OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core...

25
© 2014 International Business Machines Corporation 1 OpenPOWER: Open Source für Hardware Heiner Giefers IBM Research Zurich [email protected] Open Source Business Forum und Awards 2016 Mittwoch, 26. Oktober 2016 Haus der Universität Bern

Transcript of OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core...

Page 1: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

© 2014 International Business Machines Corporation 1

OpenPOWER:Open Source für Hardware

Heiner Giefers

IBM Research – [email protected]

Open Source Business Forum und Awards 2016

Mittwoch, 26. Oktober 2016

Haus der Universität Bern

Page 2: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

- Established in 1956

- Two Nobel Prizes (1986 and 1987)

- Today

~300 employees, ~3000 worldwide

40+ different nationalities

open innovation w/ 277 projects & 1900 partners in FP7, H2020, ...

IBM Research - Zurich

Page 3: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Today’s challenges demand innovation

Full system and stack

innovation required

Data holds

competitive value

Price/Performance

Moore’s Law

Processor

Technology

2000 2020

You are here

44 zettabytes

unstructured data

2010 2020

structured data

Data GrowthFirmware / OS

Accelerators

Software

Storage

Network

Page 4: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER, a catalyst for Open Innovation

The OpenPOWER Foundation creates an open ecosystem,

using the POWER Architecture to share expertise, investment, and

server-class intellectual property to serve the evolving needs of customers.

Performance of leading POWER architecture Broadens the capability and performance of the POWER platform

Open DevelopmentOpenPOWER enables greater innovation through

both open software and open hardware

Collaboration across multiple thought leadersCollaborative development model drives collective

thought leadership, simultaneously across multiple disciplines

Page 5: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER Foundation: 230+ Partners, 24 countries

Page 6: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Putting the “Open” in OpenPOWER

1. 2. 3.Open collaboration

and joint roadmaps

Open standards

and interfaces

Open hardware

and software

• Open technical membership organization. Welcoming new members in all areas

of the ecosystem.

• Different Membership Levels: Platinum, Gold, Silver, Associate & Academic. Annual fees ≤$100k (Free membership for ISVs and Associate/ Academic)

• Open development fosters innovation

• Joint roadmaps and joint development to accelerate time-to-market

Page 7: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Putting the “Open” in OpenPOWER

1. 2. 3.Open collaboration

and joint roadmaps

Open standards

and interfaces

Open hardware

and software

• 13 work groups developing technical specifications across the technology stack

• Free access to technical specs, design examples and documents

• POWER now supports Little Endian (ppc64le)

• Open accelerator interface: Coherent Accelerator Processor Interface (CAPI)

• OpenCAPI Consortium formed Oct. 2016 by AMD, Google, IBM, Mellanox, Micron.

OpenCAPI is an Open Interface that allows any microprocessor to attach to

• Coherent user-level accelerators and I/O devices

• Advanced memories

• Agnostic to processor architecture

Page 8: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Putting the “Open” in OpenPOWER

1. 2. 3.Open collaboration

and joint roadmaps

Open standards

and interfaces

Open hardware

and software

• POWER CPU IP licensable to members

• First “local” POWER derivative chip, POWERCORE CP1 (China)

• Open-sourcing IP for hardware accelerator design. CAPI “building blocks”

• Open-sourced (almost) complete firmware / OS stack [github.com/open-power]

• Open-sourcing many IBM tools, for example

• eCMD (IBM Systems hardware access API)

• Amester (whole system power measurement)

• Linux!

Page 9: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER Software Support

• 50+ IBM Innovation Centers, over 2,300 Linux ISVs developing on Power • Moving to little endian (almost complete)

Page 10: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER Core Technology Roadmap

2015 2016 2017

POWER8PCIe Gen3 & CAPI Interface

POWER8’NVLink & CAPI

POWER9Enhanced NVLink, OpenCAPI & PCIe Gen4

Connect-IBFDR Infiniband

PCIe Gen3

ConnectX-4EDR Infiniband

CAPI over PCIe Gen3

ConnectX-5Next-Gen Infiniband

Enhanced CAPI over PCIe Gen4Mellanox

Interconnect

IBM CPUs

NVIDIA GPUsKepler

PCIe Gen3Volta

Enhanced NVLinkPascalNVLink

GPU

FPGA

PCIe Gen3

1xGPU

FPGA

NVLINK

5x

GPUFPGA

2x

PCIe Gen4

25G Accelearator Link

7-10x

Accelerator Links

Page 11: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Coherent Accelerator Processor Interface (CAPI)

- Coherent read/write virtual memory access

- Data Caching

- Easier programming model

- Saves tens of thousands of instructions

- Enables applications not possible on I/O

PCIe

MemoryDevice

or Network

PSL

Memory Bus

CAPP

P8

Core

P8

Core

CAPI

Proxy

Core

POWER8 CPU PCIe Accelerator Card

FPGA

AFU 1

AFU 0

PSL = POWER Service

Layer

AFU = Accelerator

Functional Unit

Page 12: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

CAPI Example: Workload Optimized SystemIBM Data Engine for No-SQL

Load Balancer

500GB Cache

Node

Today’s NoSQL in memory (x86)

Infrastructure Requirements

- Large Distributed (Scale out)

- Large Memory per node

- Networking Bandwidth Needs

- Load Balancing

Uplink

WWW

500GB Cache

Node500GB Cache

Node500GB Cache

Node500GB Cache

Node

Uplink

One POWER8

Server

Flash Array w/

up to 60TB

Differentiated NoSQL

(POWER8 + CAPI Flash)

Infrastructure Attributes

- 192 threads in 4U Server drawer

- 60 TB of flash memory per 4U drawer

- Shared Memory & Cache for dynamic tuning

- Elimination of I/O and Network Overhead

- Cluster solution in a box

WWW

24:1 Reduction in

infrastructure

2.4x Price reduction

12x Less Energy

6x Less rack pace CAPI

DRAM

CAPI Flash

PCIe Flash

Access Latency

Page 13: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

More CAPI

Financial

L3 Order Book (Algo Logic)

Monte Carlo Risk Analysis (IBM)

Database

IBM Data Engine for No-SQL

(“CAPI-Flash”)

Erasure code for Hadoop

(IBM, Xilinx, Semptian)

SQL Accelerator (IBM Neteeza)

Key Value Store (KVS) (Xilinx)

Dynamic Time Warp Pattern Match (IBM)

Bitwise Encryption (DRC / SecurityFirst)

General Purpose

GZIP Compression (IBM)

Fast-Fourier Transfer (IBM)

Linear Algebra (Auviz)

JPEG Manipulation (ClusterTech)

Security

Bank Fraud Detection

In-Betweenness Djikstra (DRC)

Video Surveillance (SiliconScapes)

Digital DNA for forensics

(i-Abra)

Retail & Analytics

RegEx Text Analytics (IBM)

Mood Detection (SiliconScapes)

Real-Time Ad Auctions

(Algo-Logic)

Visually impaired assistance

(SiliconScapes)

Computer Vision & Learning

CV Library (Auviz)

People Identification

DNN Library (Auviz)

Activity Recognition (SiliconScapes)

Health Care

Light Activated Cancer Therapy

(U of Toronto)

Genomics Processing (Edico)

PairHMM Accelerator (IBM)

Page 14: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Non IBM-branded POWER Servers

Barreleye Gen1

GA 09/2016

POWER8 2-socket

mainboard, 04/2014

Many more systems available from:

+

Zaius POWER9

server (mockup)

Open Compute Project Server Design

- dual-socket POWER9 CPU

- DDR4 memory

- 48v Open Rack V2.0 compatible

- Open BMC

- OpenCAPIopencompute.org/wiki/Server/SpecsAndDesigns

openpowerfoundation.org/wp-content/uploads/2016/04/HardwareRevealFlyerFinal.pdf

Page 15: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Google's involvement in OpenPOWER

"We have ported our infrastructure onto the

POWER architecture. What that means is

that our toolchain supports POWER; for our

Google developers, enabling POWER for

their software applications is simply a matter

of modifying a config file and off they go.“

Maire Mahony

Google & Director OpenPOWER Foundation

Page 16: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

IBM OpenPOWER Servers

DDR

115 GB/s

GPU GPU

NVLINK

80 GB/s

GPU GPU

NVLINK

80 GB/s

DDR

115 GB/s

IBM Power System S822LC

for Big Data (“Firestone”)

IBM Power System S822LC

for HPC (“Minsky”)

First IBM OpenPOWER System

- Dual 8- or 10-core P8

- Up to 3.32 GHz

- 230 GB/s sustained memory bandwidth

- Starting at $9,699

NVIDIA: Tesla P100 GPU with NVLink

Ubuntu by Canonical:

Launch OS supporting NVLink and Page

Migration Engine

Wistron: Platform co-design

Mellanox: InfiniBand/Ethernet NIC

HGST: Optional NVMe Adapters

Broadcom: Optional PCIe Adapters

QLogic: Optional Fiber Channel PCIe

Samsung: 2.5” SSDs

Hynix, Samsung, Micron: DDR Memory

IBM: POWER8 CPU with NVLink

Page 17: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

IBM POWER in the Cloud

POWER8 available on SoftLayer

SaaSPaaSIaaSBare Metal

IBM Power Development Cloud

IBM Bluemix

IBM SaaS portfolio w/ 100+ offerings

Page 18: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

SuperVessel: OpenPOWER Cloud

- Open cloud platform based on Power/OpenPOWER and OpenStack technology for

business partners, developers, and university students

- The labs include: Big Data, Internet of Things, cognitive computing, acceleration

(FPGA/GPU) and virtualization on POWER

- Accelerator Cloud @ZurichResearchLab

- Demo available: zhc2.zurich.ihost.com

Page 19: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

PADC – Enhance OpenPOWER Ecosystem

• 3 PADC Centers (Power Acceleration and Design Center) involving IBM, NVIDIA and

Mellanox focused on accelerated applications and technical collaborations

• IBM HPC Benchmark center UK Science and Technology Facilities Council (STFC)

PADC to optimize Performance of OpenPOWER systems

for Modelling Simulation and Big Data Analytics

IBM PADC Montpellier joined center with NVIDIA and

Mellanox for technical collaboration, consulting,

application enablement, benchmarking, …

IBM PADC Boeblingen joined center with NVIDIA for

technical collaborations, consulting, application

enablement, system external access, …

IBM Poughkeepsie

POWER HPC

benchmark center

IBM has created 50+ innovation center to help ISVs port their

application to Linux on POWER. These centers provide:

- Technical skills and hardware access across multiple platforms,

local or remote, at little or no cost for migration to IBM platforms

- Qualify “ready for” marks to broaden marketing research

- Workshops, seminars, and more conducted by trained subject

matter experts.

Page 20: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Starting today…

Co-located with the OpenStack European Summit

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

Page 21: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Main Website openpowerfoundation.org

Heiner Giefers ([email protected])Join the rebellion!

Page 22: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Lots of resources & support for Linux ISVs and Developers

Site Ox

Free 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_partnerworld-program

IBM Innovation Centers

•All 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 Cloud

Provide free access to Power hardware to ISVs for

Portingwww.ibm.com/partnerworld/wps/servlet/ContentHandler/stg_com_sys_power-development-platform

Regional Ecosystem Initiative – Recruiting Key

Solutions Greater China, North America, Europe

Middleware and Industry Solutions

IBM DeveloperWorks

Technical resources, community, blogs, toolkits, How to

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

IBM Watson Developer’s Cloud

Access to IBM Watson for developing cognitive

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

IBM Migration Factory

Premier migration services for large applicationswww.ibm.com/systems/power/migratetoibm/index.html

Page 23: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

Innovation Centers Provide End-to-End Assistance to ISVs

For more information ibm.com/partnerworld/iic For questions, send an email to [email protected]

IBM Innovation Centers provide training and one-to-one guidance from building to marketing your solution.

• Technical skills and hardware access across

multiple platforms, local or remote, at little or

no cost for migration to IBM platforms

• Qualify "Ready for" marks to broaden

marketing reach

• Workshops, seminars, and more conducted by

trained subject matter experts.

All 50+ Innovation and Client Centers worldwide support Linux on Power

Latin America

• Mexico City

• Sao Paulo

North America

• Austin

• Cambridge• Chicago• Dallas• Silicon Valley• Toronto

Europe and Middle East• Amsterdam• Barcelona• Bratislava• Bucharest• Budapest• Copenhagen• Dublin• Ehningen• Hursley• Istanbul • Kiev• La Gaude

• London• Ljubljana• Milan• Moscow• Paris• Prague• Stockholm• Tel Aviv• Warsaw• Zurich

Africa • Casablanca• Johannesburg• Lagos• Nairobi

Japan

• Tokyo

Asia Pacific

• Bangalore

• Bangkok

• Ho Chi Minh City

• Kuala Lumpur

• Manila

• Seoul

• Shanghai

• Sydney

Page 24: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER Workgroups: Open Standards

Developer Platform

System SW

HW Architecture

Accelerator

Compliance

25g IO Compatibility

Memory

OpenPOWER I/O

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

FSI Specification FSI SpecCharter

Charter Pers Med

Charter Integrated Solutions Integrated Solutions

Personalized Medicine

SDK – Software Developer KitSP010 – Tyan OpenPOWER Customer Reference System

CAPI – Coherent Accelerator Processor Interface

AFU – Accelerator Function Unit

FSI – Field Replaceable Unit (FRU) Service Interface

OPMB – OpenPOWER Memory Bus

ABI – Application Binary Interface

2014 2015 2016

Page 25: OpenPOWER - Home > CH Open · for HPC (“Minsky”) First IBM OpenPOWER System-Dual 8- or 10-core P8-Up to 3.32 GHz-230 GB/s sustained memory bandwidth-Starting at $9,699 NVIDIA:

OpenPOWER Accelerators

Field Programmable

Gate Array (FPGA)

Graphics Processing

Unit (GPU)

Description

Reconfigurable hardware

Task customized, low latency, low

power

1000s of simple cores

High bandwidth, floating point, and

parallelism

Example

Use Cases

Compression, encryption, high speed

streaming, search, Monte Carlo

simulations

Neural networks, speech recognition,

chemistry, simulations, JAVA, Hadoop,

graphics, Monte Carlo

Power Chip

Integration

Coherent Accelerator Processor

Interface (CAPI)

NVIDA via PCIe x16

NVIDIA NVLink