Introduction to the OpenPOWER Foundation - Open Source Days event

The OpenPOWER Foundation

Dr Mandie QuartlyOpenPOWER Global Alliances, Europe, IBM@mandieq

© 2016 OpenPOWER Foundation

What would you do?

∞ 0Performance Cost


The only constant is change...

Source

http://thehipperelement.com/post/118359687821/evolution


Time

IT E

ffic

ien

cy

Multi-CoreMulti-Thread

Workload Optimised Systems

1970’s ~2004 ~2015 >>2025

Quantum, DNA, Neural or ???

computing

CLOCK ERA CLOUD ERA WOS ERA QUANTUM ERA

ScalingSingle thread


Demands and consumption of technology are changing too..


Introducing the OpenPOWER Foundation…5 Founding members in 2013


Accelerated innovation through collaboration of

partners

Amplified capabilities driving industry

performance leadership

Vibrant ecosystem through open development

What's OpenPOWER all about?

OpenPOWER is an open development community using the POWER Architecture


OpenPOWER today


Speed to innovation | Standards | Choice


Open allows you to create what you need


Open allows you to create what you need(sometimes those things can be very specific to a certain need...)


What's the best tool for the job?


What's so special about POWER anyway?


MemoryBuffer

DRAMChips

POWER8

Up to 12 Cores, up to 96 ThreadsL1, L2, L3 + L4 CachesUp to 1 TB per socket

Up to 230 GB/s sustained memory bandwidth

POWER8 Processor – the low down...


NVIDIA NVLink Technology

NVIDIA Tesla GPU NVIDIA Tesla GPU with NVLink

POWER8 POWER8with NVLink

80 GB/sPeak*

PCIe x16

Current GPU Attach

Graphics Memory

System Memory

Graphics Memory Graphics Memory

System Memory

40+40 GB/s

40+40 GB/s

16+16 GB/s

Future NVLink GPU Attachment

© 2016 OpenPOWER Foundation21

2015 2016 2017

POWER8 POWER8' POWER9

CAPI Interface NVLink Enhanced CAPI & NVLink

Connect-IBFDR Infiniband

PCIe Gen3

ConnectX-4EDR Infiniband

CAPI over PCIe Gen3

ConnectX-5Next-Gen Infiniband

Enhanced CAPI over PCIe Gen4

MellanoxInterconnect

IBM CPUs

NVIDIA GPUs KeplerPCIe Gen3

VoltaEnhanced NVLink

PascalNVLink

Systems

Joint roadmaps looking forward...


Typical I/O Model FlowTypical I/O Model Flow

Flow with a Coherent ModelFlow with a Coherent ModelShared Mem.

Notify AcceleratorShared Mem.

Notify Accelerator AccelerationAcceleration Shared MemoryCompletion

Shared MemoryCompletion

CAPP PCIe

POWER8 Processor

DD CallDD Call Copy or PinSource DataCopy or PinSource Data

MMIO NotifyAccelerator

MMIO NotifyAccelerator AccelerationAcceleration Poll / Int

CompletionPoll / Int

CompletionCopy or Unpin

Result DataCopy or Unpin

Result DataRet. From DD

CompletionRet. From DD

Completion

FPGA

Coherent Accelerator Processor Interface (CAPI)


Linux support for POWER

Same source and distribution release schedules as x86Simplified x86 application migration with little endian distributions Enterprise support for all three from IBM or distributors


Back to the original question...

∞ 0Performance Cost


A spectrum of “caring”...

“I couldn't care less about the

platform”

“I want to change the specifics of <stack part>”


● Change is inevitable● Open is good● What's the best tool for the job?

OpenPOWER is all about ecosystem... => collaboration through open development => speed to innovation => price / performance


Join the rebellion!

Main website → openpowerfoundation.orgOverview → ibm.biz/openpower_overview

[email protected]

Reference material


New IBM “LC” Power Systems for Linux

S812LC CPUCPU

MemoryMemory

1x POWER8 CPU 10 cores, 2.9-3.3 GHzUp to 1TB Memory

115 GB/s Memory Bandwith14 Drives (84TB, HDD, SSD)

Optimized for Hadoop, Spark

S822LC CPUCPU

MemoryMemory

Built for CPU Performance, CAPI

Differentiation

CPUCPU2x POWER8 CPU

10 cores each, 2.9-3.3 GHzUp to 1TB Memory

230 GB/s Memory BandwidthOpt. CAPI-attached accelerators

S822LCHPC

CPUCPU

MemoryMemoryBuilt for GPU Acceleration

CPUCPU 2x POWER8 CPU 10 cores each, 2.9-3.3 GHz

Up to 1TB Memory230 GB/s Memory Bandwidth

2x NVIDIA Tesla K80 GPUsGPUGPU GPUGPU


Data Engine for NoSQL with 40TB CAPI-attached flash

Load Balancer

500GB Cache Node

10Gb Uplink

POWER8 Server

Flash Array w/ up to 40TB

After: NoSQL POWER8 + CAPI Flash

WWW

10Gb Uplink

WWW

Backup Nodes

500GB Cache Node500GB Cache

Node500GB Cache Node500GB Cache

Node

Before: NoSQL in memory (x86)

24U

4U

Less is More24:1 physical server consolidation =

6x less rack space

24:1 server consolidation

3x lower cost per user


Membership Level

Annual Fee$ USD FTEs Technical Steering Committee Board / Voting position

Platinum $100k 10 One seat per member not otherwise represented

Includes board positionIncludes TSC position

Gold $60k 3 May be on TSC if Work group lead

Gold members may elect one board representative per three gold members

Silver $20k $5k if <300 employees 0 May be on TSC if

Work group leadSliver members may elect one board representative for all silver members

Associate & Academic $0 0 May be on TSC if

Work group leadMay be elected to one community observer, non-voting Board seat

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 agreement, Bylaws, and IP Rights Policy available for reviewwww.openpowerfoundation.org

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.

Membership Options


Engage in the OpenPOWER communityA

CT

ION

AC

TIO

NO

UT

CO

ME

OU

TC

OM

EIN

TE

RE

SIN

TE

RE

STT

Technology and Software Innovators

Discuss areas of collaboration and synergy in

OpenPOWER.

Sign up for membership and join a work group.

Build technical and business relationships.

Develop collaborative innovations

with compelling value

Innovators, Integrators, and Partners

Deep understanding of workload demands and

consumption preferences of end users.

Team with innovators and end users to drive

requirements, engage in specific projects.

Deliver collaborative innovations


End Users

Strategic imperatives and workload demands for performance and cost

optimization.

Require open software and systems with choice and

flexibility.

Engage directly on system design options.

Deploy collaborative innovations



OpenPOWER Work Group Roadmap

2014 2015 2016

Developer Platform

System SW

HW Architecture

Accelerator

Compliance

Proposed Work Groups Integrated Solutions

Pers Med

SP010 – Tyan OpenPOWER Customer Reference System CAPI – Coherent Accelerator Processor InterfaceABI – Application Binary Interface

AFU – Accelerator Function UnitFSI – Field Replaceable Unit (FRU) Service InterfaceSDK – Software Developer Kit

25g IO Compatibility

FSI Spec

Memory

OpenPOWER I/O

Work Group

Charter Compliance SpecificationDraft Review WG Spec

CompSTD

CharterOpenPOWER ISA Profile V1IO Device Architecture V2Coherent Accel Intf Arch

OpenPOWER ISA Profile V2IO Device Architecture V3Coherent Accel Intf Arch

Charter P8 SP010Data

P8 2U2SReference

P8+ 1U1SReference

P8+ 2U2SReference

OPMB Intf. Spec V1Charter

Charter CAPI AFU Intf Spec V1

OpenCL SDK

CAPI AFU Intf Spec V2

Charter

CAPI LinuxSDK 64b ABI

Platform Ref

Sys I/O Enablement GuideCharter

Charter 25g IO Spec

OPMB – OpenPOWER Memory Bus


Designed for DataOptimised price-performance for the unique needs of Spark

S812LC delivers optimized Spark price-performance based on 10 SparkBench benchmarks

– Complete the same Spark workloads for less than ½ the cost of Intel Xeon E5-2690 v3 systems o 2.3X BETTER performance per dollar spent

– 94% more Spark workloads in the same rack space versus Intel Xeon E5-2690 v3 systems o 1.94X BETTER performance per system (10 core S812LC vs 24

core DL380)

IBM S812LC10c/80t

• All results are based on IBM Internal Testing of 10 SparkBench benchmarks consisting of SQL RDD Relation, Twitter, Pageview Streaming, PageRank, Logistic Regression, SVD++, TriangleCount, SVM, MF, SQL Hive

• IBM Power System S812LC 10 cores / 80 threads, POWER8; 2.9GHz, 256 GB memory, Ubuntu 15.04, Spark 1.4, OpenJDK 1.8• Intel Xeon HP DL380; 24 cores / 48 threads, E5-2690 v3; 2.3GHz , 256 GB memory. Ubuntu 15.04, Spark 1.4, OpenJDK 1.8• Pricing is based on list prices of HP DL380 and estimated prices of IBM Power S82LC

HP DL380E5-2690 v3

24c/48t

IBM S812LC10c/80t

HP DL380E5-2690 v3

24c/48t


• 50 IBM Innovation Centers • Compelling PoCs • Support for little endian applications

Over 1,800 Linux ISVs developing on Power

Big Data & Machine Learning

Big Data & Machine LearningCloudCloud Mobile EnterpriseMobile EnterpriseHPCHPC

miniDFTminiDFTCTHCTH

BLASTBLASTBowtieBowtieBWABWA

FASTAFASTAHMMERHMMER

GATKGATKSOAP3SOAP3

STAC-A2STAC-A2SHOCSHOC

Graph500Graph500IlogIlog

CHARMMCHARMMGROMACSGROMACS

NAMDNAMDAMBERAMBER

RTMRTMGAMESSGAMESS

WRFWRFHYCOMHYCOMHOMMEHOMME

LESLESMiniGhostMiniGhostAMG2013AMG2013

OpenFOAMOpenFOAM