Documente Academic
Documente Profesional
Documente Cultură
Joe Armstrong
jdarmstr@us.ibm.com
POWER9
Extreme Analytics
Optimization
POWER8 Extreme Big Data
More Cores Optimization
SMT+++ On-chip accelerators
POWER7/7+ Reliability ++
FPGA Support
45/32 nm Transactional Memory
POWER6/6+ Eight Cores
PCIe Acceleration
On-Chip eDRAM
65/65 nm Power-Optimized Cores
POWER5/5+ Memory Subsystem ++ 200+ systems in test
130/90 nm Dual Core SMT++
High Frequencies Reliability +
Dual Core Virtualization + VSM & VSX
Enhanced Scaling Memory Subsystem + Protection Keys+
SMT Altivec
Distributed Switch + Instruction Retry
Core Parallelism + Dynamic Energy Mgmt
FP Performance + SMT +
Memory Bandwidth + Protection Keys
Virtualization
Technology
22nm SOI, eDRAM, 15 ML 650mm2
Caches
Cores • 512 KB SRAM L2 / core
• 12 cores (SMT8) Core Core Core Core Core Core • 96 MB eDRAM shared L3
Accelerators
SMP Links
• 8 dispatch, 10 issue, • Up to 128 MB eDRAM L4
16 exec pipe L2 L2 L2 L2 L2 L2 (off-chip)
• 2X internal data flows/queues 8M L3
Regio Memory
• Enhanced prefetching n
• Up to 230 GB/s
• 64K data cache, L3
Mem. Cache & Chip Interconnect
Ctrl. Mem. Ctrl. sustained bandwidth
32K instruction cache
Bus Interfaces
Accelerators PCIe
SMP Links
L2 L2 L2 L2 L2 L2 • Durable open memory attach
• Crypto & memory expansion interface
• Transactional Memory • Integrated PCIe Gen3
Core Core Core Core Core Core
• VMM assist • SMP Interconnect
• Data Move / VM Mobility • CAPI (Coherent Accelerator
Processor Interface)
Energy Management
• On-chip Power Management Micro-controller
• Integrated Per-core VRM
• Critical Path Monitors
4
POWER8 Core
Power
780+ Power
Power 795
770+
Power
750+ / 760+
Power
720+/740+
Power
710+/730+
p260+
p24L
Power
780+ Power
Power 795
770+
Power
750+ / 760+
Power
720+/740+
Power
710+/730+
p260+
p24L
Power Systems
S824
2-socket, 4U
Up to 24 cores
Power 1 TB memory
11 PCIe Gen 3
Power Systems S814 AIX, IBM i, Linux
1-socket, 4U CAPI support (4)
Power Systems S822 Up to 8 cores PowerVM
Systems 2-socket, 2U 512 GB memory
Power Systems Up to 20 cores 7 PCIe Gen 3
S822L 1 TB memory AIX, IBM i, Linux
S812L 2-socket, 2U 9 PCIe Gen 3 CAPI support (2)
POWER8 processor AIX & Linux PowerVM
1-socket, 2U Up to 24 cores
POWER8 processor
CAPI support (4)
1 TB memory
Linux only 9 PCI Gen3 slot PowerVM
CAPI support (2) Linux only
CAPI support (4)
PowerVM & PowerKVM
Power S812L: Announced in April: have now shipped across Geos between
7/20 and 9/10
Doubling the memory capacity to 2 TB in the S824 ( through RPQ 8A2232 ) GA: Dec
128GB DIMMS will either be to have 1 TB or 2 TB configurations, no mix and match of DIMM sizings.
If buy 1 TB now can MES upgrade to different add’l DIMMS later
12
© 2014 International Business Machines Corporation 12
Nvidia K40 GPU
Systems
Up to 2 K40 GPU in S824L
GPU Spec
Kepler-2 architecture GPU
ASIC: GK110B
PCIe interface
PCIe Gen3 x16
Full length / double wide PCIe form factor
Plugs in using existing double wide cassette
Power
235W Max power draw :75W via PCIe slot
plus 160W via 8-pin Aux. cable.
OS support
Ubuntu 14.10 or later
http://www.nvidia.com/object/tesla-servers.html
13
© 2014 International Business Machines Corporation 13
POWER S824 More Memory
RPQ announce planned mid October
2X max memory GA in December
Up to 2 TB RPQ Number 8A2232
Rules:
a) No Mixing of DIMMs sizes. ( i.e when order 128GB DIMM, no other size ( 64GB,
32GB, 16GB ) can be mixed with )
b) Only qty 8 or qty 16 DIMMs ( no other qty accepted, so when order 128GB DIMM,
customer can either order them by quantity of 8 = 1TB or quantity of 16=2TB )
c) No MES Upgrade provided for installed systems.
d) Only supported on Models S824, no other Scale-out servers.
14
© 2014 International Business Machines Corporation 14
NEBS: Differentiated Value for Telecommunications Clients
Designed for clients that require hardened infrastructures because of the industries they serve
or data center environments where the equipment is located.
Carrier-grade platforms for NGN infrastructure and application deployment, designed for
extreme shock, vibration and thermal conditions which exceed normal datacenter design
standards.
Announcement Date: October 6
General available Date : October 31st
Eligible model : Power S822L and S822
Certifications: NEBS Level-3 and ETSI
15
© 2014 International Business Machines Corporation 15
Power Scale-out Servers
Power Systems
S824
Power 2-socket, 4U
Up to 24 cores
Systems 1 TB memory
Power S824L 11 PCIe Gen 3
2-socket, 4U AIX, IBM i, Linux
Power Systems S814 CAPI support (4)
Power 1-socket, 4U
Up to 24 cores
PowerVM
Systems S822 Linux
Systems Up to 8 cores NVIDIA GPU
Power Systems 2-socket, 2U 512 GB memory CAPI support(2)
S822L Up to 20 cores 7 PCIe Gen 3
S812L 2-socket, 2U 1 TB memory AIX, IBM i, Linux
POWER8 processor 9 PCIe Gen 3
1-socket, 2U
CAPI support (2)
Up to 24 cores
POWER8 processor
AIX & Linux PowerVM
1 TB memory
Linux only 9 PCI Gen3 slot CAPI support (4)
CAPI support (2) Linux only PowerVM
CAPI support (4)
PowerVM & PowerKVM
POWER8
Was just Statement of direction in April 2014 POWER8
PCIe Gen3
Transport for encapsulated messages
FPGA or ASIC
Application
Attach flash memory to POWER8
Read/Write
Syscall via CAPI coherent Attach
FileSystem
strategy() iodone() 20K Application
Instructions Posix Async aio_read()
LVM I/O Style API aio_write()
strategy() iodone()
User Library
Disk & Adapter DD < 500 Shared
Pin buffers, Interrupt, Instructions Memory
Work Queue
Translate, unmap,
Map DMA, unpin,Iodone
Start I/O scheduling
The Solution: POWER8 + CAPI FLASH as RAM The POWER8 + CAPI Flash as RAM Advantage:
- Up to 40 TB in 4U • New FLASH as RAM for Redis in-memory apps
WWW
• Provides means for large FLASH exploitation
• Lower cost memory, greater workload density
Power S822L/S812L • Dramatically reduce costs to deliver services
Ubuntu 14.10 • Can be offered as a cloud-based service or as
an on-premise solution for enterprises
4U
Expanded
Must select one Base Split Function *
12 SFF SAS bays 6+6 SFF SAS bays 18 SFF SAS bays
1 SAS controller 2 SAS controllers Dual SAS controllers
4U server No write cache No write cache 7.2GB cache
RAID-0,1,5,6,10 RAID-0,1,5,6,10 RAID-0,1,5,6,10
DVD bay DVD bay DVD bay
External SAS ports
8-bay SSD cage
Easy Tier function
Sweet SSD enhancement for
AIX / Linux / VIOS
S824 & S822
Using Easy Tier function
Intelligent SAS controller in backplane places data with high read activity
and low write activity on the drive to allow many years of service without
replacement. (replacement covered by IBM service agreement)
List price qty 3 177GB SSD for RAID-5T2 $ 3,930 63% lower
List price qty 3 387GB SSD for RAID-5T2 $10,764
Prices are USA list prices and subject to change. Reseller prices can vary.
Now on POWER8 servers most disk shipped with 528 byte sectors
Provides additional level of protection to client
Newer generation SAS adapters/controllers get essentially the same
performance
Save clients time from reformatting drives to 528-byte sectors
© 2014 International Business Machines Corporation IBM & Business Partner Confidential until Announcement 29
Power E870
Power E880
Power
780+ Power
Power 795
770+
Power
750+ / 760+
Power
720+/740+
Power
710+/730+
p260+
p24L
No Primary Node
Midplane
Service Processors
Clocks
Oscillators
Large Memory
E870
795 E880
19” Rack
Modular design
770 Up to 4 CEC drawers/nodes
780 PCIe slots in the nodes
19” Rack
Great memory
Up to 16TB (E880) or 4TB (E870)
Faster memory - 1600 MHz
Up to 85.3 GB / Core
Fans
4 POWER8 SCMs
Power Supplies
5U Enclosure
No integrated SAS bays or SAS controller in node
No integrated DVD bay or DVD controller in node
No integrated Ethernet port in node
No tape bay in node
Power E870: IBM plans to enhance the Power E870 with greater I/O and
memory capacity and flexibility with:
Support for concurrent maintenance on the Power E870 I/O Expansion
Drawer by enabling hot add and repair capabilities
The ability to scale up to 4TBs of memory per Power E870 processor node
E870
2U drawer
Ops panel
Rear View
Redundant Hot plug
Redundant Power Clock Battery Redundant Power
Optional DVD
#EU13 1.6 meter USB cable #EBK4
Rear of sys cntrl unitK has USB connector for cabling ease
Node
7U in 19” rack
Node
Node
Node
Node
Node
17U in 19” rack
Node
Node
Node
Node
22U in 19” rack
Service Processor
(FSP) Cables
Clock Cables
SMP Cables
11 11
10 FSP / Clk Sys Cntrl Unit 10
9 B B
9
5u space reserved for
5 E E E E 5
8
e
a 8 N N N N 8
e
a 8 System node
7 r r System node 7
p 9 0 0 0 0 9 p
Drw 2
6 a a 6
5 w 9 B J J B 9 w 5 2u space reserved for PDU’s
4 4
PDU PDU
3 3
PDU PDU
2 2
1 2u reserved for 1 2u reserved for
cable handling cable handling
Add 8-inch rack extender (#ERG0) to rear of 42U rack. Provides more
room for cables.
Generally recommended, but not required
STRONGLY urged if there are a lot of I/O cables
Strongly urge leaving 2U space open at top and/or bottom for cable
egress
E870
32-core 4.02 GHz 359,000
64-core 4.02 GHz 711,000
SMT4 would
be somewhat
E880 lower
“L4 cache”
DDR Interfaces
16MB
Scheduler & POWER8
Memory
Management Cache Link
Intelligence Moved into Memory
• Previously on POWER7+ chip onto buffer
Processor Interface
• High speed interface
Performance Value
Prices are USA list prices and subject to change. Reseller prices may vary.
POWER7+ POWER8
DDR2 1066MHz DDR3 1600 MHz
Feature GB per $ / GB Feature GB per $ / GB
code feature 100% activated code feature 100% activated
#EM40 32 GB $203 -- -- -- No compare
Prices are USA list prices and subject to change. Reseller prices may vary.
Prices calculated using static activations, not mobile activations.
POWER8 SCM
One CDIMM card One CDIMM card
POWER7
POWER6
POWER5
POWER8 E870
POWER7+ 770
POWER7 770
POWER6 570
GB/Sec
DCM SCM
x16 x8 x16
x16 x16
x8
Front view
Feat #EMX0
12 PCIe Gen3 slots
4U drawer
+ + +
0 1 1
0 1
2 2
2
3
4
0 or 2 PCIe Gen3 I/O
Drawers in 2014
0, 2 or 4 PCIe Gen3 I/O Drawers in 2014
(max 2 per node)
---------------
Net +10 additional slots to the system
with each PCIe3 I/O Drawer
Sample quantities of PCIe PCIe slots per PCIe slots per PCIe slots per 4-node
drawers on the server 1-node server 2-node server server (2015)
0 8 16 32
2 28 36 52
24 drwr + 4 node 24 drwr + 12 node 24 drwr + 28 node
4 48 (SOD) 56 72
48 drwr + 0 node 48 drwr + 8 node 48 drwr + 24 node
8 n/a 96 (SOD) 112
96 drwr + 0 node 96 drwr + 16 node
POWER8 PCIe-attached Gen3 I/O drawer has two fan-out modules and each fan-
out module has 32GB/s
A per slot average of 5+GB/s
#EMX0
One #5877
system
system system node
node node
system system
node node
PCIe I/O
Drwr
PCIe I/O
Drwr PCIe I/O
PCIe I/O Drwr
Drwr
PCIe I/O
Drwr
Notes:
With two system nodes it is a good practice (but not required) to
attach the two fan-out modules in one I/O drawer to different
system nodes. Combined with placing redundant PCIe adapters
in different fan-out modules, system availability is enhanced.
PCIe I/O drawer can be in the same or different rack as the
system nodes. If large numbers of I/O cables are attached to
PCIe adapters, it’s nice to have the I/O drawer in a different rack
for cable management ease
System control unit not shown for visual simplicity
Invalid in 2014
PCIe I/O
PCIe I/O
Drwr
Just one drawer Drwr
not supported
© 2014 International Business Machines Corporation IBM & Business Partner Confidential until Announcement 80
SOD More PCIe Adapters for PCIe Gen3 Drawer
© 2014 International Business Machines Corporation IBM & Business Partner Confidential until Announcement 82
Slide added after 18 Sept
However in I/O drawer fan-out module, bandwidth of one system node PCIe slot is
shared by six PCIe slots. It is possible to have multiple high bandwidth adapters in
the fan-out module which together require more bandwidth than available. For
example a 2-port 40Gb Ethernet adapter uses a lot of the available bandwidth
assuming it is really busy. But often clients don’t run their PCIe adapters that
heavily from a bandwidth perspective and it is not a concern
#EMX0 PCIe Gen3 Drawer bandwidth is far, far better than in POWER7+
#5802/5877 12X-attached I/O Drawer. One or two PCIe Gen1 drawers attach to a
GX++ bus and share 10 or 20 PCIe Gen1 x8 slots across that one GX++ bus. A
GX++ bus has a theoretical max of 20GB/s (duplex burst). A fan-out module has a
theoretical max of 32GB/s (duplex burst) to share across just 6 PCIe Gen3 x8 or
x16 slots. One PCIe Gen3 I/O drawer with two fan-out modules has max of
64GB/s (220% larger than a #5802/5877 with a dedicated GX++ slot) .
The PCIe Spreadsheet has sizing factors for individual adapters relating to
bandwidth. Use the same sizing factors for the PCIe adapters for
POWER7/POWER7/POWER8+ servers, but instead of the max GX++ subtotal
value of “60 Gb/s” used for POWER7/POWER7+, use “96 Gb/s” for one EMXF fan-
out module.
IBM’s statements regarding its plans, directions, and intent are subject to change or withdrawal
without notice at IBM’s sole discretion. Information regarding potential future products is intended to
outline our general product direction and it should not be relied on in making a purchasing decision.
The information mentioned regarding potential future products is not a commitment, promise, or legal
obligation to deliver any material, code or functionality. Information about potential future products
may not be incorporated into any contract. The development, release, and timing of any future
features or functionality described for our products remains at our sole discretion.
IBM i
IBM i 7.2 TR1 or later
IBM i 7.1 TR9 or later
Linux
Red Hat Enterprise Linux 6.5 or later
SUSE Linux Enterprise Server 11 SP3 and later SPs
VIOS
VIOS 2.2.3.4 with ifix IV63331 or later
Firmware 8.2.0
E880
795
770
570
750 4-socket
560
740
550
720 6/8 core
Scale-out
520 4-core 1-S / 2-S
720 4 core
520 2-core
730
520 1-core
710
780 780
“B” Mdl “C” Mdl E880
780+
“D” Mdl
570
E870
770 770+
770 “D” Mdl
“C” Mdl
“B” Mdl
570 1 780+
2 E880
770+ E870
Words from IBM Business Practices: “The client must buy the POWER7+ upgrade for purpose of
using the POWER7+ 770/780, not solely as a step to the E870/E880. And, they must get business
value out of the POWER7+ 770+/780 for at least 90 days. Finally, the upgrade to the POWER8 server
needs to be sold separately / independently of the upgrade to POWER7+ 770/780.”
Additional notes:
Assuming historical practices maintained, upgrades from a POWER6 570 to a POWER7+
770/780 will be withdrawn when IBM withdraws the sales of new serial number 770/780.
Conversions from a POWER6 9406-MMA to a POWER6 9117-MMA are NOT considered a
step requiring a significant time pause in the above “no one-step” rule. However
manufacturing systems won’t accept a 9117-MMA770/780 order until 94069117 MES
shows as installed.
770+
2 E870
770
Words from IBM Business Practices: “The client must buy the POWER7+ upgrade
for purpose of using the POWER7+ 770/780, not solely as a step to the E870/E880.
And, they must get business value out of the POWER7+ 770+/780 for at least 90
days. Finally, the upgrade to the POWER8 server needs to be sold separately /
independently of the upgrade to POWER7+ 770/780.”
AIX 7.1
AIX 7.1 AIX 7.1 AIX 7.1 AIX 7.1 AIX 7.1
AIX 6.1
AIX 6.1 AIX 6.1 AIX 6.1 AIX 6.1 AIX 6.1
AIX 5.3
AIX 5.3 AIX 5.3 IBM i 7.1 IBM i 7.1 *
IBM i 7.2 IBM i 7.2 *
Linux Linux Linux Linux Linux
Power
Enterprise
Pools
Pool Totals
Sys A Sys B Sys C Sys D
64-core E880 96-core 795 96-core 780 128-core 795
4.35 GHz 3.7 GHz 3.7 GHz 4.0 GHz Activations:
Activations: Activations: Activations: Activations: 96 static
10 static 30 static 16 static 40 static 160 mobile
40 mobile 40 mobile 20 mobile 60 mobile
14 “dark” 26 “dark” 60 “dark” 28 “dark” 128 “dark”
Example
starting
point
Pool Totals
Sys A Sys B Sys C Sys D
64-core E880 96-core 795 96-core 780 128-core 795
4.35 GHz 3.7 GHz 3.7 GHz 4.0 GHz Activations:
Activations: Activations: Activations: Activations: 96 static
10 static 30 static 16 static 40 static 160 mobile
0 mobile 55 mobile 45 mobile 60 mobile
54 “dark” 11 “dark” 35 “dark” 28 “dark” 128 “dark”
Going back to
initial starting
point and
moving the
activations
differently
Move 15 activations
Not so dark
Memory
Activations:
Mobile memory activations Xx GB static
work the same way as mobile Xxx GB mobile
processor core activations Xxx GB “dark”
Software Tier
AIX = Medium IBM i = P30
System node
Width: 445 mm (17.5 in.)
Depth: 902 mm (35.5 in.)
Height: 219 mm (8.6 in.) 5 EIA units
Weight: 75.7 kg (167 lb)
9117-MMD 9119-MME
Power 770 Power E870
CPU Sockets per Node 4 4
Max processor nodes 4 2
Max number sockets 16 8
Max Cores 64 80
Max Frequency 3.8 GHz 4.19 GHz
Max Memory 1 TB per node 2 TB per node
Memory per core 31.3 GB 50 GB
Memory Bandwidth (peak) 272 GB/s per node 922 GB/s per node
111
© 2014 International Business Machines Corporation 111
POWER 795/780+ 3-Hop 128-way Topology
32-way Drawer
9117-MMD 9119-MME
Power 770 Power E870
CPU Sockets per Node 4 4
Max processor nodes 4 2
Max number sockets 16 8
Max Cores 64 80
Max Frequency 3.8 GHz 4.19 GHz
Max Memory 1 TB per node 4 TB per node SOD
Memory per core 31.3 GB 100 GB
Memory Bandwidth (peak) 272 GB/s per node 922 GB/s per node
Memory Bandwidth (peak) 272 GB/s per node 272 GB/s per node 922 GB/s per node
80 GB/s per node 80 GB/s per node 256 GB/s per node
I/O Bandwidth (peak)
(GX) (GX) (PCIe Gen3)
Max PCIe I/O drws 16 (4 per Node) 16 (4 per Node) 4 (2 per Node)
160 - in IO drws 160 - in IO drws 48 in IO drws
Max PCIe I/O Slots
24 - internal 24 - internal 8 - internal
Memory Bandwidth (peak) 272 GB/s per node 272 GB/s per node 922 GB/s per node
80 GB/s per node 80 GB/s per node 256 GB/s per node
I/O Bandwidth (peak)
(GX) (GX) (PCIe Gen3)
Max PCIe I/O drws 16 (4 per Node) 16 (4 per Node) 15 (4 per Node SOD)
160 - in IO drws 160 - in IO drws 192 in IO drws SOD
Max PCIe I/O Slots
24 - internal 24 - internal 0 - internal
Memory Bandwidth (peak) 272 GB/s per node 272 GB/s per node 922 GB/s per node
80 GB/s per node 80 GB/s per node 256 GB/s per node
I/O Bandwidth (peak)
(GX) (GX) (PCIe Gen3)
Max PCIe I/O drws 16 (4 per Node) 16 (4 per Node) 15 (4 per Node SOD)
160 - in IO drws 160 - in IO drws 192 in IO drws SOD
Max PCIe I/O Slots
24 - internal 24 - internal 0 - internal
IBM Confidential
© 2014 International Business Machines Corporation 120
Power E880 vs Power 795 2014
9119-FHB 9110-MHE
Power 795 Power E880
CPU Sockets per Node 4 4
Max Procesor Nodes 8 2 (4 in 2015)
Max Cores 256 64 (128 or SOD 192 in 2015)
Max Frequency 4.0 GHz 4.35 GHz
Inter CEC Drw SMP Bus (A Bus) 336 GB/s per node 307 GB/s per node
Intra CEC Drw Bus (X bus) 576 GB/s per node 922 GB/s per node
4 TB per node
2 TB per node
Max Memory 8 TB max system 2014
16 TB max system
(16 TB max in 2015)
Memory per core 64 GB 125 GB
Memory Bandwidth (peak) 546 GB/s per node 922 GB/s per node
Memory Bandwidth per core
17 GB/sec 19 GB/sec
(peak)
256 GB/s per node (PCIe
I/O Bandwidth (peak) 80 GB/s per node (GX)
Gen3)
4 (2 per Node)
Max PCIe I/O drws 32 (4 per Node)
(16 – 4 per node 2015)
48 in 2014
Max PCIe I/O Slots 640
192 SOD 2015
© 2014 International Business Machines Corporation 121
Power E880 vs Power 795
2015
9119-FHB 9110-MHE E880
Power 795 48-node SOD
CPU Sockets per Node 4 4
Max Procesor Nodes 8 4
Max Cores 256 192 SOD
Max Frequency 4.0 GHz TBD ~4GHz est
Inter CEC Drw SMP Bus (A Bus) 336 GB/s per node 307 GB/s per node
Intra CEC Drw Bus (X bus) 576 GB/s per node 922 GB/s per node
2 TB per node 4 TB per node
Max Memory
16 TB max system 16 TB max systems
Memory per core 64 GB 85 GB
Memory Bandwidth (peak) 546 GB/s per node 922 GB/s per node
Memory Bandwidth per core
17 GB/sec 19 GB/sec
(peak)
256 GB/s per node (PCIe
I/O Bandwidth (peak) 80 GB/s per node (GX)
G3)
Max PCIe I/O drws 32 (4 per Node) 16 (4 per Node)
Max PCIe I/O Slots 640 192 SOD
The Power Architecture and Power.org wordmarks and the Power and Power.org logos and related marks are trademarks and service marks licensed by Power.org.
UNIX is a registered trademark of The Open Group in the United States, other countries or both.
Linux is a registered trademark of Linus Torvalds in the United States, other countries or both.
Microsoft, Windows and the Windows logo are registered trademarks of Microsoft Corporation in the United States, other countries or both.
Intel, Itanium, Pentium are registered trademarks and Xeon is a trademark of Intel Corporation or its subsidiaries in the United States, other countries or both.
AMD Opteron is a trademark of Advanced Micro Devices, Inc.
Java and all Java-based trademarks and logos are trademarks of Sun Microsystems, Inc. in the United States, other countries or both.
TPC-C and TPC-H are trademarks of the Transaction Performance Processing Council (TPPC).
SPECint, SPECfp, SPECjbb, SPECweb, SPECjAppServer, SPEC OMP, SPECviewperf, SPECapc, SPEChpc, SPECjvm, SPECmail, SPECimap and SPECsfs are
trademarks of the Standard Performance Evaluation Corp (SPEC).
NetBench is a registered trademark of Ziff Davis Media in the United States, other countries or both.
AltiVec is a trademark of Freescale Semiconductor, Inc.
Cell Broadband Engine is a trademark of Sony Computer Entertainment Inc.
InfiniBand, InfiniBand Trade Association and the InfiniBand design marks are trademarks and/or service marks of the InfiniBand Trade Association.
Other company, product and service names may be trademarks or service marks of others.
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, AIX
Version 4.3, AIX 5L or AIX 6 were used. All other systems used previous versions of AIX. The SPEC CPU2006, SPEC2000, 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 Enterprise Edition V7.0 for AIX, XL C/C++ Enterprise Edition V7.0 for AIX, XL FORTRAN Enterprise Edition V9.1 for AIX, XL C/C++
Advanced Edition V7.0 for Linux, and XL FORTRAN Advanced Edition V9.1 for Linux. The SPEC CPU95 (retired in 2000) tests used preprocessors, KAP 3.2 for
FORTRAN and KAP/C 1.4.2 from Kuck & Associates and VAST-2 v4.01X8 from Pacific-Sierra Research. The preprocessors were purchased separately from these
vendors. Other software packages like IBM ESSL for AIX, MASS for AIX and Kazushige Goto’s BLAS Library for Linux were also used in some benchmarks.
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
NotesBench http://www.notesbench.org
VolanoMark http://www.volano.com
STREAM http://www.cs.virginia.edu/stream/
SAP http://www.sap.com/benchmark/
Oracle Applications http://www.oracle.com/apps_benchmark/
PeopleSoft - To get information on PeopleSoft benchmarks, contact PeopleSoft directly
Siebel http://www.siebel.com/crm/performance_benchmark/index.shtm
Baan http://www.ssaglobal.com
Microsoft Exchange http://www.microsoft.com/exchange/evaluation/performance/default.asp
Veritest http://www.veritest.com/clients/reports
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 January 15, 2008
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, AIX
Version 4.3 or AIX 5L were used. All other systems used previous versions of AIX. The SPEC CPU2000, 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 Enterprise Edition V7.0 for AIX, XL C/C++ Enterprise Edition V7.0 for AIX, XL FORTRAN Enterprise Edition V9.1 for AIX, XL C/C++ Advanced Edition V7.0 for Linux,
and XL FORTRAN Advanced Edition V9.1 for Linux. The SPEC CPU95 (retired in 2000) tests used preprocessors, KAP 3.2 for FORTRAN and KAP/C 1.4.2 from Kuck &
Associates and VAST-2 v4.01X8 from Pacific-Sierra Research. The preprocessors were purchased separately from these vendors. Other software packages like IBM
ESSL for AIX, MASS for AIX and Kazushige Goto’s BLAS Library for Linux were also used in some benchmarks.
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/
Veritest http://www.veritest.com/clients/reports
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
ABAQUS http://www.abaqus.com/support/sup_tech_notes64.html
select Abaqus v6.4 Performance Data
ANSYS http://www.ansys.com/services/hardware_support/index.htm
select “Hardware Support Database”, then benchmarks.
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
HMMER http://hmmer.janelia.org/ Revised January 15, 2008
http://powerdev.osuosl.org/project/hmmerAltivecGen2mod
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.
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.
========================================================================
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