HPC Advisory Council March 21, 2016

Calista Redmond President, OpenPOWER Foundation Director, IBM OpenPOWER Global Alliances credmond@us..com @Calista_Redmond

© 2016 OpenPOWER Foundation Data is the new basis of competitive value

Oil & Gas

Healthcare

Retail Banking

2 © 2016 OpenPOWER Foundation Data is growing exponentially and demands new system approaches (technology and strategy) 44 zettabytes

unstructured data You are here

structured data

2010 2020

3 © 2016 OpenPOWER Foundation Data on the Rise and is the New Basis of Competitive Value . The world is awash in data – sensor, modeling, social . Pools of dark data – collected but un-analyzed

Oil and Gas Genomics Financial Analysis

Personalized medicine •15 PBs of survey data • Next Gen Sequence data doubling •10+ months to process new survey • 4 TB/day every 5 months data once acquired • 10k to 100k databases • Some leading institutions @ •100x compute needed for deep water • Multi PB volume >100PB in 1-2 years imaging • 400M tasks per day • Projects increasingly National •Network and high bandwidth storage • <100ms response required scale (UK, China, UAE, US) intensive • Tight coupling of near real time • Large cluster and Large SMP •Long term storage needs analytics and trading systems to needs •Multi-TB data sets for visualization underlying data sets. • High performance file system and •1 TB per oilfield per day scalable storage •2 TB per rig per day •80% Dark Data* • Multi-stage workflow process for personalized treatment Firms of all sizes are experiencing data challenges as they focus on insight 4 © 2016 OpenPOWER Foundation From High Performance Computing to High Performance Insights Massive data requirements drive a composable architecture for big data, complex analytics, modeling and simulation. The DCS architecture will appeal to segments experiencing an explosion of data and the associated computational demands

Business Business Intelligence Analytics Technical Computing Oil and Gas

Financial Social Analytics Analytics Climate & Environment Science

Life Sciences

Cloud Meets Data Centric Cognitive Systems Big Data • Cloud • Analytics • Watson 5 © 2016 OpenPOWER Foundation OpenPOWER, a catalyst for Open Innovation

Market Shifts Moore’s law no longer Growing workload Numerous IT Mature Open satisfies performance gain demands consumption models software ecosystem

OpenPOWER Strategy Vibrant ecosystem through open Accelerated innovation through Amplified capabilities driving development collaboration of partners industry performance leadership

Industry adoption, Open choice

Cloud Computing High Performance Domestic Hyperscale & Large scale Computing & Analytics Datacenters IT Agendas

OpenPOWER is an open development community, using the POWER Architecture to serve the evolving needs of customers. 6 © 2016 OpenPOWER Foundation OpenPOWER collaboration with HPC Advisory Council

OpenPOWER is pleased to announce our membership in HPCAC • Furthering cross-community collaboration opportunities in HPC

Contribution of OpenPOWER systems to HPCAC lab for benchmarking and demonstration • 4 systems 2x 10c with 256GB memory with dual K80 GPU acceleration

7 © 2016 OpenPOWER Foundation 8 © 2016 OpenPOWER Foundation OpenPOWER Development Community

9 © 2016 OpenPOWER Foundation Fueling the OpenPOWER Community

10 © 2016 OpenPOWER Foundation OpenPOWER Foundation reach

190+ members 20+ technologies revealed

24 countries 6 continents 100+ innovations under way

11 © 2016 OpenPOWER Foundation So what? What difference does it make?

12 © 2016 OpenPOWER Foundation POWER8: Processor Performance Leadership & Accelerator Interfaces

Faster Cores Larger Caches 3x Higher Memory 8 Threads Per Core Direct Accelerator Bandwidth, 1 TB Interconnect Memory per Socket

POWER8 12 Cores, 96 Threads

DRAM Memory 4 Level Large Caches Chips Buffer Up to 1 TB per socket Up to 230 GB/s sustained 13 © 2016 OpenPOWER Foundation Accelerates Technology roadmap

Mellanox Interconnect Connect-IB ConnectX-4 ConnectX-5 FDR Infiniband EDR Infiniband Next-Gen Infiniband PCIe Gen3 CAPI over PCIe Gen3 Enhanced CAPI over PCIe Gen4

NVIDIA GPUs Kepler Pascal Volta PCIe Gen3 NVLink Enhanced NVLink

POWER8 POWER8 with NVLink POWER9 IBM CPUs OpenPower Enhanced CAPI Interface CAPI & NVLink

2015 2016 2017

IBM Systems

14 © 2016 OpenPOWER Foundation Application Benchmarks

15 © 2016 OpenPOWER Foundation Portfolio of HPC Solutions

• High Performance Processors & Systems Processors & • Accelerator, networking, storage integration via CAPI & NVLink Systems • Innovative solutions like CAPI Flash

• Platform LSF & Symphony workflow and resource management HPC • Compilers: gcc, IBM XLC, PGI Fortran/C/C++, Java, OpenACC, Software OpenMP • Debuggers, Profilers, Math libraries, MPI & HPC apps

High Performance • Highest Performance HPC Storage: Elastic Storage Server File System & • High Performance Spectrum Scale (GPFS) Parallel File System Storage • Flash Storage

16 © 2016 OpenPOWER Foundation IBM Research Paving the Path to Next-Generation HPC

Data Centric System Scalable High Programming Enhancing Open Node, & Processor Performance Storage Models for Exascale Interconnects Innovations & File Systems

POWER

17 © 2016 OpenPOWER Foundation Cross community collaboration is essential

18 © 2016 OpenPOWER Foundation OpenPOWER and are teaming to deliver price performance advantage on Power

• Contributing OpenPOWER Open Compute reference design via collaboration between IBM and Rackspace.

• Contributing OpenPOWER from IBM to ensure fully open opportunity for OpenPOWER Open Compute system

• Engaging cross community opportunities via Open Compute role on advisory group to OpenPOWER Foundation along with joint activities

19 © 2016 OpenPOWER Foundation • Over 2,200 ISVs developing on Power Accelerates • 50 IBM Innovation Centers software • Compelling PoCs innovation • Support for little endian applications Big Data & Machine HPC Cloud Mobile Enterprise Learning CHARMM miniDFT GROMACS CTH NAMD BLAST AMBER Bowtie RTM BWA GAMESS FASTA WRF HMMER HYCOM GATK HOMME SOAP3 LES STAC-A2 MiniGhost SHOC AMG2013 Graph500 OpenFOAM Ilog Major Linux Distros

20 © 2016 OpenPOWER Foundation Superior Cloud economics, powering Hyperscale datacenters

SoftLayer Bare Metal POWER cloud

Rackspace joins OpenPOWER, declares Open Compute OpenPOWER system

Docker, OpenStack HEAT, and Chef Server for Power Open Source Docker

Bluemix Cloud Integration and SQLDB Services with Power Systems for SOE/SOR integration

Power Cloud environments in Europe

IBM Watson Developer’s Cloud for cognitive computing www.ibm.com/smarterplanet/us/en/ibmwatson/developers.html

Power Dev Cloud free POWER access to ISVs for Porting hwww.ibm.com/partnerworld/wps/servlet/ContentHandler/stg_com_sys_power-development-platform

21 © 2016 OpenPOWER Foundation Accelerates Infrastructure Standards

11 Work Groups 2014 2015 2016 Personalized Medicine Charter Pers Med Integrated Solutions Integrated Solutions Charter CAPI Linux SDK System SW Charter 64b ABI Platform Ref

25g IO Compatibility Charter 25g IO Spec

OpenPOWER I/O Charter Sys I/O Enablement Guide CAPI AFU Intf Spec V1 CAPI AFU Intf Spec V2 Accelerator Charter OpenCL SDK

Memory Charter OPMB Intf. Spec V1

FSI Specification Charter FSI Spec

P8 SP010 P8 2U2S P8+ 1U1S P8+ 2U2S Charter Developer Platform Data Reference Reference Reference OpenPOWER ISA Profile V1 OpenPOWER ISA Profile V2 Charter IO Device Architecture V2 IO Device Architecture V3 HW Architecture Coherent Accel Intf Arch Coherent Accel Intf Arch

Compliance Specification Comp Compliance Charter Draft Review WG Spec STD

SP010 – Tyan OpenPOWER Customer Reference System AFU – Accelerator Function Unit OPMB – OpenPOWER Memory Bus SDK – Software Developer Kit CAPI – Coherent Accelerator Processor Interface FSI – Field Replaceable Unit (FRU) Service Interface ABI – Application Binary Interface 22 © 2016 OpenPOWER Foundation Engage in the OpenPOWER community

Technology and Software Innovators, Integrators,

Innovators and Partners End Users INTEREST

Strategic imperatives and Discuss areas of collaboration Deep understanding of workload workload demands for and synergy in OpenPOWER. demands and consumption performance and cost preferences of end users. optimization. Sign up for membership and

ACTION join a work group. Team with innovators and end Require open software and users to drive requirements, systems with choice and Build technical and business engage in specific projects. flexibility. relationships. Engage directly on system design options.

Develop Deliver Deploy collaborative innovations with collaborative innovations with collaborative innovations with OUTCOME compelling value compelling value compelling value

23 © 2016 OpenPOWER Foundation Thank you.

Calista Redmond President, OpenPOWER Foundation Director, IBM OpenPOWER Global Alliances

[email protected] @Calista_Redmond 24 © 2016 OpenPOWER Foundation People Can't Stop Talking About OpenPOWER

According to results shared at the STAC Summit on June 4, an IBM Both the current results and future potential are so promising that we are Power8-based system server delivered more than twice the preparing to build an OpenPOWER-based, Open Compute platform. “performance of the best-in-class x86 counterpart when running a set “And it will run OpenStack services. - Aaron Sullivan, Senior Director & of standard financial industry benchmarks. - HPC Wire Distinguished Engineer, Rackspace

We estimate that a physical Power Systems server can support “People ask me if we would switch to Power, and the answer is twice as many virtual environments – and therefore double the absolutely,” Urs Hölzle, , said emphatically and unequivocally. “number of client systems – as a physical x86 server. - ““Even for a single generation.” - The Platform

The architectures of the last fifteen or more years are clearly not going to take us past the next several years, so there needs to be a The $320 million deal [U.S. Department of Energy's Sierra and Summit “step change and there are lots of options but for the workloads we’re supercomputers] is a validation of sorts of IBM’s recent strategic shift to open looking at ahead, we are confident about this architecture. - Cliff “up its POWER computing architecture for Brereton, STFC license and use by other companies looking for alternatives to an Intel-controlled world. - Forbes

While other vendors are struggling to establish themselves in China, I still see the POWER8 processor as one of the most IBM seems to have found the secret with OpenPOWER. Not only powerful processors in the industry – and with the “does this seem to be cementing IBM inside the tough Chinese “ accelerators being developed in the OpenPOWER technology market but the benefits, such as SuperVessel, are now ecosystem, it is going to be hard for other processors starting to be felt (including Intel and ARM silicon) to match some of the across the whole of IBM. acceleration speeds that will be achieved by POWER8 - Enterprise Times and successive generations. - Clabby Analytics

25 © 2016 OpenPOWER Foundation The IBM Power Systems Linux Portfolio Enterprise & IFLs PurePower • Enterprise level • Converged robustness and IFL The L Line infrastructure offering capability • Enterprise level RAS The LC Line • Rapid time to value • Solution editions for in for single system memory databases deployments and simplicity of • Design and TCA cost management • (HANA, DB2 BLU) optimized for cognitive workloads on scale out • Solutions for Big Data running deployments (cloud & Analytics and cluster) • Hosted cloud and hybrid cloud solutions • Solutions for cognitive, Community / 3rd HPC and Big Data IBM Support Party Support • Rapid deployments and POCs Supported by Canonical

26 © 2016 OpenPOWER Foundation OpenPOWER expands POWER opportunity

Customize and optimize infrastructure Developer resources for custom tuning for most compelling business value • SuperVessel cloud system access with • FPGA acceleration, leveraging open CAPI acceleration options in US and China processor connect • Linux software porting and optimization • GPU acceleration, will leverage NVLINK • OpenPOWER Foundation work groups and processor attach in 2016 technical resources • Advanced memory options from Samsung, Micron, SK Hynix Linux on Power dedicated ISV support • Leadership interconnect from Mellanox • Over 1,900 Linux ISVs developing on Power • 50 IBM Innovation Centers Cloud options expand adoption • Compelling Proof-of-Concepts • IBM Softlayer • Support for little endian applications • OVH and Online in Europe • Rackspace Open Compute coming soon

On-going engagements with nearly every major hyperscale client in the U.S. and China

27 © 2016 OpenPOWER Foundation 8x faster than x86 Ivy Bridge on pattern extraction

10x faster text search than CPU only

28 © 2016 OpenPOWER Foundation GPUdb in-memory Accelerated Database Ultrafast ingest and analysis of billions of objects

• In-memory distributed database using GPUs for processing

• Order of magnitude gains: • Performance over CPU solutions • Power reduction • Cost savings

• Anticipate order of magnitude gains with POWER based NVLink solutions in 2016

29 © 2016 OpenPOWER Foundation 10x higher throughput, 10x lower latency accelerating NoSQL workloads

Turbo LAMP Stack for mobile and web apps

(Ubuntu) Orchestration PHP (Zend)

- PHP (Zend) Juju Juju

Apache Web Server ariaDB (MariaDB) (Ubuntu) M PHP (Zend)Linux (Ubuntu)

High Speed Data Transfer (Mellanox) Turbo LAMP Turbo

Future Dramatically less infrastructure Dramatically faster responsiveness to customers 30 © 2016 OpenPOWER Foundation Storage Enablement for OpenPOWER

2-3x Port Density with 50% Lower 10x Performance over typical SATA Power SSD’s

Series 7 & 8 – RAID/HBA Portfolio Ecosystem of Industry Solutions Industry’s only 16-24 port • Density native SAS/SATA adapters

• Flexibility HBA, Caching HBA, or RAID PMC Flashtec Samsung SSD Memblaze SSD • Performance Leading Throughput & IOPS NVRAM 2.5” PCIe Card Up to 10M IOPS 700K IOPS Up to 8TB

31 © 2016 OpenPOWER Foundation Up to 36x performance improvement for Key-Value Store acceleration

• Accelerated appliance with novel data-flow implementation of Memcached on FPGA. • Up to 36x improved performance and power response times in microsecond range. • CAPI integration of memory allows both host memory and coherent- attached flash to be used as value store.

32 © 2016 OpenPOWER Foundation Data Engine for NoSQL with 40TB CAPI-attached flash 24:1 3x server consolidation lower cost per user

Before: NoSQL in memory (x86) After: NoSQL POWER8 + CAPI Flash

WWW WWW

10Gb Uplink 10Gb Uplink

POWER8 Server Load Balancer 4U 24U Flash Array w/ up 500GB Cache to 40TB 500GBNode Cache 500GBNode Cache 500GBNode Cache 500GBNode Cache Node

Backup Nodes Less is More 24:1 physical server consolidation = 6x less rack space

33 © 2016 OpenPOWER Foundation 8x performance improvement GPU acceleration for Java on segmentation using accelerated machine learning for clustering with Hadoop / Mahout

Best-in-class ingredients • IBM POWER8 • IBM Java • NVIDIA CUDA GPU acceleration • Ubuntu Little Endian Linux for POWER

34 © 2016 OpenPOWER Foundation Critical workloads run on Linux on Power

Web, Java Apps and Infrastructure Analytics & Research

HPC applications for Life Sciences

• Highly threaded • Compute intensive • Throughput oriented • High memory bandwidth • Scale out capable • Floating point • High quality of service • High I/O rates

Business Applications Database

• High quality of service • Handle peak workloads • Scalability • Scalability • Flexible infrastructure • High quality of service • Large memory footprint • Resiliency and security

35 © 2016 OpenPOWER Foundation Solution stacks continuing to grow

• ISV community of 1,900+ • All major Linux distros • Open sourced POWER8 Cloud firmware stack Software Existing Standard Operating Open • Resources for porting Environment Source (System Mgmt) Software and optimizing: Communities Operating OpenPOWERFoundation.org System / KVM

New OSS Firmware OpenPOWER Firmware Community

Hardware OpenPOWER Technology

36 © 2016 OpenPOWER Foundation Industry trends drive innovation beyond the chip Microprocessors alone no longer drive sufficient Cost/Performance improvements

System Stack Applications and Services

Systems Management & Processors Cloud Deployment Systems Acceleration & Semiconductor Technology HW/SW Optimization Firmware, Operating System and

Processors

Semiconductor Technology Use Cases

Workload Acceleration Services Delivery Model Advanced Memory Tech POWER8 Linux Network & I/O Accel NEW OpenPOWER

System stack innovations are required to drive Cost/Performance 37 © 2016 OpenPOWER Foundation Membership Options

Anyone may participate in OpenPOWER. Membership levels are designed for those that are investing to grow and enhance the OpenPOWER community and its proliferation within the industry.

. The OpenPOWER Foundation is a Not-for-profit entity with a Board of Directors and a Technical Steering Committee. • Membership levels provide either a default Board of Director position (Platinum) or an opportunity to be elected to the Board (Gold, Silver, and Assoc/Academic members). The Bylaws include additional governance detail. • Technical Steering Committee is formed from Work group Leads and Platinum members. . Membership options include Platinum, Gold, Silver, and Associate / Academic memberships • Annual fee and dedicated full-time equivalent (FTEs) - verification of FTEs on honor system • Contributors, committers, Work group leads and project leads influence Technical Steering Committee • Associate / Academic level is not available to corporations

Membership Annual Fee FTEs Technical Steering Committee Board / Voting position Level $ USD One seat per member not otherwise Includes board position Platinum $100k 10 represented Includes TSC position

May be on TSC if Gold members may elect one board Gold $60k 3 Work group lead representative per three gold members

$20k May be on TSC if Sliver members may elect one board Silver $5k if <300 0 Work group lead representative for all silver members employees Associate & $0 0 May be on TSC if May be elected to one community Academic Work group lead observer, non-voting Board seat

Membership agreement, Bylaws, and IP Rights Policy available for review www.openpowerfoundation.org 38 © 2016 OpenPOWER Foundation OpenPOWER Open Software and University Cloud Environments

Oregon State North America http://osuosl.org/services/powerdev

UNICAMP Brazil, SA http://openpower.ic.unicamp.br/mini Brno University / cloud/index.html • OpenPOWER Platforms RedHat. Czech Republic • Open Stack Software https://fit-rhlab.rhcloud.com • University research • Open Development & Ecosystem Support SuperVessel IIT Bombay, India USA and China 3Q, 2015 www.ptopenlab.com

HPC Center University of Texas- TACC 3Q, 2015

39 © 2016 OpenPOWER Foundation Lots of resources & support for Linux ISVs and Developers

IBM PartnerWorldTechnical Support IBM DeveloperWorks • IBM Innovation Centers Technical resources, community, blogs, toolkits, How to • Free access to Power Hardware articles, beta code www.ibm.com/developerworks/linux/ • Free porting assistance • Free -based development environment www.ibm.com/partnerworld/wps/servlet/ContentHandler/pw_com_pwp_partnerw Regional Ecosystem Initiative – Recruiting Key orld-program Solutions Greater China, North America, Europe IBM Migration Factory Middleware and Industry Solutions Premier migration services for large applications www.ibm.com/systems/power/migratetoibm/index.html IBM Innovation Centers IBM Watson Developer’s Cloud .All 50+ centers worldwide now support Linux on Power Access to IBM Watson for developing cognitive .One-stop for ISVs, developers computing applications .HW access, technical support, demos, toolkits, Hands-on labs www.ibm.com/smarterplanet/us/en/ibmwatson/developers.html www.ibm.com/systems/power/software/linux/centers

IBM Power Development Cloud Site Ox Provide free access to Power hardware to ISVs for Porting Free On-demand cloud-based development platform hwww.ibm.com/partnerworld/wps/servlet/ContentHandler/stg_com_sys_power- using Linux on POWER8 for 2 weeks development-platform www.siteox.com

40 © 2016 OpenPOWER Foundation IBM Innovation Centers Provide End-to-End Assistance to ISVs

IBM Innovation Centers provide training and one-to- one guidance from building to marketing your solution. Japan North America • Tokyo • Austin

• Cambridge • Technical skills and hardware access across • Chicago • Dallas multiple platforms, local or remote, at little or • Silicon Valley • Toronto no cost for migration to IBM platforms Latin America Europe and Middle East Africa Asia Pacific • Amsterdam • Mexico City • London • Casablanca • Bangalore • • Qualify "Ready for" marks to broaden Barcelona • • Johannesburg • Sao Paulo Ljubljana • Bangkok • Bratislava • Milan • Lagos • • Bucharest Ho Chi Minh City marketing reach • Moscow • Nairobi • Budapest • Kuala Lumpur • Paris • Copenhagen • Prague • Manila • Dublin • Stockholm • Seoul • Workshops, seminars, and more conducted by • Ehningen • Tel Aviv • Shanghai • Hursley • Warsaw • Sydney • Istanbul trained subject matter experts. • Zurich • Kiev • La Gaude

All 50+ Innovation and Client Centers worldwide support Linux on Power For more information ibm.com/partnerworld/iic For questions, send an email to [email protected] 41 © 2016 OpenPOWER Foundation Data Centric: Minimizing Data Movement, Optimizing Workflows • Locate Computing where the data resides and relative to the task required • Within the system • Within the Data Center / Cloud location, or across locations as optimal • Shifting from algorithms to workflows Seismic Data Processing Workflow and Data Lifecycle

Data Input

Cold Storag e © 2016 OpenPOWER Foundation Different Solutions for Different Types of Workflows

Data Centric Conceptual View of Data Intensive- Applications Integer Workflow Conceptual View of Data Fusion/Uncertainty Quantification Workflow Low Spatial Locality

Floating Point Integer OPS OPS

High Spatial Conceptual View of Data Locality Intensive with Floating Point Workflow

Region defined by Compute Centric LINPACK Applications DCS Workflows: Mixed Compute Capabilities Required

Oil and Gas

4-40 Racks

Reservoir Seismic Analytics Capability:

• Complex code Financial Analytics Massively Parallel • Data Dependent Code 2-20 Racks Compute Capability: Paths / Computation Value At Risk • Lots of indirection / Analytics • Simple kernels, pointer chasing • Ops dominated (e.g. • Often Memory System DGEMM, Linpack) Latency Dependent All Source Analytics • Simple data access • C++ templated codes 1-5+ Racks patterns. • Limited opportunity for Graph • Can be preplanned for vectorization Analytics high performance. • Limited scalability Image Analysis • Limited threading Science opportunity 1-100’s Racks

Throughput Capability

44 12/9/13