Mandie Quartly, Ph.D. - IBM · © 2016 OpenPOWER Foundation …no i tadnuo FRWOEPn Oepeh tgn icudo...

Mandie Quartly, Ph.D.OpenPOWER 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


Source

Houston, we have a problem...

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


Time

IT E

ffic

ien

cy

Multi-Core

Multi-Thread

Workload

Optimised

Systems

1970’s ~2004 ~2015 >>2025

Quantum, DNA,

Neural or ???

computing

CLOCK ERA CLOUD ERA WOS ERA QUANTUM ERA

Scaling

Single thread


Introducing the OpenPOWER Foundation…5 Founding members in 2013


Accelerated innova�on

through collabora�on of

partners

Driving industry

performance leadership

Vibrant ecosystem

through open

development

What's OpenPOWER all about?

OpenPOWER is an open development community using the POWER Architecture


This is what a revolu�on looks like


Speed to innovation | Standards | Choice


Open allows you to create what you need


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...

Now with added

PowerAccel!


Field ProgrammableGate Array

GraphicsProcessingUnit

Descrip*onRecon�gurable hardware

Task customized, low latency, low power

1000s of simple cores

High bandwidth, $oa�ng point, and parallelism

ExampleUse Cases

Compression, encryp�on, high speed streaming, search, Monte Carlo simula�ons

Deep neural networks, speech recogni�on, chemistry, simula�ons, JAVA, Hadoop, graphics

Power ChipIntegra*on

Coherent Accelerator Processor Interface (CAPI)

NVIDIA NVLink

Workload accelerators and POWER


System

Bottleneck

Graphics

Memory

New

POWER8

CPU

New

POWER8

CPU

DDR4DDR4

P100

Tesla

GPU

P100

Tesla

GPU

NV

Lin

k NV

Lin

k

NVLink

115GB/s

80 GB/s

80 G

B/s

CPUCPUDDR4DDR4

GPUGPU

PC

Ie

32

GB

/s

Graphics

Memory

Graphics

Memory

Current CPU – GPU

PCIe Data Pipe

POWER8 with NVLink Data Pipe

P100

Tesla

GPUz

P100

Tesla

GPUz

POWER8 with NVLink = 2.8x bandwidth


Typical I/O Model FlowTypical I/O Model Flow

Flow with a Coherent ModelFlow with a Coherent ModelShared Mem.

Notify Accelerator

Shared Mem.

Notify Accelerator AccelerationAccelerationShared Memory

Completion

Shared Memory

Completion

CAPP PCIe

POWER8 Processor

DD CallDD CallCopy or Pin

Source Data

Copy or Pin

Source DataMMIO Notify

Accelerator

MMIO Notify

AcceleratorAccelerationAcceleration

Poll / Int

Completion

Poll / Int

CompletionCopy or Unpin

Result Data

Copy or Unpin

Result DataRet. From DD

Completion

Ret. From DD

Completion

FPGA

Coherent Accelerator Processor Interface (CAPI)


Linux support for POWER

Same source and distribution release schedules as x86

Simplified x86 application migration with little endian distributions

Enterprise support for all three from IBM or distributors


So what's new?


Vision Execution Adoption

Incorporated December 2013

Opening the door, a

license to innovate

OpenPOWER Summit March 2015

Game changing

solutions

OpenPOWER Summit April 2016

From proof of concept

to industry revolution

OpenPOWEROpenPOWER

Open InnovationOpen Innovation


POWER8 Architecture POWER9 Architecture

2014POWER8

12 cores

22nm

New Micro-Architecture

New ProcessTechnology

2016POWER8w/ NVLink

12 cores

22nm

EnhancedMicro-

ArchitectureWith NVLink

2017P9 SO24 cores

14nm


Direct attachmemory


TBDP9 SU12 cores

14nm

EnhancedMicro-

Architecture

BufferedMemory

POWER6 Architecture POWER7 Architecture

2007POWER6

2 cores

65nm



2008POWER6+

2 cores

65nm+

EnhancedMicro-

Architecture

EnhancedProcess

Technology

2010POWER7

8 cores

45nm



2012POWER7+

8 cores

32nm

EnhancedMicro-

Architecture


Focus on EnterpriseTechnology and Performance Driven

Focus on Scale-Out and EnterpriseCost and Acceleration Driven

2018 - 20P9 SO

10nm - 7nm

Existing Micro-

Architecture

FoundryTechnology

Partner ChipPOWER8/9

2020+

New Micro- Architecture

NewTechnology

POWER10

Future

T

B

D

POWER Processor Roadmap


Zaius 1.25 OU• 2 POWER9 CPUS

• 32 DDR4 DIMM SLOTS

• 2X G4 PCIE X16 FHFL

SLOTS

• 1X G4 X16 HHHL SLOT

• 1X G4 X16 OCP MEZ

• 1X M.2 SATA PORT

• 1X SATA PORT

• 15X 2.5” SAS/SATA/NVME

SLOTS

• BMC W/GBE LOM

• “DISKLESS” OPTION


OpenPOWER Innovation in the DesignPower Systems S822LC for High Performance Computing (aka Minsky)

NVIDIA:

Tesla P100 GPU Accelerator with

NVLink (GPU↔GPU & GPU↔CPU)

Ubuntu by Canonical:

Launch OS supporting NVLink and

Page Migration Engine

Wistron: Platform co-design

Mellanox: InfiniBand/Ethernet

Connectivity in and out of server

HGST: Optional NVMe Adapters

Broadcom: Optional PCIe Adapters

QLogic: Optional Fiber Channel PCIe

Samsung: 2.5” SSDs

Hynix, Samsung, Micron: DDR4

IBM: POWER8 CPU with NVLink

S822LC For High

Performance Computing

• 2 socket, 2U

• POWER8 with NVIDIA

NVLink

• Up to 4 integrated NVIDIA

“Pascal” P100 GPUs

S822LC For Big Data

• 2 socket, 2U

• Storage-centric and high

through-put workloads

• Big data acceleration with

work CAPI and GPUs

S821LC

• 1 socket, 2U

• Storage rich for big

data applications

• Memory Intensive

workloads

S822LC

• 2 socket, 2U

• Memory Intensive

workloads

S812LC

• 2 sockets, 1U

• Dense computing

NEWNEW

NEWNEW

NEWNEW

High Performance

Computing

Introducing the IBM Power Systems LC LineOpenPOWER servers for cloud and cluster deployments that are different by design

Who is using OpenPOWER and how?

STFC Hartree Centre

Power Acceleration and Design CentreCollaboration with IBM and NVidia

Panther System: 32x POWER8 + NVidia K80 GPUs

“take advantage of IBM and NVIDIA’s technological leadership in supercomputing and the Hartree Centre's expertise and experience in delivering solutions to real-world problems”


Tencent

Integration of OpenPOWER systems into their Hyper Scale Data Centres for Big Data workloads

Recent results running Spark Terasort:3x better performance vs former x86 infrastructure, with 2/3rd fewer servers.

512x SuperMicro POWER8 servers


https://www.youtube.com/watch?v=GZAFzlWN8FU

KineticaGPU-powered database

“insights in real time from live data”

Kinetica on OpenPOWER = 2.6X faster than x86

POWER8 with NVLink + NVidia P100 GPUs compared to x86 + NVidia K80 GPUs

https://www.youtube.com/watch?v=GZAFzlWN8FU


Video analytics using FGPAConnected via CAPI – only with POWER8

CAPI = 30 fpsSoftware = 2 fps

ibm.biz/powercapi

http://ibm.biz/powercapi



A spectrum of “caring”...

“I couldn't care less about the

platform”

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

I want to know more / how do I get started?

http://www.softlayer.com/info/power8 http://www-03.ibm.com/systems/power/hardware/

Like hardware? Like virtual?

http://www.softlayer.com/info/power8

http://www-03.ibm.com/systems/power/hardware/


Come and find out more...

Co-located with the OpenStack European Summit & no cost to attend!

http://openpowerfoundation.org/openpower-summit-europe/

http://openpowerfoundation.org/openpower-summit-europe/


Join the rebellion!Main website → openpowerfoundation.orgOverview → ibm.biz/openpower_overview

[email protected] | @mandieq

http://www.openpowerfoundation.org/

http://ibm.biz/openpower_overview

Reference material


Engage in the OpenPOWER communityA

CT

ION

AC

TIO

NO

UT

CO

ME

OU

TC

OM

EIN

TE

RE

ST

INT

ER

ES

T

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



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 lead

Sliver members may elect one board representative for all silver

members

Silver ISV$0 if ISV is <300

employees 0 May be on TSC if Work group lead

Sliver members may elect one board representative for all silver

members

Associate & Academic

$0 0 May be on TSC if Work group lead

May be elected to one community observer, non-voting Board seat

Membership agreement, Bylaws, and IP Rights Policy available for review

www.openpowerfoundation.org

Membership agreement, Bylaws, and IP Rights Policy available for review

www.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

New




2014 2015 2016

Developer Platform

System SW

HW Architecture

Accelerator

Compliance

SP010 – Tyan OpenPOWER Customer Reference System

CAPI – Coherent Accelerator Processor InterfaceAFU – Accelerator Function Unit

FSI – Field Replaceable Unit (FRU) Service Interface

25g IO Compatibility

Memory

OpenPOWER I/O

12 Work Groups

CharterCompliance SpecificationDraft Review WG Spec

CompSTD

Charter

OpenPOWER 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

CharterCAPI 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

ABI – Application Binary Interface

FSI Specification FSI SpecCharter

SDK – Software Developer Kit

Charter Pers Med

CharterIntegrated Solutions Integrated Solutions

Personalised Medicine

OpenPOWER Ready

Physical Science

Charter

Charter


• Contributing OpenPOWER Open Compute

reference design via collaboration between IBM

and Rackspace.

• Contributing OpenPOWER firmware 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

OpenPOWER and Open Compute Project are teaming to deliver price performance advantage on Power


Coming soon: Barreleye

In partnership with Avago, IBM,

Mellanox, PMC & Samsung


Introducing OpenPOWER Ready

● Products / solutions may display badge

● Participants request to use it● Compatibility● Not Certification or Compliance

● OpenPOWER Ready™ List

http://openpowerfoundation.org/technical/openpower-ready/

http://openpowerfoundation.org/technical/openpower-ready/


Maximum Performance, Minimum Change

From highly parallelised languages to memory-intensive apps, running the same code on POWER can dramatically increase performance – and with all major Linux distros supported, switching is easy.

Accelerate Beyond Limits

POWER’s deeply integrated accelerator support means code can push beyond the limits of what’s possible with standard processors.

Work Less, Do More

With better performance and scale, you can spend less time building / testing / optimising and more time developing the next new thing.

Rule the Stack – No black boxes

With OpenPOWER’s fully open architecture, developers rule the stack – hardware, )rmware, and software are open and ready for optimisation.

POWER8: Designed for Big Data

SMT8 – 64+ threads per CPU

Massive memory bandwidth & workspace

Why Develop on OpenPOWER?


http://developers.openpowerfoundation.org/

http://developers.openpowerfoundation.org/

© 2016 IBM Corporation

POWER8 => built for performance

* Intel calls this Hyper-Threading Technology (No HT and with HT)

*32KB running in “Non-RAS mode” 16KB results in better RAS

**85GB running in “Non-RAS mode” and dual-device error NOT supported

1 = The Micro-architecture features above contributed to premium performance for the POWER8 systems compared to Ivy Bridge systems

Sandy Bridge EP

E5-x6xx

Ivy Bridge EP

E5-26xx v2

Ivy Bridge EX

E7-88xx v2

Haswell EPE5-26xx v3

POWER 7+ Systems

POWER8

Clock rates (GHz) 1.8–3.6 1.7-3.7 1.9-3.4 1.6-3.5 3.1–4.4 GHz 3.0-4.1 GHz

SMT options 1,2* 1, 2* 1, 2* 1, 2* 1, 2, 4 1, 2, 4, 8

Max Threads / sock 16 24 30 36 32 96

Max L1 Cache 32KB 32KB* 32KB* 64 KB 32KB 64KB

Max L2 Cache 256 KB 256 KB 256 KB 256KB 256 KB 512 KB

Max L3 Cache 20 MB 30 MB 37.5 MB 45 MB 80 MB 96 MB

Max L4 Cache 0 0 0 0 0 128 MB

Memory Bandwidth 31.4-51.2 GB/s

42.6-59.7 GB/s

68-85**

GB/s

51-68

GB/s

100 – 180 GB/sec

115 - 410 GB/sec


Motherboard Chassis Rack

OpenPOWER Open Compute Project systems coming to market


• Over 2,500 Linux ISVs developing on Power

• 50 IBM Innova�on Centers

• Compelling PoCs

• Support for li6le endian applica�ons

Accelerated software innovation

Big Data & Machine

Learning

Big Data & Machine

LearningCloudCloud Mobile EnterpriseMobile Enterprise

Major Linux Distros

HPCHPC

miniDFTminiDFT

CTHCTH

BLASTBLAST

Bow�eBow�e

BWABWA

FASTAFASTA

HMMERHMMER

GATKGATK

SOAP3SOAP3

STAC-A2STAC-A2

SHOCSHOC

Graph500Graph500

IlogIlog

CHARMMCHARMM

GROMACSGROMACS

NAMDNAMD

AMBERAMBER

RTMRTM

GAMESSGAMESS

WRFWRF

HYCOMHYCOM

HOMMEHOMME

LESLES

MiniGhostMiniGhost

AMG2013AMG2013

OpenFOAMOpenFOAM


Site OxFree On-demand cloud-based development platform

using Linux on POWER8 for 2 weekswww.siteox.com

IBM PartnerWorldTechnical Support

•IBM Innovation Centers•Free access to Power Hardware•Free porting assistance•Free Eclipse-based development environmentwww.ibm.com/partnerworld/wps/servlet/ContentHandler/pw_com_pwp_p

artnerworld-program

IBM Innovation CentersAll 50+ centers worldwide now support Linux on Power

One-stop for ISVs, developers

HW access, technical support, demos, toolkits, Hands-on

labswww.ibm.com/systems/power/software/linux/centers

IBM Power Development CloudProvide free access to Power hardware to ISVs for

Portinghwww.ibm.com/partnerworld/wps/servlet/ContentHandler/stg_com_sys_powe

r-development-platform

Regional Ecosystem Initiative – Recruiting Key

SolutionsGreater China, North America, Europe

Middleware and Industry Solutions

IBM DeveloperWorksTechnical resources, community, blogs, toolkits, How to

articles, beta codewww.ibm.com/developerworks/linux/

IBM Watson Developer’s CloudAccess to IBM Watson for developing cognitive

computing applicationswww.ibm.com/smarterplanet/us/en/ibmwatson/developers.html

IBM Migration FactoryPremier migration services for large applicationswww.ibm.com/systems/power/migratetoibm/index.html

Lots of resources & support for Linux ISVs and Developers


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 Pascal GPUs

3 PCIe slots, 3 CAPI enabled, IB Add-in

Air or water cooled

Default 3 year 9x5 warranty, 100% CRU

Unprecedented performance and application gains with the new

POWER8 with NVLink processor —delivering 2.8X the CPU-

GPU bandwidth compared to x86 based systems

CPU - GPU NVLink: POWERAccel IO Capability not

available on x86

Simpler programming and larger data sets: GPU

access to system memory with page migration

Water cooled: Improves data center efficiency and enable

sustained performance with CPU (Turbo)/GPU (Boost)

performance

Power Systems S822LC for High Performance Computing

45

First in the Industry with CPU-GPU NVLink for bandwidth differentiation


Power Systems S822LC for Big Data


Designed from the ground up for Big Data Workloads

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 K80 GPU capable

Default 3 year 9x5 warranty, 100% CRU

The S822LC for Big Data leverages the best processor in

the industry for big data with POWER8 resulting in superior

data through-put for Linux workloads

Leadership data through-put in a storage rich flexible

server design resulting in an average of 80% better price-

performance on Open Source Data Bases

Leadership virtualization density with 42% More VM’s per

Server than Intel Xeon E5-2690 v4 systems

Acceleration of big data workloads with the POWERAccel

family of technologies


System Details 2 socket, 1U Up to 20 cores (2.09-2.32Ghz)

512 GB Memory (16 DIMMs)

115 GB/sec max sustained memory bandwidth)

4x3.5” SATA drives, 32 TB Storage 4 PCIe slots, 3 CAPI enabled

1 Nvidia K80 GPU capable

Default 3 yr 9x5 Warranty 100% CRU

For compute-Intensive workloads, 2 POWER8

processors can now be accessed in a 1U server

design

Delivering nearly 7,000 threads in a rack for

superior container deployments, that’s 80%

more than HP DL360 V4 systems

Ideal for dense database, virtualization and

container deployments requiring a balanced

system design of through-put and density

Superior 1U data throughput solution-specific

hardware accelerators and the POWERAccel

family of technologies

The Power Systems S821LC Drives the most compute in the smallest rack space

Date post:	23-Sep-2019
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

Mandie Quartly, Ph.D. - IBM · © 2016 OpenPOWER Foundation …no i tadnuo FRWOEPn Oepeh tgn icudo...

Documents