM EXPRESSLY DISCLAIMS ALL WARRANTIES,
EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE.
The provision of the information contained h erein is not intended to, and does not, grant any right or license under any IBM patents, copyrights, trademarks or other
intellectual property right.
IBM, the IBM logo, ibm.com, Aspera®, Bluemix, Blueworks Live, CICS, Clearcase, Cognos®, DOORS®, Emptoris®, Enterprise Document Management System™,
FASP®, FileNet®, Global Business Services ®, Global Technology Services ®, IBM ExperienceOne™, IBM SmartCloud®, IBM Social Business®, Information on
Demand, ILOG, Maximo®, MQIntegrator®, MQSeries®, Netcool®, OMEGAMON, OpenPower, PureAnalytics™, PureApplication®, pureCluster™, PureCoverage®,
PureData®, PureExperience®, PureFlex®, pureQuery®, pureScale®, PureSystems®, QRadar®, Rational®, Rhapsody®, Smarter Commerce®, SoDA, SPSS,
Sterling Commerce®, StoredIQ, Tealeaf®, Tivoli®, Trusteer®, Unica®, urban{code}®, Watson, WebSphere®, Worklight®, X-Force® and System z® Z/OS, are
trademarks of International Business Machines Corporation, registered in many jurisdictions worldwide. Other product and service names might be trademarks of
IBM or other companies. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at: www.ibm.com/legal/copytrade.shtml.