IBM Power microprocessors explained

IBM Power microprocessors (originally POWER prior to Power10) are designed and sold by IBM for servers and supercomputers.[1] The name "POWER" was originally presented as an acronym for "Performance Optimization With Enhanced RISC". The Power line of microprocessors has been used in IBM's RS/6000, AS/400, pSeries, iSeries, System p, System i, and Power Systems lines of servers and supercomputers. They have also been used in data storage devices and workstations by IBM and by other server manufacturers like Bull and Hitachi.

The Power family was originally developed in the late 1980s, and remains under active development. In the beginning, they implemented the POWER instruction set architecture (ISA), which evolved into PowerPC and later into Power ISA. In August 2019, IBM announced it would open source the Power ISA.[2] As part of the move, it was also announced that administration of the OpenPOWER Foundation will now be handled by the Linux Foundation.

History

Early developments

The 801 research project

See main article: IBM 801. In 1974 IBM started a project to build a telephone switching computer that required, for the time, immense computational power. Since the application was comparably simple, this machine would need only to perform I/O, branches, add register-register, move data between registers and memory, and would have no need for special instructions to perform heavy arithmetic. This simple design philosophy, whereby each step of a complex operation is specified explicitly by one machine instruction, and all instructions are required to complete in the same constant time, would later come to be known as RISC. When the telephone switch project was canceled, IBM retained the design for the general purpose processor and named it 801 after building #801 at Thomas J. Watson Research Center.

The Cheetah project

By 1982 IBM continued to explore the superscalar limits of the 801 design by using multiple execution units to improve performance to determine if a RISC machine could maintain multiple instructions per cycle. Many changes were made to the 801 design to allow for multiple execution units and the Cheetah processor has separate units for branch prediction, fixed-point, and floating-point execution. By 1984 CMOS was chosen because it allows improved circuit integration and transistor-logic performance.

The America project

In 1985, research on a second-generation RISC architecture started at the IBM Thomas J. Watson Research Center, producing the "AMERICA architecture". In 1986, IBM Austin started developing the RS/6000 series computers based on that architecture. This was to become the first POWER processors using the first POWER ISA.

POWER

See main article: POWER1. The first IBM computers to incorporate the POWER ISA are the RISC System/6000 or RS/6000 series. They were released in February 1990. These RS/6000 computers were divided into two classes, POWERstation workstations and POWERserver servers. The first RS/6000 CPU has 2 configurations, called the "RIOS-1" and "RIOS.9" (or more commonly the POWER1 CPU). A RIOS-1 configuration has a total of 10 discrete chips: an instruction cache chip, fixed-point chip, floating-point chip, 4 data L1 cache chips, storage control chip, input/output chips, and a clock chip. The lower cost RIOS.9 configuration has 8 discrete chips: an instruction cache chip, fixed-point chip, floating-point chip, 2 data cache chips, storage control chip, input/output chip, and a clock chip.

The POWER1 is the first microprocessor that used register renaming and out-of-order execution. A simplified and less powerful version of the 10 chip RIOS-1 was made in 1992, for lower-end RS/6000s. It uses only one chip and is called RISC Single Chip or RSC.

POWER1 processors

POWER2

See main article: POWER2. IBM started the POWER2 processor effort as a successor to the POWER1. By adding a second fixed-point unit, a second powerful floating point unit, and other performance enhancements and new instructions to the design, the POWER2 ISA had leadership performance when it was announced in November 1993. The POWER2 was a multi-chip design, but IBM also made a single chip design of it, called the POWER2 Super Chip or P2SC that went into high performance servers and supercomputers. At the time of its introduction in 1996, the P2SC was the largest processor with the highest transistor count in the industry and was a leader in floating point operations.

POWER2 processors

PowerPC

See main article: PowerPC. In 1991, Apple looked for a future alternative to the CISC-based Motorola 68000 series platform, and Motorola experimented with a RISC platform of its own, the 88000. IBM joined the discussion and the three founded the AIM alliance to build the PowerPC ISA, heavily based on the POWER ISA, but with additions from both Apple and Motorola. It was to be a complete 32/64 bit RISC architecture, and to range from very low end embedded microcontrollers to the very high end supercomputer and server applications.

After two years of development, the resulting PowerPC ISA was introduced in 1993. A modified version of the RSC architecture, PowerPC added single-precision floating point instructions and general register-to-register multiply and divide instructions, and removed some POWER features. It also added a 64-bit version of the ISA and support for SMP.

The Amazon project

See main article: IBM RS64. In 1990, IBM wanted to merge the low end server and mid range server architectures, the RS/6000 RISC ISA and AS/400 CISC ISA into one common RISC ISA that could host both IBM's AIX and OS/400 operating systems. The existing POWER and the upcoming PowerPC ISAs were deemed unsuitable by the AS/400 team so an extension to the 64-bit PowerPC instruction set was developed called PowerPC AS for Advances Series or Amazon Series. Later, additions from the RS/6000 team and AIM Alliance PowerPC were included, and by 2001, with the introduction of POWER4, they were all joined into one instruction set architecture: the PowerPC v.2.0.

POWER3

See main article: POWER3. The POWER3 began as PowerPC 630, a successor of the commercially unsuccessful PowerPC 620. It uses a combination of the POWER2 ISA and the 32/64-bit PowerPC ISA set with support for SMP and single-chip implementation. It was used to great extent in IBM's RS/6000 computers, and the second generation version, the POWER3-II, is the first commercially available processor from IBM using copper interconnects. The POWER3 is the last processor to use a POWER instruction set, and all subsequent models use the PowerPC instruction sets.

POWER3 processors

POWER4

See main article: POWER4. The POWER4 merged the 32/64 bit PowerPC instruction set and the 64-bit PowerPC AS instruction set from the Amazon project to the new PowerPC v.2.0 specification, unifying IBM's RS/6000 and AS/400 families of computers. Besides the unification of the different platforms, POWER4 was also designed to reach very high frequencies and have large on-die L2 caches. It is the first commercially available multi-core processor and came in single-die versions as well as in four-chip multi-chip modules. In 2002, IBM also made a cost- and feature-reduced version of the POWER4 called PowerPC 970 by Apple's request.

POWER4 processors

POWER5

See main article: POWER5. The POWER5 processors built on the popular POWER4 and incorporated simultaneous multithreading into the design, a technology pioneered in the PowerPC AS based RS64-III processor, and on-die memory controllers. It was designed for multiprocessing on a massive scale and came in multi-chip modules with onboard large L3 cache chips.

POWER5 processors

Power ISA

See main article: Power ISA. A joint organization was founded in 2004 called Power.org with the mission to unify and coordinate future development of the PowerPC specifications. By then, the PowerPC specification was fragmented since Freescale (née Motorola) and IBM had taken different paths in their respective development of it. Freescale had prioritized 32-bit embedded applications and IBM high-end servers and supercomputers. There was also a collection of licensees of the specification like AMCC, Synopsys, Sony, Microsoft, P.A. Semi, CRAY, and Xilinx that needed coordination. The joint effort was not only to streamline development of the technology but also to streamline marketing.

The new instruction set architecture was called Power ISA and merged the PowerPC v.2.02 from the POWER5 with the PowerPC Book E specification from Freescale as well as some related technologies like the Vector-Media Extensions known under the brand name AltiVec (also called VMX by IBM) and hardware virtualization. This new ISA was called 'Power ISA v.2.03 and POWER6 was the first high end processor from IBM to use it. Older POWER and PowerPC specifications did not make the cut and those instruction sets were henceforth deprecated for good. There is no active development on any processor type today that uses these older instruction sets.

POWER6

See main article: POWER6. POWER6 was the fruit of the ambitious eCLipz Project, joining the I (AS/400), P (RS/6000) and Z (Mainframe) instruction sets under one common platform. I and P was already joined with the POWER4, but the eCLipz effort failed to include the CISC based z/Architecture and where the z10 processor became POWER6's eCLipz sibling., a separate line of processors implementing z/Architecture continue to be developed by IBM, with the latest being the IBM Telum.[3]

Because of eCLipz, the POWER6 is an unusual design as it aimed for very high frequencies and sacrificed out-of-order execution, something that has been a feature for POWER and PowerPC processors since their inception. POWER6 also introduced the decimal floating point unit to the Power ISA, something it shares with z/Architecture.

With the POWER6, in 2008 IBM merged the former System p and System i server and workstation families into one family called Power Systems. Power Systems machines can run different operating systems like AIX, Linux, and IBM i.

POWER6 processors

POWER7

See main article: POWER7. The POWER7 symmetric multiprocessor design was a substantial evolution from the POWER6 design, focusing more on power efficiency through multiple cores, simultaneous multithreading (SMT), out-of-order execution and large on-die eDRAM L3 caches. The eight-core chip could execute 32 threads in parallel, and has a mode in which it could disable cores to reach higher frequencies for the ones that are left. It uses a new high-performance floating point unit called VSX that merges the functionality of the traditional FPU with AltiVec. Even while the POWER7 run at lower frequencies than POWER6, each POWER7 core performs faster than its POWER6 counterpart.

POWER7 processors

POWER8

See main article: POWER8. POWER8 is a 4 GHz, 12 core processor with 8 hardware threads per core for a total of 96 threads of parallel execution. It uses 96 MB of eDRAM L3 cache on chip and 128 MB off-chip L4 cache and a new extension bus called CAPI that runs on top of PCIe, replacing the older GX bus. The CAPI bus can be used to attach dedicated off-chip accelerator chips such as GPUs, ASICs and FPGAs. IBM states that it is two to three times as fast as its predecessor, the POWER7.

It was first built on a 22 nanometer process in 2014.[4] [5] [6] In December 2012, IBM began submitting patches to the 3.8 version of the Linux kernel, to support new POWER8 features including the VSX-2 instructions.[7]

POWER9

See main article: POWER9. IBM spent much time designing the POWER9 processor according to William Starke, a systems architect for the POWER8 processor.[8] The POWER9 is the first to incorporate elements of the Power ISA version 3.0 that was released in December 2015, including the VSX-3 instructions, and also incorporates support for Nvidia's NVLink bus technology.[9] [10]

The United States Department of Energy together with Oak Ridge National Laboratory and Lawrence Livermore National Laboratory contracted IBM and Nvidia to build two supercomputers, the Sierra and the Summit, based on POWER9 processors coupled with Nvidia's Volta GPUs. The Sierra went online in 2017 and the Summit in 2018.[11] [12] [13]

POWER9, which was launched in 2017, is manufactured using a 14 nm FinFET process, and comes in four versions, two 24 core SMT4 versions intended to use PowerNV for scale up and scale out applications, and two 12 core SMT8 versions intended to use PowerVM for scale-up and scale-out applications. Possibly there will be more versions in the future since the POWER9 architecture is open for licensing and modification by the OpenPOWER Foundation members.[14]

Power10

See main article: Power10. Power10 is a CPU introduced in September 2021. It is built on a 7 nm technology.[15] [16]

Devices

Name Image ISA Bits Cores Fab Transistors Die size L1 L2 L3 Clock Package Introduced
RIOS-1POWER32 bits11.0 μm6.9 M1284 mm2KB I
64 KB D
n/an/a20–30 MHz10 chips
in CPGA
on PCB
1990
RIOS.9POWER32 bits11.0 μm6.9 M8 KB I
32 KB D
n/an/a20–30 MHz8 chips
in CPGA
on PCB
1990
POWER1+POWER32 bits16.9 M8 KB I
64 KB D
n/an/a25–41.6 MHz8 chips
in CPGA
on PCB
1991
POWER1++POWER32 bits16.9 M8 KB I
64 KB D
n/an/a25–62.5 MHz8 chips
in CPGA
on PCB
1992
RSCPOWER32 bits10.8 μm1 M226.5 mm28 KB
unified
n/an/a33–45 MHz201 pin CPGA1992
POWER2POWER232 bits10.72 μm23 M1042.5 mm2
819 mm2
32 KB I
128 - 265 KB D
n/an/a55–71.5 MHz6–8 dies
on ceramic 734 pin MCM
1993
POWER2+POWER232 bits10.72 μm23 M819 mm232 KB I
64 - 128 KB D
0.5 - 2 MB
external
n/a55–71.5 MHz6 chips
in CBGA
on PCB
1994
P2SCPOWER232 bits10.29 μm15 M335 mm232 KB I
128 KB D
n/an/a120–135 MHzCCGA1996
P2SC+POWER232 bits10.25 μm15 M256 mm232 KB I
128 KB D
n/an/a160 MHzCCGA1997
RAD6000POWER32 bits10.5 μm1.1 M8 KB unifiedn/an/a20–33 MHzRad hard1997
POWER3POWER2
PowerPC 1.1
64 bits10.35 μm15 M270 mm232 KB I
64 KB D
1–16 MB
external
n/a200–222 MHz1088 pin CLGA1998
POWER3-IIPOWER2
PowerPC 1.1
64 bits10.25 μm Cu23 M170 mm232 KB I
64 KB D
1–16 MB
external
n/a333–450 MHz1088 pin CLGA1999
POWER4PowerPC 2.00
PowerPC-AS
64 bits2180 nm174 M412 mm264 KB I
32 KB D
per core
1.41 MB
per core
32 MB
external
1–1.3 GHz1024 pin CLGA
ceramic MCM
2001
POWER4+PowerPC 2.01
PowerPC-AS
64 bits2130 nm184 M267 mm264 KB I
32 KB D
per core
1.41 MB
per chip
32 MB
external
1.2–1.9 GHz1024 pin CLGA
ceramic MCM
2002
POWER5PowerPC 2.02
Power ISA 2.03
64 bits2130 nm276 M389 mm232 KB I
32 KB D
per core
1.875 MB
per chip
32 MB
external
1.5–1.9 GHzceramic DCM
ceramic MCM
2004
POWER5+PowerPC 2.02
Power ISA 2.03
64 bits290 nm276 M243 mm232 KB I
32 KB D
per core
1.875 MB
per chip
32 MB
external
1.5–2.3 GHzceramic DCM
ceramic QCM
ceramic MCM
2005
POWER6Power ISA 2.0364 bits265 nm790 M341 mm264 KB I
64 KB D
per core
4 MB
per core
32 MB
external
3.6–5 GHzCLGA
OLGA
2007
POWER6+Power ISA 2.0364 bits265 nm790 M341 mm264 KB I
64 KB D
per core
4 MB
per core
32 MB
external
3.6–5 GHzCLGA
OLGA
2009
POWER7Power ISA 2.0664 bits845 nm1.2 B567 mm232 KB I
32 KB D
per core
256 KB
per core
32 MB
per chip
2.4–4.25 GHzCLGA
OLGA
organic QCM
2010
POWER7+Power ISA 2.0664 bits832 nm2.1 B567 mm232 KB I
32 KB D
per core
256 KB
per core
80 MB
per chip
2.4–4.4 GHzOLGA
organic DCM
2012
POWER8Power ISA 2.0764 bits6
12
22 nm??
4.2 B
362 mm2
649 mm2
32 KB I
64 KB D
per core
512 KB
per core
48 MB
96 MB
per chip
2.75–4.2 GHzOLGA DCM
OLGA SCM
2014
POWER8
with NVLink
Power ISA 2.0764 bits1222 nm4.2 B659 mm232 KB I
64 KB D
per core
512 KB
per core
48 MB
96 MB
per chip
3.26 GHzOLGA SCM2016
POWER9 SUPower ISA 3.064 bits12
24
14 nm8 B32 KB I
64 KB D
per core
512 KB
per core
120 MB
per chip
~4 GHz2017
Power10Power ISA 3.164 bits15
30
7 nm18 B602 mm248 KB I
32 KB D
per core
2 MB
per core
120 MB
per chip
3.5 to 4GHzOLGA SCM
OLGA DCM
2021
Name Image ISA Bits Cores Fab Transistors Die size L1 L2 L3 Clock Package Introduced

See also

External links

Notes and References

  1. Web site: IBM Power10. IBM. December 29, 2021.
  2. Web site: Morgan . Timothy . Big Blue Open Sources Power Chip Instruction Set . nextplatform.com . August 20, 2019 . Stackhouse Publishing Inc . August 20, 2019.
  3. Web site: 2022-04-29 . IBM's New Telum Chip Reboots the Mainframe . 2022-05-05 . IEEE Spectrum . en.
  4. http://www.itjungle.com/tfh/tfh070912-story01.html The Four Hundred-Some Insight Into Those Future Power7+ Processors
  5. http://www.profi-ag.de/wps/wcm/connect/0750ad004e9016be9fdc9f7e9d9936e7/Praesentation-PROFIAG-IBMPower7.pdf IBM Power Systems 2013.
  6. Web site: IBM POWER8 - Announce / Availability Plans . August 11, 2018 . https://web.archive.org/web/20140524004044/http://komplex-it.dk/media/128719/ibm_power8.pdf . May 24, 2014 . dead .
  7. http://lkml.indiana.edu/hypermail/linux/kernel/1212.1/03373.html Linux-Kernel Archive: [git pull] Please pull powerpc.git next branch]
  8. https://www.theregister.co.uk/2013/08/27/ibm_power8_server_chip/ You won't find this in your phone: A 4 GHz 12-core Power8 for badass boxes
  9. https://sourceware.org/ml/binutils/2015-11/msg00071.html Add full Power ISA 3.0 / POWER9 binutils support
  10. http://wccftech.com/nvidia-volta-gpus-ibm-power9-cpus-deliver-300-petaflops-performance-2017-summit-sierra-supercomputers/ NVIDIA Volta GPUs and IBM Power9 CPUs To Deliver Up To 300 PetaFlops of Performance in 2017 With Summit and Sierra Supercomputers
  11. http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers NVIDIA Volta, IBM POWER9 Land Contracts For New US Government Supercomputers
  12. https://www.olcf.ornl.gov/summit/ ORNL Summit home page
  13. https://www.llnl.gov/news/next-generation-supercomputer-coming-lab Lawrence Livermore signs contract with IBM
  14. https://www.theregister.co.uk/2016/04/07/open_power_summit_power9/ Power9: Google gives Intel a chip-flip migraine, IBM tries to lures big biz
  15. https://www.anandtech.com/show/13740/ibm-samsungs-7nm-euv-power-z-cpu IBM to use Samsung 7nm EUV for Next-Gen POWER and z CPUs
  16. https://www.nextplatform.com/2015/08/10/ibm-roadmap-extends-power-chips-to-2020-and-beyond/ IBM Roadmap Extends Power Chips To 2020 And Beyond