Slide 1IBM 4,914
Samsung 3,611
Microsoft 2,906
Canon 2,206
Matsushita 1,829
Toshiba 1,696
Sony 1,680
Intel 1,537
Other companies of note:
SUN …. 562, up from 509….. Moves them up to 32nd from 35th a year ago
GE ….. 979…..
Honeywell …. 655
Watson
IBM Research is building a computer system able to compete with humans at the game of Jeopardy: Human vs. Machine contest.
Code-named “Watson,” Computer is designed to rival the human mind
Answering questions in the Jeopardy! format poses a grand challenge in computer science:
Broad range of topics , such as history, literature, politics, popular culture and science
Nature of the clues, requires analyzing subtle meaning, irony, riddles and other complexities
Based on the science of Question Answering (QA), differs from conventional search
Critical for implementing useful business applications such as:
Customer relationship management
POWER7 System Architecture
Enhances memory implementation
PCIe, SAS / SATA
Built in Virtualization
Availability
Operating Systems
Major Features:
PowerVM Virtualization
Performance
Performance leadership in a variety of workloads
Best Performance per core
Memory and IO bandwidth
Dynamic movement of Partitions and Applications
Reduce infrastructure costs
Hot Add support / Concurrent Maintenance
Alternate Process Recovery
567mm2 Technology:
Transistors: 1.2 B
12 execution units per core
4 Way SMT per core – up to 4 threads per core
32 Threads per chip
L2: 256 KB per core
L3: Shared 32MB on chip eDRAM
Dual DDR3 Memory Controllers
Scalability up to 32 Sockets
360 GB/s SMP bandwidth/chip
Binary Compatibility with POWER6
eDRAM technology
IBM’s eDRAM technology benefits: Greater density, Less power requirements, Fewer soft errors, and Better performance
Enables POWER7 to provide 32MB of internal L3 Cache
L3 Cache critical to balanced design / performance:
6:1 Latency improvement for L3 accesses vs external L3
2X Bandwidth improvement with on chip interconnect. 32B busses to and from each core
No off chip driver or receivers in L3 access path.
eDRAM is nearly as fast as conventional SRAM but requires far less space
1/3 the space of conventional 6T SRAM implementation
1/5 the standby power
1.5 Billion reduction in transistors
IBM is effectively doubling microprocessor performance beyond what classical scaling alone can achieve,” said Dr. Subramanian Iyer, DE (Distinguished Engineer)
DT
1 Branch
6 Wide Dispatch
Out of Order Execution
POWER7 continues to support VMX / Extends SIMD support with VSX
2 VSX units that can each handle 2 Double-Precision FP instructions
8 FLOPS per cycles
VSX units can also handle 4 Single Precision instructions per cycle
VSX instruction set support for vector and scalar instructions
L2 Cache
AIX Release/TL
64 / 128
64 / 128
64 / 128
64 / 128
256 / 1024
64 / 128
256 / 1024
4 / 6 / 8 Core options
45nm technology
3rd Generation Multi-Threading
New Power Bus
Energy Optimized Design
Multiple Memory Controllers
DDR3 memory support
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
Aggregation of L3 Caches of unused cores.
TurboCore chips have a 2X the L3 Cache per Chip available
4 TurboCore Chips L3 = 32 MB
Performance gain over POWER6.
Chips run at higher frequency:
Power reduction of unused cores.
With “Reboot”, System can be reconfigured to 8 core mode.
ASM Menus
Memory Interface
Operating System Support
Linux
Spread work among cores to execute in appropriate threaded mode
Can dynamical shift between modes as required: SMT1 / SMT2 / SMT4
LPAR-wide SMT controls
Requires use of “Resource Groups”
Standard Cache Option
All cores active
© 2010 IBM Corporation
IBM Power Systems
Power your planet
POWER7 Multi-threading Options
MaxCore option
Distributed Enterprise/SMB, Infrastructure Consolidation
Targets infrastructure consolidation, distributed enterprise and SMB
Targets sweet spot of performance, scalability and reliability
High-density, low-power options
Glass ceramic targets leadership performance, scalability and reliability
Designed for enterprise database, ERP, CRM and decision support
Ideal for mission-critical and highly virtualized environments
Supports scalability up to 32 sockets
Massive Scale-Out
Targets unparalleled capacity for modeling complex systems and compute-intensive research
Scales up to 256 w eight-core processors per server and networked clusters of thousands of servers driving >PFLOP capability
2/4s Blades and Racks
3363 Pins
1-4 Socket System
POWER7 Processor Offerings
Cores / Socket
4 TurboCore
EnergyScale
EnergyScale is IBM Trademark. It consists of a built-in Thermal Power Management Device (TPMD) card and Power Executive software.
IBM Systems Director is also required to manage Energy-Scale functions.
EnergyScale is used to dynamically optimizes the processor performance versus processor power and system workload.
IBM Systems Director is also required to manage AEM functions and supports the following functions:
Power Trending
Thermal Reporting
Energy Trending
EnergyScale provides continuous collection of real-time server energy consumption. This energy usage data may be displayed or exported by IBM Systems Director Active Energy Manager.
Administrators may use such information to predict data center energy consumption at various times of the day, week, or month.
Thermal Reporting
*
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Definition of Terms
StaticEnergy Saver Mode
Static Energy Saver lowers the processor frequency and voltage on an Power 750 a fixed amount, reducing the energy consumption of the system while still delivering predictable performance.
This percentage is predetermined to be within a safe operating limit and is not user configurable.
Active Energy Manager is the recommended user interface to enable/disable Energy Saver mode.
Energy Saver could be enabled based on regular variations in workloads, such as predictable dips in utilization over night, or over weekends.
*
Definition of Terms continued
Dynamic Energy Saver Mode
Dynamic Energy Saver varies processor frequency and voltage based on the utilization of the Power 750 POWER7 processors.
The user must configure this setting from Active Energy Manager.
Processor frequency and utilization are inversely proportional for most workloads, implying that as the frequency of a processor increases, its utilization decreases, given a constant workload.
Dynamic Energy Saver takes advantage of this relationship to detect opportunities to save power, based on measured real-time system utilization.
When a system is idle, the system firmware will lower the frequency and voltage to Static Energy Saver values.
When fully utilized, the maximum frequency will vary, depending on whether the user favors power savings or system performance.
If an administrator prefers energy savings and a system is fully-utilized, the system will reduce the maximum frequency to 95% of nominal values.
If performance is favored over energy consumption, the maximum frequency will be at least 100% of nominal.
Dynamic Energy Saver is mutually exclusive with Static Power Saver mode.
*
Power Capping enforces a user specified limit on energy consumption.
The user must set and enable an energy cap from the Active Energy Manager user interface.
In most data centers and other installations, when a machine is installed, a certain amount of energy is allocated to it.
Generally, the amount is what is considered to be a “safe” value, and it typically has a large margin of reserved, extra energy that is never used. This is called the margined power.
*
“Soft” Energy Capping
There are two power ranges into which the power cap may be set .
When a power cap is set in the guaranteed range (described above), the system is guaranteed to use less energy than the cap setting.
Setting a energy cap in this region allows for the recovery of the margined power, but in many cases cannot be used to save energy.
Soft power capping extends the allowed energy capping range further, beyond a region that can be guaranteed in all configurations and conditions.
*
Processor Nap
The IBM POWER7 processor uses a low-power mode called Nap that stops processor execution when there is no work to do on that processor core.
The latency of exiting Nap falls within a partition dispatch (context switch) such that the Hypervisor firmware can use it as a general purpose idle state.
When the Operating System detects that a processor thread is idle, it yields control of a hardware thread to the Hypervisor.
The Hypervisor immediately puts the thread into Nap.
*
Processor Nap
If the processor core is in a shared processor pool (the set of cores being used for micro-partition dispatching) and there is no micro-partition to dispatch, the Hypervisor puts the second thread into Nap mode.
By entering Nap mode, it allows the hardware to clock off most of the circuits inside the processor core.
Reducing active energy consumption by turning off the clocks allows the temperature to fall, which further reduces leakage (static) power of the circuits causing a cumulative effect.
*
Energy-Optimized Fan Control and Altitude Input
On the Power 750, firmware will dynamically adjust fan speed based on energy consumption, altitude, ambient temperature, & energy savings modes.
Systems are designed to operate in worst-case environments, in hot ambient temperatures, at high altitudes, & with high power components.
In a typical case, one or more of these constraints are not valid.
When no power savings setting is enabled, fan speed is based on ambient temperature, and assumes a high-altitude environment.
When a power savings setting is enforced (either Static Energy Saver or Dynamic Energy Saver) fan speed will vary based on power consumption, ambient temperature, & altitude (if available).
*
Processor Folding
Processor Folding is a consolidation technique that dynamically adjusts, over the short-term, the number of processors available for dispatch to match the number of processors demanded by the workload.
As the workload increases, the number of processors made available increases; as the workload decreases, the number of processors made available decreases.
*
TPMD card is part of the base hardware configuration.
Residing on the processor planar
TPMD function is comprised of a risk processor and data acquisition
TPMD monitor power usage and temperatures in real time
Responsible for thermal protection of the processor cards
Can adjust the processor power and performance in real time.
If the temperature exceeds an upper (functional) threshold, TPMD actively reduces power consumption by reducing processor voltage and frequency or throttling memory as needed.
If the temperature is lower than upper (functional) threshold, TPMD will allows POWER7 cores to “Over clock” if workloads demands are present.
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Monitor energy consumption to allow better utilization of available energy resources.
Can trend actual energy consumption and corresponding thermal loading of IBM Systems running in their environment with their applications. :
Allocate less power and cooling infrastructure to IBM servers
Lower power usage on select IBM servers
Plan for the future by viewing trends of power usage over time
Determine power usage for all components of a rack
Retrieve temperature and power information via wireless sensors
Collect alerts and events from facility providers related to power and cooling equipment
Better understand energy usage across your data center.
Identify energy usage
Monitor IT costs across components
Manage by department and/or user
*
32-ways (8 cores/processor card x 4 processor cards)
Industry Standard RDIMM, DDR3 1066 Mbps with enhanced memory RAS features including 64-byte marking ECC code, and ChipKill detection and correction.
512 GB maximum (16GB/DIMM x 8 DIMMs/processor card x 4 processor cards)
8 hot plug and front access SFF SAS DASD.
1 slim media bay for DVD.
1 half high bay for tape drive.
Hot plug 3 PCIe slots and two PCIX slots with Enhanced Error Handling.
One GX+ slot and one GX++ slot (not hot pluggable)
Hot plug and redundant power.
Hot plug and redundant cooling.
Support for Logical Partitioning (LPAR) and Dynamic LPAR (DLPAR).
Embedded SAS and SATA
Embedded four 1 Gigabit Ethernet devices or two 10 Gigabit Ethernet devices
Embedded USB
Service Processor FSP-1 for enhanced reliability and remote system management
Rack mountable drawer
POWER7 Architecture
6 Cores @ 3.3 GHz 8 Cores @ 3. 0, 3.3, 3.55 GHz Max: 4 Sockets
DDR3 Memory
System Unit SAS SFF Bays
Up to 8 Drives (HDD or SSD) 73 / 146 / 300GB @ 15k (2.4 TB) (Opt: cache & RAID-5/6)
System Unit IO Expansion Slots
PCIe x8: 3 Slots (2 shared) PCI-X DDR: 2 Slots 1 GX+ & Opt 1 GX++ 12X cards
Integrated SAS / SATA
Integrated Virtual Ethernet
System Unit Media Bays
IO Drawers w/ PCI slots
PCIe = 4 Max: PCI-X = 8 MAX
Cluster
Redundant Power and Cooling
Yes (AC or DC Power) Single phase 240 VAC or -48 VDC
Certification (SoD)
EnergyScale
© 2010 IBM Corporation
IBM Power Systems
Power your planet
8-core 3.55 GHz #8336 – 4 per server
All processor cards on the same server must be identical feature code
Processor
VRM
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
DIMM
DIMM
SN
RJ45
RJ45
ENET
PHY
RJ45
RJ45
ENET
PHY
POWER7
Chip
POWER7
Chip
POWER7
Chip
POWER7
Chip
Power
Power 755
Power 750
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Power 750 Memory
DIMMS: 4GB, 8GB and 16GB
Plugged in pairs. 1 feature code = 1 pair
Min = 1 feature per SERVER, but min 1 feat per Proc card recommended
Can NOT mix different size DIMMs on same processor card
Can have different size DIMMs on same server.
The following is for ONE processor card in the Power 750
One proc card GB memory capacity with
DIMM size
1 Pair
2 Pair
3 Pair
4 Pair
4 GB
POWER7
Mem
Cntrl
Each Nova Chip (Read/Write Buffer) supports two DIMMS
Chip Bandwidth
** Pass thru bus
Intra-Node Buses
6.4 GB/sec
20 GB/sec 10 GB/sec ( Shared )
GX Bus Slot 1 GX Bus Slot 2 Internal IO Slots Total IO Bandwidth
20 GB/sec 5** GB/sec 5** GB/sec 30 GB/sec
© 2010 IBM Corporation
IBM Power Systems
Power your planet
All PCIe or PCI-X slots are hot-pluggable
All PCIe or PCI-X slots Enhanced Error Handling (EEH)
Slot #
Description
Speed
N/A
Standard PCI Short card
Short card
Standard PCI Short card
Short card
Standard PCI Long card
FC Order #
Available
12X
4
Available
12X
4
Available
12X
48
7314-G30
Supported
SCSI
24
One PCIe IO Drawer
Two PCIe IO Drawers
Three PCIe IO Drawers
Four PCIe IO Drawers
20
GB
20
GB
20
GB
5**
GB
20
GB
5**
GB
PCIe
750
PCIe
PCIe
750
PCIe
750
PCIe
PCIe
750
PCIe
PCIe
PCIe
PCIe
Two IO Drawers
Three IO Drawers
Five IO Drawers
Rules:
Can not mix PCI-X and PCIe drawers on the same 12X loop
GX+ or GX++ can attach either PCIe or PCI-X drawers
750
PCI-X
PCI-X
PCIe
PCIe
Split DASD Backplane Feature
FC 3669 internal SAS cable for Split DASD mode is used to put Lahaina in Split DASD mode.
Where 4 SFF DASD on left (from front view) are assigned to the integrated SAS controller, and 4 SFF DASD on right are assigned to the external rear SAS port.
A PCIe or PCI-X SAS adapter (such as FC 5900 or FC 5901) can access the right 4 SFF DASD via an external SAS cable as shown on a picture below.
*
These four (D3-D6) are assigned to SAS controller
These four (D7-D10) are
assigned to external rear
Alternative to DLT, VXA, DAT72 or 8mm tape drives
Faster than tape / 20MB/s Sustained Transfer Rate
Lower total cost of ownership
USB drives have longer life than tape cartridges
No cleaning cartridges
Inexpensive docking stations
NOT designed to be used as a regular disk drive.
*
Operating voltage:
Power Factor: 0.97
Power-source Loading
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Up to 256 GB Memory 32 DIMM slots
Up to 512 GB Memory 32 DIMM slots
DDR2 DIMMS
DDR3 DIMMs
8 SFF SAS disk/SSD
Commercial focus
IVE: Dual Gb Optional: Quad Gb, or 10 Gb
IVE: Quad Gb Optional: Dual 10 Gb
TPMD
Minimum hardware definition
No free processor activations
1 proc card, SFF backplane, DVD-RAM, 1 memory feature, 2 power supplies, operator panel, IVE daughter card, and 1 or 2 disk/SSD unless using SAN, 2 power cords
750 rule: 100% processor activations, no CUoD/CoD
IBM Edition (generic to AIX, IBM i, Linux)
Minimum edition definition entitles client to ½ the processor cores activated at no charge
Available at initial purchase only (later activations chargeable)
Edition memory minimum = 4 GB per core (or more)
Edition I/O minimum = 2 disk or 2 SSD or 2 FC or 2 FCoE (or more)
Do not need licensing for unused cores
© 2010 IBM Corporation
IBM Power Systems
Power your planet
© 2010 IBM Corporation
IBM Power Systems
Power your planet
POWER7 / POWER6 Comparison
Power 750: 32 Cores Power 550: 8 Cores Power 560: 16 Cores Active
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Performance* / KW
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Only an 8-core 3.3GHz will be offered
Valid configuration is 32-core 3.3GHz (i.e. 4 processor cards).
No 16GB DIMM - Maximum memory is 256GB.
No IBM i O/S support
No PowerVM features (i.e. no LPAR or DLPAR)
No RAID feature (CCIN 57B7 & 57B8)
No Split Disk feature
No IB 12x SDR adapter (CCIN 1817)
*
5.3 / 6.1 RHEL / SLES
( 10 nodes per Rack )
4U x 28.8” depth
DDR3 Memory
System Unit SAS SFF Bays
Up to 8 disk or SSD 73 / 146 / 300GB @ 15K (up to 2.4TB)
System Unit Expansion
PCIe x8: 3 Slots (1 shared) PCI-X DDR: 2 Slots GX++ Bus
Integrated Ports
Integrated Ethernet
System Unit Media Bay
Cluster
Redundant Power
Yes (AC or DC Power) Single phase 240vac or -48 VDC
Certifications (SoD)
EnergyScale
© 2010 IBM Corporation
IBM Power Systems
Power your planet
IB-DDR
Interconnect
Up to 10 Nodes per Rack
Air cooled
1H / 2010
Operating Systems
HPC Stack Levels
xCAT v2.3.x GPFS v3.3.x PESSL v3.3.x LL v4.1.x PE v5.2.x
ESSL
Compilers
© 2010 IBM Corporation
IBM Power Systems
Power your planet
Power 755 Memory
POWER5+ 575 (1.9GHz)
Power 575 (4.7GHz)
Power 755 (3.3GHz)
755 offers the same core count per node
40% better performance per node
1/3 the power per node
37% less floor space for a 64 node configuration.
Power 755
Power 575
Feature
755
750
Processors
32-core @ 3.3 GHz
32-core @ 3.55 GHz 6 / 12 / 18 / 24-core @ 3.3 GHz 8 / 16 / 24 / 32-core @ 3.0 GHz
Memory
GX slot support
Yes – IB clustering
Virtualization
8 SFF SAS HDD / SDD 10k and 15K SFF drives
8 SFF SAS HDD / SDD 10k and 15K SFF drives Optional RAID
Internal Tape
IBM i Support H/W Raid Cards
Gap 3.3GHz up to drive upsell to faster performance and limit buy down from Saturn
Power 550 with 3.5/4.2/5.0GHz has delivered significant incremental GP
using this model.
- ~70% buy faster frequency
- From third to first in $25k to $100k UNIX price band with this pricing
strategy
Power 750 MES Upgrade
Upgrades consist of a box of parts and may be installed by the customer if they so choose.
*
Power 770
Processor Technology
L3 Cache
On Chip
Std: Quad 1Gb Opt: Dual 10Gb + Dual 1 Gb
Std: Four Quad 1Gb Opt: Four x Dual 10Gb + Dual 1 Gb
USB
3
12
Max: 4 PCIe, 8 PCI-X
Max: 16 PCIe, 32 PCI-X
© 2010 IBM Corporation
IBM Power Systems
Power your planet
L3 Cache
On Chip
Std: Quad 1Gb Opt: Dual 10Gb + Dual 1 Gb
Std: Four Quad 1Gb Opt: Four x Dual 10Gb + Dual 1 Gb
USB
3
12
Max: 4 PCIe, 8 PCI-X
Max: 16 PCIe, 32 PCI-X
© 2010 IBM Corporation
IBM Power Systems
Power your planet
POWER7 Modular Layout
Power 780 Processor Options (2 Sockets per enclosure )…