In the System p 575 (9125-F2A), IBM plans to deliver
a 32-core, 4.7 GHz POWER6 high-bandwidth supercomputing
node, ideal for many high-performance computing applications.
The p575 will be packaged in a super-dense 2U form factor
with up to 14 nodes installed in a 42U-tall, 24-inch
water-cooled rack. Multiple racks of p575 nodes can be
combined to provide a broad range of powerful cluster
solutions.
Plans are for the symmetric multiprocessor (SMP) node to
use innovative, 64-bit, dual-core POWER6
microprocessors in a 4.7 GHz 32-core configuration. Also
planned is for each microprocessor dual-core chip
to be supported by 32 MB of Level 3 cache, 8 MB of Level 2
cache, and up to 4 DDR2 memory DIMMs. Each 32-core node will
include 64 slots for memory DIMMs. Memory sizes will be
offered from 4 GB up to 256 GB. With the optional I/O
Assembly with PCI and I/O drawer attachment capability,
along with an I/O drawer, up to 24 PCI cards
and up to 18 disk drives per node will be available.
Other planned features include:
Two integrated SAS controllers
Integrated service processor
Four 10/100/1000 Ethernet ports
Optional Dual 10 Gb Optical Ethernet
Optional Advanced POWER Virtualization
Optional dual 2-port 4x Host Channel Adapter
The p575 is planned to support AIX and Linux operating
systems. These operating systems are planned to run
simultaneously in different partitions within the p575 node.
32 MB of Level 3 cache and 8 MB of Level 2 cache per
microprocessor dual-core chip
Up to 256 GB of DDR2 memory per node
Up to 5093 GB of internal disk storage
Up to 18 hot-swap disk bays and 22 hot-swap PCI-X slots
Up to four hot-swap PCI-E slots
Redundant rack power subsystem
Dynamic logical partitioning (DLPAR)
Optional Advanced POWER Virtualization
Previews provide insight into IBM plans and direction.
Availability, prices, ordering information, and terms and
conditions will be provided when the product is announced.
All statements regarding IBM's plans, directions, and intent
are subject to change or withdrawal without notice, and
represent goals and objectives only.
The p575 is planned to be characterized by innovative,
elegant conceptual design and packaging. Mounted in a sleek
2U enclosure, the modular p575 is planned to allow users
to deploy up to 14 water-cooled nodes in a single, 24-inch
system frame.
The heart of the supercomputing node is planned to contain
32 POWER6 processors on 16 dual-core microprocessor
chips. Each microprocessor dual-core chip is planned to be
supported by 32 MB of Level 3 cache, 8 MB of Level 2 cache,
and up to four DDR2 memory DIMMs. Each 32-core node will include
64 slots for memory.
To increase the rack density of the p575 supercomputing node,
the microprocessors are planned to be cooled with an innovative
modular water cooling design and distribution system. Included
in the system side water cooling loop is an integrated rear
door heat exchanger that is planned to significantly reduce the
heat load to air from the system.
At the back of the node enclosure, selectable I/O assemblies
are planned to deliver two small form factor disk bays, four
10/100/1000 integrated Ethernet ports, and optional 10 Gb optical
Ethernet and optional 12X ports for attaching an I/O expansion
drawer. Up to two of the optional dual 2-port 4x Host Channel
Adapters are planned for the high-performance interconnect.
The p575 is planned to be capable of supporting AIX and Linux
operating systems. These operating systems are planned to run
simultaneously in different partitions within the p575 node.
The p575 is planned to be positioned as a substantially
enhanced POWER6 follow-on to the p5-575 POWER5+ cluster node.
Like the p5-575, the p575 is positioned as an extremely
effective solution to the requirements of the most demanding,
memory bandwidth-intensive HPC applications.
This preview provides insight into IBM plans and direction.
Availability, prices, ordering information, and terms and
conditions will be provided when the product is announced.
All statements regarding IBM's plans, directions, and intent
are subject to change or withdrawal without notice, and
represent goals and objectives only.
Minimum system configuration:
The planned minimum configuration for a System p5 575 (9125-F2A)
includes the following items (for the first node in a rack).
Features identified with an "*" are planned to allow alternate
features to be selected for minimum configuration requirements.
1X - System p 575 (9125-F2A)
1X - 0/32 Core 4.7 GHz POWER6 CPU (#7298)
32X - One CPU activation for #7298 (#7299)
1X - I/O Assembly without PCI capability (#6319) *
Some of the following features are planned to be announced with
the initial p575 product announcement or some may be announced at
a later time. They are presented here to give a better
understanding of the Preview product.
A 2U-tall central electronics complex (CEC) housing the system
backplane, power supply, cooling components (blowers and
water-cooled coldplates and manifolds), and system electronic
components. A 0/32 core POWER6 processor with 32 processor
activations and 64 DDR2 memory DIMM slots are planned to be on
the system backplane.
Selectable I/O assemblies: One assembly is planned to have
PCI and I/O
drawer attachment capability (#6399), and one will not have PCI
capability (#6319). Both I/O assemblies are planned to have four
10/100/1000 Ethernet ports per node and two small form factor disk
bays.
The I/O assembly with PCI capability is planned to support up
to two I/O riser features. The I/O riser features have either
PCI-E (8x and 16x) slots, or have one PCI-X slot and one PCI-E-16x
slot.
Up to one 12X attached I/O drawer containing 20 PCI-X slots and
16 hot-swap disk bays, when connected to the I/O assembly (#6399) is
planned. A p575 may be connected to one half of an I/O drawer in
dual-loop mode. The second half of an I/O drawer may be connected
to another p575 node. Attached I/O drawers must be in the same rack
as the node.
A planned 42U-tall, 24-inch System Rack (#5770) ordered on the
first p575 (9125-F2A) node in the system.
A planned redundant power subsystem housed in the top 10U of
the rack.
The p575 CEC is planned to be a 2U-tall, water-cooled, 24-inch
rack-mounted device. It will house the system processors, memory,
system support processor, power supply, blowers, water cooling
hardware, and selectable I/O assemblies with disk bays and (HMC)
connectivity.
The System Rack with base power (#5770) is planned to be a
24-inch rack with an integrated power subsystem to support the p575
nodes. It will provide 42U of rack space and will house the nodes, the
water cooling option, I/O drawers, and the power subsystem.
The water cooling option (#6872) is planned to include the modular
water units, system side supply and return water manifolds, and
associated hoses.
The System Rack with base power (#5770) must have door assemblies.
Doors kits containing front and rear doors will be available in either
slim line or acoustic styles. The door kits include a rear door with
an integrated rear door heat exchanger. The integrated rear door heat
exchanger is on the system side cooling loop.
The Slim Line Door Kit will provide a minimized footprint for use
where conservation of space is desired.
The Acoustic Door Kit will provide additional acoustic dampening
for use where a quieter environment is desired.
The height and weight of the System Rack with base power (#5770)
may require special handling when shipping by air, when moving under
a low doorway, or when using certain elevators. The Compact Handling
Option (#7960), Shipping Depth Reduction Option (#7964), and Shipping
Weight Reduction Option (#6850) are planned to be available if
deemed required during the Systems Assurance Review.
The System Rack with base power (#5770) is planned to utilize
redundant power throughout its design. It will implement redundant
bulk power assemblies, Bulk Power Regulators, power controllers,
Power Distribution Assemblies, and associated cabling. The power
subsystem will provide redundant 350 V dc power to the 575 nodes and
I/O drawers. These bulk power assemblies are mounted in front and rear
positions and occupy the top 10U of the rack. The feature 5770
includes the
system rack, the two bulk power controller assemblies, the first two
bulk power regulators, and the bulk power Ethernet hubs.
Power and Ethernet cable features will be used to connect the dc
power converters to the bulk power assembly and to the Ethernet hubs
in the System Rack. These cable groups contain two UPICs
(universal power input cables) and two Ethernet cables.
The initial order racking sequence is planned to have the water
cooling option (#6872) in the bottom 4U, followed by any I/O drawer
features in the configuration, with the p575 nodes placed above
any I/O drawers.
Bulk Power Regulators (#6333) are planned to interface to the bulk
power assemblies to help ensure proper power is supplied to the
system components. Bulk Power Regulators are always installed in pairs
in the front and rear bulk power assemblies to provide redundancy.
The number of Bulk Power Regulators required is
configuration-dependent based on the number of p575 nodes and I/O
drawers installed.
Two bulk power regulators are included in the System Rack with base
power (#5770). Additional bulk power regulators are ordered
as feature 6333.
Up to four line cords are planned to be required per System Rack
with base power (#5770), and these line cords must be identical
features.
The p575 is planned to be capable of operating voltages
(3-phase V ac at 50/60 Hz) of 200 to 240 V, 380 to 415 V, or 480 V.
Up to four optional Bulk Power Distribution Assemblies (#6328)
are planned to be required for system racks with a maximum configuration.
Logical partitioning (LPAR) is planned to allow the p575 node
resources to be allocated and for multiple instances of the supported
operating systems to be run simultaneously on a single node.
LPAR allocation, monitoring, and control is planned to be provided
by the HMC.
Each LPAR is planned to function under its own instance of the
operating system.
Advanced POWER Virtualization allows partitions to be created that
are in units of less than one CPU (sub-CPU LPARs) and allows the
same system I/O to be virtually allocated to these partitions.
With Advanced POWER Virtualization, the processors on the system can
be partitioned into as many as 10 LPARs per processor.
Optional Advanced POWER Virtualization includes Partition Load Manager,
which
will provide cross-partition workload management across the system LPARs.
An encrypted key is planned to be supplied to the customer and
installed on the system, authorizing the partitioning at the
subprocessor level.
Using Advanced POWER Visualization, the p575 is planned to be
divided into as many as 320 LPARs. System resources can be
dedicated to each LPAR.
Each p575 node must be connected to an HMC for system control, LPAR,
and service functions. The HMC is
planned to be capable of supporting multiple POWER6 nodes.
Each p575 node is planned to be connected to two HMCs for
redundancy, if desired.
Each p575 connects to the integrated Ethernet hubs within
the System Rack (#5770). The hubs are planned to be connected to the
HMC through two Ethernet Cable, HMC Attaches (#7801 or #7802).
The p575 is planned to utilize optional 4U-tall remote I/O
drawers (#5798) for additional directly attached PCI-X adapters
and SCSI disk capabilities.
Each I/O drawer is planned to be divided into halves. Each half
contains 10 blind-swap PCI-X slots and two Ultra3 SCSI 4-pack
backplanes for a total of 20 PCI slots and up to 16 hot-swap disk
bays per drawer.
A maximum of one I/O drawer can be connected to a p575 node.
One single-wide, blind-swap cassette (equivalent to those in #4599)
will be provided in each PCI-X slot of the I/O drawer. Cassettes
not containing an adapter will be shipped with a filler card
installed to ensure proper environmental characteristics for the
drawer. If additional single-wide, blind-swap cassettes are needed,
feature number 4599 should be ordered.
All 10 PCI-X slots on each I/O drawer planar is planned to be
capable of supporting either 64-bit or 32-bit PCI or PCI-X adapters.
Each I/O drawer planar is planned to provide 10 PCI-X slots
capable of supporting 3.3 V signaling PCI or PCI-X adapters operating at
speeds up to 133 MHz.
Each I/O drawer planar incorporates two integrated Ultra3 SCSI
adapters for direct attachment of the two 4-pack hot-swap backplanes
in that half of the drawer. These adapters do not support external
SCSI device attachments.
A system I/O drawer will always be connected to the model 575 CEC
via the 575 I/O assembly (#6399). Drawer connections are always
made in loops to help protect against a single point-of-failure
resulting from an open, missing, or disconnected cable. Systems with
non-looped configurations could experience degraded performance and
serviceability.
I/O drawers are planned to be connected to the CEC in either
single-loop or dual-loop mode. Dual-loop mode is recommended whenever
possible as it will provide the maximum bandwidth between the I/O
drawer and the CEC.
Single-loop mode connects an entire I/O drawer to the CEC via one
12X loop (two ports). The two I/O planars in the I/O drawer are
connected together via a short 12X cable (#1829). Single-loop
connection requires one loop (two ports) per I/O drawer.
Dual-loop mode connects a single I/O planar in the drawer to the
CEC.
A minimum of two identical internal small form factor SAS hard
disk drives are planned to be required per p575 node. It is highly
recommended that these disks be used as mirrored boot devices.
This configuration will provide service personnel the maximum amount
of diagnostic information if the system encounters errors in the boot
sequence.
All European, Middle Eastern and African Countries.
The data in this letter is subject to the disclaimer in Letter ZS90-0112,
which is available from the same IBM announcement letters database.
This announcement is provided for your information only.
For additional information, please contact your IBM Representative or
IBM Business Partner as appropriate.
