L3 Cache on Chip Аппаратные Ускорители

IBM Power Systems: представляем процессоры POWER7+ и новое поколение серверов класса

предприятия

Алексей Бойко POWER Product Manager [email protected] Немного истории

1974-1978 – проект 801.

John Cocke – «отец» IBM RISC архитектуры

Source:JOHN COCKE AND V. MARKSTEIN –The evolutions of RICS technology at IBM

Специализированные команды для операций с памятью — чтения или записи.

Операции вида «прочитать-изменить-записать» отсутствуют.

Любые операции «изменить» выполняются только над содержимым регистров (т. н. архитектура load-and-store).

Большое количество регистров общего назначения (32 и более).

Отсутствие поддержки операций вида «изменить» над укороченными типами данных — байт, 16-битное словами.

Отсутствие микропрограмм внутри самого процессора.

© 2012 IBM Corporation RISCRISC процессорыпроцессоры *ARM (Adwanced RISC Machines)— Apple iPods /iPhone/iPad Palm and PocketPC PDAs RIM BlackBerry smartphone/email devices Nintendo HTC(Qualcom proc) *PowerPC (Apple–IBM–Motorola alliance) *Motorola 68000 *MIPS(SGI computers and the PlayStation, PlayStation 2, Nintendo 64, PlayStation Portable game consoles) *IBM POWER(IBM's supercomputers, midrange servers and workstations, in Nintendo's Gamecube and Wii, Microsoft's Xbox 360 and Sony's PlayStation 3 game consoles, EMC's DMX range of the Symmetrix SAN, and in many embedded applications like printers and cars. * SPARC, by Oracle/Sun Microsystems and Fujitsu * Hewlett-Packard's PA-RISC, also known as HP-PA * Alpha * Hitachi's SuperH

© 2012 IBM Corporation ЕстьЕстьRISCRISC процессоры процессоры лили жизньжизнь нана нана Марсе?Марсе?МарсеМарсе «Because of this, he added that an Intel Core i7 processor, for example, wouldn't survive a day on Mars because of the extremes in temperature.»

The single-board computer aboard NASA's Curiosity rover is built around a PowerPC RAD750 microprocessor provided by BAE Systems.

ComputerWorld August 8 2012 http://www.computerworld.com/s/article/9230060/NASA_Your_smartphone_is_as_smart_as_the_Curiosity_rover

CBS News http://www.cbsnews.com/network/news/space/home/spacenews/files/081012_msl_update.html

45% POWER6 Live Partition 40% Mobility POWER5 Micro-Partitioning 35% POWER7 Workload Optimized 30% Leadership POWER4 Dynamic LPARs HP -16% YTY* 25%

20% Oracle -24% YTY* HP Sun/Oracle IBM

15% 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 6 6 7 7 7 7 8 8 8 8 9 9 9 9 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Source: IDC Worldwide Quarterly Server Tracker, http://www.idc.com/getdoc.jsp?containerId=IDC_P348 * As reported last fiscal quarter

~ 3года новое поколение ~ 18месяцев “+”“+” модели

POWER8 POWER7/7+ Most POWER6/6 Most POWERful, + POWER5/5 POWERful & Scalable Scalable and Exclusive + Fastest Processor in Performance POWER4/4 Processor Industry. 4,6,8 Core . More Cores Hardware . 32MB On-Chip . Larger Cache + Virtualization In Industry eDRAM . 4th Gen SMT . Dual Core . Power Optimized . Reliability ++ . for.Dual Unix Core & Quad Linux High Frequencies Cores . Accelerators + First Dual Core . . . Core Md Virtualization + Mem Subsystem + more… . + . 22nm in Industry .Enhanced Scaling Memory Subsystem . Dual Core + . 4 Thread SMT++ High Level design .2 Thread SMT . . Chip Multi . Reliability + complete .Distributed Switch + Altivec . VSM & VSX and in implementation Processing . .Core Parallelism + Instruction Retry . Protection Keys+ phase . Distributed Switch . .FP Performance + Dyn Energy Mgmt Power. 45nm 7+ 32nm IBM is the leader . Shared L2 . . .Memory bandwidth 2 Thread SMT + Faster . Dynamic LPARs . . Very large + Protection Keys in Processor and (32) . 65nm cache . .130nm, 90nm . Accelerators Server design 180nm,2001 130nm 2004 2007 2010 Binary Compatible & Increased Core Performance © 2012 IBM Corporation POWER7+

POWER7 POWER7+POWER7 45 nm 32 nm

В 2.5 раза больше L3 cache On chip аппаратные ускорители

Physical Design: SMP Fabric • 8 cores with integrated Cache, Memory Core Core A Core Core Controllers, and Accelerators c c • 3 / 4 / 6 / 8 Core options L2 L2 E L2 L2 • 32nm technology n g L3 Cache L3 Cache Features: M M • 2.5X increase in L3 Cache C Power Bus C •eDRAM technology L3 Cache L3 Cache G •Higher Frequencies X L2 L2 L2 L2 • Memory Compression Engine B • Active Memory Expansion with no u Core Core s Core Core processor overhead penalty • Encryption / Cryptography Support SMP Fabric • Random Number Generator • Enhanced Energy / Power Gating • 1/20 LPAR Core Granularity with FW7.6 • 2X SPFP performance

POWER5 POWER5+ POWER6 POWER7 POWER7+ Technology 130nm 90nm 65nm 45nm 32nm Size 389 mm2 245 mm2 341 mm2 567 mm2 567 mm2 Transistors 276 M 276 M 790 M 1.2 B 2.1 B Cores 2 2 2 8 8 1.65 1.9 4 - 5 3 – 4 3.6 – 4.4+ Frequencies GHz GHz GHz GHz GHz 256 KB 256 KB L2 Cache 1.9MB Shared 1.9MB Shared 4MB / Core per Core per Core L3 Cache 36MB 36MB 32MB 4MB / Core 10MB / Core Memory Cntrl 1 1 2 / 1 2 / 1 2 / 1 Architecture Out of Order Out of Order In of Order Out of Order Out of Order LPAR 10 / Core 10 / Core 10 / Core 10 / Core 20 / Core

2.1Млрд транзисторов 567 mm2 5.4Млрд транзисторов 950 mm2 IBM’s eDRAM выигрыш: . Большая плотность: 1/3 площади от 6T SRAM . Меньшее энергопотребление: 1/5 SRAM . Меньше ошибок: в 250x меньше чем в SRAM . Большая производительность

12 © 2012 IBM Corporation POWER7+ процесссоры и архитектура До 40% больше производительности, идеальны для транзакционных и бизнес-аналитических нагрузок Большая масштабируемость и гибкость • 20 Виртуальных машин на ядро • Elastic Capacity on Demand Выше производительность • 10 MB L3 Cache • up to 4.4 GHz POWER7+ processors • Memory Compression Accelerator • Random number generator • Enhanced Single Precision Floating Point performance • Enhanced GX system bus SMP Fabric

Core Core Core Core Улучшенная доступность Acc L2 L2 L2 L2 • Self-healing capability for L3 Cache functions Eng • Dynamic processor fabric bus repair L3 Cache L3 Cache M Power Bus M • Processor re-initialization C C L3 Cache G L3 Cache • X Hardware encryption support for AIX L2 L2 L2 L2 SMP BFabric u Энергоэффективность Core Core s Core Core • Delivering 5x more performance per watt • Enhanced energy / power gating POWER7+ 32 nm

. Возможность расширения 2 5 0 памяти до 125% 125%

2 0 0 . Клиент получает больше памяти используя меньше ресурсов POWER7+ 1 5 0

. Больше циклов процессора 1 0 0 для основных задач

5 0

0 No AME P7 AME P7+ AME

Технологии Выигрыш клиента

. Crypto Offload Accelerators provide . Cryptography can be applied to a broader set of cryptographic engines to relieve the data creating a stronger security ecosystem. P7+ processor from the . Two primary AIX security applications will be performance intensive enabled by default: cryptographic algorithms of AES, • Encrypted File Systems, protecting your SHA, and RSA. data in storage or on backup media • IPsec, protecting your data over the network. • Processor performance remains focused on business applications.

. High quality random numbers . At the core of cryptography is the need for high generated with high performance quality entropy or random numbers. via the RNG offload feature of the . With the POWER7+ RNG offload providing this P7+ processor. entropy, security is enhanced and processor performance remains focused on your business applications.

Power 795

Power 780

Power 775

Power 770

Power 750 Power 720 / 740 IBM PureFlex System

IBM Flex Power 755 System p460 IBM Flex System p260 BladeCenter Power 710 / 730 PS700/PS701/ IBM Flex PS702/PS703/PS704 System p24L

© 2012 IBM Corporation Новые серверы IBM Power на процессорах POWER7+ Power 770 Power 780 •POWER7+ processors • POWER7+ processors •До 48 ядер @ 4.2 GHz • До 64 ядер @ 4.4 GHz •До 64 ядер @ 3.8 GHz • До128 ядер @ 3.7 GHz •12 или 16 core 4U узлы • 16 или 32 core 4U узлы •До 4x узлов в системе • До 4x узлов в системе •Dynamic Platform Optimizer • Dynamic Platform Optimizer

Increased performance and scale  Increased scalability to 128-cores Now up to 20 VMs per-core  Increased performance per-core Increased energy efficiency  Now up to 20 VMs per-core Elastic Capacity on Demand  Increased energy efficiency Enterprise RAS  Built-in Elastic Capacity on Demand  Share resources in Power Systems Pool  Enterprise RAS

Категории Power 770 Power 770+ Преимущество Processor • POWER7 45nm • POWER7+ 32 nm  Faster performance L3 cache • 4 MB per core • 10 MB per core  Faster performance • 64 cores @ 3.3 GHz • 64 cores @ 3.8 GHz Configurations • 48 cores @ 3.7 GHz • 48 cores @ 4.2 GHz  Faster performance • Software memory • Hardware assisted memory Hardware compression compression and AIX file  Faster performance accelerators • Software encryption system encryption • Random number generator Other processor • Enhanced Single Precision  Faster performance features Floating Point performance • Self-healing capability for L3 cache functions RAS • Dynamic processor fabric bus  Better availability repair • Processor re-initialization

Energy • Enhanced energy power gating  More energy efficient • 10 VMs per core • 20 VMs per core Virtualization  More efficient for virtualization • 8 concurrent LPMs • 16 concurrent LPMs  More efficient for virtualization Software • Dynamic Platform Optimizer  Less complex to manage CoD • Easier CoD enablement keys  Less complex to manage • >1,300 direct disk drives • New 1U SSD storage enclosure  Greater performance/density Storage • >3,000 direct attach disk drives  Higher capacity

Category POWER7 780 POWER7+ 780 Benefit

Processor • POWER7 45nm • POWER7+ 32nm Faster performance

L3 Cache • 4MB per core • 10MB per core Faster performance

Faster performance Configurations • 96 cores @ 3.44GHz • 128 cores @ 3.72GHz • 64 core @ 3.92GHz • 64 cores @ 4.42GHz Greater scalability

Hardware • Memory Compression Faster performance Accelerators • AIX file system & IPsec encryption • Random number generator Other Processor • Enhanced GX system buses Faster performance Features • Enhanced single precision floating Point performance • Multi-system pooling of compute resources RAS Better availability • Dynamic processor fabric bus repair Energy • Enhanced energy power gating More energy efficient • 10 VMs per Core • 20 VMs per Core More efficient for Virtualization • 8 Concurrent LPMs • 16 Concurrent LPMs virtualization

More efficient for Software • Dynamic Platform Optimizer virtualization Less complex to manage

• Elastic COD processor and COD memory days built-in Greater flexibility • Flexible COD enablement keys

© 2012 IBM Corporation Новые свойства Power 795 Power 795 • High-end POWER7 server • Up to 256 cores @ 4.0 GHz • Up to 128 cores @ 4.25 GHz • 1 - 8 processor books • Up to 1,000 VMs • Dynamic Platform Optimizer • Active Memory Mirroring

✓ Now up to 16 TB Memory ✓ Now up to 20 VMs per-core ✓ New Integrated Bus & I/O adapters ✓ Built-in Elastic Capacity on Demand ✓ Share resources in a Power Systems Pool ✓ Enterprise RAS

21 18 октября, 2012 © 2012 IBM Corporation Power Systems top all others in SAP Sales and Distribution Standard Application 2-Tier benchmark Simple, single Power 780 delivers over 2X the users and 88% more users per-core than HP DL980 G7 & 46% more users and 191% more users per-core than the Oracle M9000

96 -c ore S Overall A P ow e r 78 0 P

w ith D

57,024 P OWE R 7 + E ® #1 R SAP SD Users & DB 2 10 P

#1 6

16-processor B

#1 12-processor c

#1 8-processor a

4-processor s

Fujitsu RX500 S7 Power 750 HP DL980 G7 Power 780 Sun M9000 Power 780 Power 795 Power 795 Xeon POWER7 Xeon POWER7 SPARC POWER7+ POWER7 POWER7 SQL Server 2008 DB2 9.7 SQL Server 2008 DB2 9.7 Oracle 10g DB2 10 DB2 9.7 DB2 9.7 12,900 users 17,312 users 25,160 users 37,000 users 39,100 users 57,024 users 70,032 users 126,063 users 4/32/64 4/32/128 8/80/160 8/64/256 64/256/512 12/96/384 16/128/512 32/256/1024 # 2012029 # 2011043 # 2011021 # 2010013 # 2008042 # (1) # 2010042 # 2010046 (1) IBM Power 780 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; certification number not available at press time and can be found at sap.com/benchmark . IBM Power System 780, 12 processors / 96 cores / 384 threads, POWER7+ 3.72 GHz, 1536 GB memory, 57,024 SD benchmark users, running AIX® 7.1 and DB2® 10, dialog resp.: 0.98s, line items/hour: 6,234,330, Dialog steps/hour: 18,703,000, SAPS: 311,720, DB time (dialog/ update): 0.009s / 0.014s, CPU utilization: 99%, Certification #: (1). Results valid as of 09/17/2012. Source: http://www.sap.com/benchmark. Additional details on following pages.

more 594 users/core A +88% P

users/per-core D

than HP DL980 G7 R

314 users/core e

r 152 users/core k

Sun M9000 - SPARC HP DL980 G7 - Xeon Power 780 - POWER7+ Oracle 10g SQL Server 2008 DB2 10 39,100 users 25,160 users 57,024 users 64/256/512 8/80/160 12/96/768 # 2008042 # 2011021 # (1) (1) IBM Power 780 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; certification number not available at press time and can be found at sap.com/benchmark . IBM Power System 780, 12 processors / 96 cores / 384 threads, POWER7+ 3.72 GHz, 1536 GB memory, 57,024 SD benchmark users, running AIX® 7.1 and DB2® 10, dialog resp.: 0.98s, line items/hour: 6,234,330, Dialog steps/hour: 18,703,000, SAPS: 311,720, DB time (dialog/ update): 0.009s / 0.014s, CPU utilization: 99%, Certification #: (1). Results valid as of 09/17/2012. Source: http://www.sap.com/benchmark. Additional details on following pages.

 Быстрое проникновение облаков

 Формирование бизнес преимуществ на основе данных и аналитики

 Повышенное внимание к безопасности

 Проникновение социальных сетей и мобильных приложений

 Консолидация и интеграция IT инфраструктуры

Cloud Как перенести инвестиции IT с Только 1 из 5 компаний выделяет расходы обслуживания на инновации и на инновации IT сервисы?

Data Как разобраться с данным и 48% - рост данных в 2012 по сравнению с использовать их для 2011 преимущества над конкурентами? Security Как обезопасить информацию, имеющую значение для вашего 60,000 ежедневных атак на IT инфраструктуры бизнеса и ваших клиентов?

Безопасную облачная инфраструктура позволяет Cloud максимально эффективно доставить бизнес приложения

Критически важные данные,результаты их анализа доступны в Data реальном времени в масштабах всего предприятия

Обеспечение полной безопасности данных, снижение Security рисков и повышения соответствия требованиям государства

high performance analytics, high performance computing & high performance UNIX

IBM Watson LLNL Sequoia  Performance Победитель Jeopardy!Jeopardy! Самый быстрый Супекромпьтер  Scalability  Virtualization  Availability  Security

consolidation 1.7X 80% 16 petaflops 125% more database reduction in floor space LLNL Sequoia more SAP SD Users performance per core reduction in energy than best Intel 27 reduction in licensed cores © 2012 IBM Corporation  Power Systems, PowerVM, AIX, IBM I, Linux, IBM SW  Технологии и решения, которые:

. Разработаны вместе 80% 70% . Произведены вместе 60% 50% . Протестированы вместе 40% 30% . Поставляются вместе 20% 10% . Поддерживаются вместе 0%

IBM Smart Analytics IBM Lotus Domino IBM Rational System Consolidation Developer on Power for Power

SAP on IBM DB2 and Power Systems

IBM DB2 pureScale IBM Systems Director IBM WebSphere Editions Application Server Parallelization

Вопросы?

[email protected]

(495)-775-8800 ext 2904

The IBM benchmarks results shown herein were derived using particular, well configured, development-level and generally-available computer systems. Buyers should consult other sources of information to evaluate the performance of systems they are considering buying and should consider conducting application oriented testing. For additional information about the benchmarks, values and systems tested, contact your local IBM office or IBM authorized reseller or access the Web site of the benchmark consortium or benchmark vendor.

IBM benchmark results can be found in the IBM Power Systems Performance Report at http://www.ibm.com/systems/p/hardware/system_perf.html .

For a definition/explanation of each benchmark and the full list of detailed results, visit the Web site of the benchmark consortium or benchmark vendor.

TPC http://www.tpc.org SPEC http://www.spec.org LINPACK http://www.netlib.org/benchmark/performance.pdf Pro/E http://www.proe.com GPC http://www.spec.org/gpc VolanoMark http://www.volano.com STREAM http://www.cs.virginia.edu/stream/ SAP http://www.sap.com/benchmark/ Oracle, Siebel, PeopleSoft http://www.oracle.com/apps_benchmark/ Baan http://www.ssaglobal.com Fluent http://www.fluent.com/software/fluent/index.htm TOP500 Supercomputers http://www.top500.org/ Ideas International http://www.ideasinternational.com/benchmark/bench.html Storage Performance Council http://www.storageperformance.org/results

Revised December 2, 2010

IBM benchmark results can be found in the IBM Power Systems Performance Report at http://www.ibm.com/systems/p/hardware/system_perf.html .

All performance measurements were made with AIX or AIX 5L operating systems unless otherwise indicated to have used Linux. For new and upgraded systems, the latest versions of AIX were used. All other systems used previous versions of AIX. The SPEC CPU2006, LINPACK, and Technical Computing benchmarks were compiled using IBM's high performance C, C++, and FORTRAN compilers for AIX 5L and Linux. For new and upgraded systems, the latest versions of these compilers were used: XL C for AIX v11.1, XL C/C++ for AIX v11.1, XL FORTRAN for AIX v13.1, XL C/C++ for Linux v11.1, and XL FORTRAN for Linux v13.1. Linpack HPC (Highly Parallel Computing) used the current versions of the IBM Engineering and Scientific Subroutine Library (ESSL). For Power7 systems, IBM Engineering and Scientific Subroutine Library (ESSL) for AIX Version 5.1 and IBM Engineering and Scientific Subroutine Library (ESSL) for Linux Version 5.1 were used.

For a definition/explanation of each benchmark and the full list of detailed results, visit the Web site of the benchmark consortium or benchmark vendor. SPEC http://www.spec.org LINPACK http://www.netlib.org/benchmark/performance.pdf Pro/E http://www.proe.com GPC http://www.spec.org/gpc STREAM http://www.cs.virginia.edu/stream/ Fluent http://www.fluent.com/software/fluent/index.htm TOP500 Supercomputers http://www.top500.org/ AMBER http://amber.scripps.edu/ FLUENT http://www.fluent.com/software/fluent/fl5bench/index.htm GAMESS http://www.msg.chem.iastate.edu/gamess GAUSSIAN http://www.gaussian.com ANSYS http://www.ansys.com/services/hardware-support-db.htm Click on the "Benchmarks" icon on the left hand side frame to expand. Click on "Benchmark Results in a Table" icon for benchmark results. ABAQUS http://www.simulia.com/support/v68/v68_performance.php ECLIPSE http://www.sis.slb.com/content/software/simulation/index.asp?seg=geoquest& MM5 http://www.mmm.ucar.edu/mm5/ MSC.NASTRAN http://www.mscsoftware.com/support/prod%5Fsupport/nastran/performance/v04_sngl.cfm STAR-CD www.cd-adapco.com/products/STAR-CD/performance/320/index/html NAMD http://www.ks.uiuc.edu/Research/namd Revised December 2, 2010 HMMER http://hmmer.janelia.org/ http://powerdev.osuosl.org/project/hmmerAltivecGen2mod 32 18 октября, 2012 © 2012 IBM Corporation Notes on performance estimates rPerf for AIX rPerf (Relative Performance) is an estimate of commercial processing performance relative to other IBM UNIX systems. It is derived from an IBM analytical model which uses characteristics from IBM internal workloads, TPC and SPEC benchmarks. The rPerf model is not intended to represent any specific public benchmark results and should not be reasonably used in that way. The model simulates some of the system operations such as CPU, cache and memory. However, the model does not simulate disk or network I/O operations.

. rPerf estimates are calculated based on systems with the latest levels of AIX and other pertinent software at the time of system announcement. Actual performance will vary based on application and configuration specifics. The IBM eServer pSeries 640 is the baseline reference system and has a value of 1.0. Although rPerf may be used to approximate relative IBM UNIX commercial processing performance, actual system performance may vary and is dependent upon many factors including system hardware configuration and software design and configuration. Note that the rPerf methodology used for the POWER6 systems is identical to that used for the POWER5 systems. Variations in incremental system performance may be observed in commercial workloads due to changes in the underlying system architecture.

All performance estimates are provided "AS IS" and no warranties or guarantees are expressed or implied by IBM. Buyers should consult other sources of information, including system benchmarks, and application sizing guides to evaluate the performance of a system they are considering buying. For additional information about rPerf, contact your local IBM office or IBM authorized reseller.

======

CPW for IBM i

Commercial Processing Workload (CPW) is a relative measure of performance of processors running the IBM i operating system. Performance in customer environments may vary. The value is based on maximum configurations. More performance information is available in the Performance Capabilities Reference at: www.ibm.com/systems/i/solutions/perfmgmt/resource.html

Revised April 2, 2007