IBM BladeCenter GPU Expansion Blade and GPU Expansion Blade II

IBM Redbooks Product Guide

Abstract

The IBM BladeCenter GPU Expansion Blade and GPU Expansion Blade II provide the capability to attach next-generation graphics processing unit (GPU) technology on supported server blades. This capability is ideal for many applications written to take advantage of acceleration and visualization performance advantages that are offered in general-purpose computing on GPUs. These products offer the choice of NVIDIA Tesla M2070, M2070Q, M2075, and M2090 GPU. In addition, the IBM BladeCenter GPU Expansion Blade is stackable, allowing clients to stack up to four GPU Expansion Blades on a single-base compute blade, thereby offering a unique density advantage versus the competition.

Contents


The IBM® BladeCenter® GPU Expansion Blade and GPU Expansion Blade II provide the capability to attach next-generation graphics processing unit (GPU) technology on supported server blades. This capability is ideal for many applications written to take advantage of acceleration and visualization performance advantages that are offered in general-purpose computing on GPUs. These products offer the choice of NVIDIA Tesla M2070, M2070Q, M2075, and M2090 GPU. In addition, the IBM BladeCenter GPU Expansion Blade is stackable, allowing clients to stack up to four GPU Expansion Blades on a single-base compute blade, thereby offering a unique density advantage versus the competition. Figure 1 shows the IBM BladeCenter GPU Expansion Blade.

IBM BladeCenter GPU Expansion Blade (attached to a HS22 blade server)
Figure 1. IBM BladeCenter GPU Expansion Blade (attached to a HS22 blade server)


Did you know?

General purpose GPUs are more than just high performance graphics engines. In fact, they do not connect to computer displays and they do not have video out capabilities, Instead, they perform general purpose graphical, scientific, and engineering computing and allow the offloading of compute-intensive tasks across all industries, including life sciences, fluid dynamics, finance, data analytics, atmospheric modeling, as well as large scale graphic rendering. These types of applications operate on data that is broken down easily into small chunks that can be operated on in parallel. GPUs are designed for massive parallelism. Whereas a traditional CPU might have four processing cores, the NVIDIA GPUs have more than 100 times that many cores (up to 512 cores), providing a peak double-precision capacity of 665 Gigaflops (billions of floating point operations per second) instead of approximately 50 Gigaflops. The CPU and GPU work together in this computing model. The main sequential part of the application runs on the CPU, and the computationally-intensive part runs on the GPU.


Part number information

Table 1. Ordering part number and feature code
DescriptionPart numberFeature code
IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M207046M6740*5090
IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M2070Q46M6772A10R
IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M207546M6771A25F
IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M2070Q68Y7479A246
IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M207568Y7478A245
IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M209000D6881A2VW
* Special bid or Configure-to-Order (CTO) only

These part numbers include the following items:
  • IBM BladeCenter GPU Expansion Blade or GPU Expansion Blade II with a NVIDIA GPU installed
  • Documentation CD that contains the Installation and User's Guide
  • Warranty information and Safety flyer
  • Important Notices document
  • Technical Update flyer

Figure 2 shows the IBM BladeCenter GPU Expansion Blade attached to an HS22 blade server.


Figure 2. IBM BladeCenter GPU Expansion Blade (attached to a HS22 blade server)


Features

The expansion blades have the following features:
  • Support for a high-performance NVIDIA adapter in each expansion blade

    The expansion blade comes with NVIDIA Tesla M2070, NVIDIA Tesla M2070Q, NVIDIA Tesla M2075, or NVIDIA Tesla M2090 adapter card. The use of these expansion blades enables the use of the specialized adapters that would not otherwise be available to BladeCenter customers.

  • Ability to stack up to four expansion blades on a single base blade

    You can attach up to four expansion blades, thereby maintaining the BladeCenter density advantage while still having the ability to access the power of these high-performance GPUs as needed without the need for additional server blades and the added complexity and cost that brings. The first expansion blade connects to the server blade using the CFFh expansion slot of the server blade. The second expansion blade attaches to the CFFh connector on the first expansion blade, and so on.

    Note: You cannot attach an IBM BladeCenter PCI Express Gen 2 Expansion Blade or PCI Express Gen 2 Expansion Blade II to the same stack as GPU Expansion Blades.

  • CFFh slot still available

    The CFFh expansion connector is accessible on the top-most expansion blade (see Figure 3), even with four expansion blades attached. This lets you maintain the integrated networking capabilities of the blade server when installed in a BladeCenter H or BladeCenter HT chassis.


Figure 3 shows the layout of the expansion blade.


Figure 3. Layout of the expansion blade


Specifications

The expansion blade has the following specifications:
  • 30 mm expansion blade, attaches to a supported server blade through the PCIe x16 CFFh connector
  • Standard NVIDIA GPU computing module
  • Up to four expansion blades can be attached to the base server. With four expansion blades attached, the total width of the server is 150 mm.
  • The CFFh slot remains usable on the top-most expansion blade, for use in BladeCenter H and HT configurations. Supported CFFh cards are listed in Table 4.

The NVIDIA Tesla M2070, M2070Q, and M2075 cards have the following specifications:
  • NVIDIA Fermi GPU engine
  • 448 processor cores operating at 1.15 GHz
  • 6 GB GDDR5 (graphics DDR) memory operating at 1.566 GHz
  • PCIe x16 host interface
  • 225W (TDP) power consumption
  • Double Precision floating point performance (peak): 515 Gflops
  • Single Precision floating point performance (peak): 1.03 Tflops

The NVIDIA Tesla M2070Q combines the Fermi GPU with a NVIDIA Quadro advanced visualization engine in the same GPU. NVIDIA Tesla M2075 supports Dynamic Power Scaling compared to M2070.

The NVIDIA Tesla M2090 card has the following specifications:
  • NVIDIA Fermi GPU engine
  • 512 processor cores operating at 1.3 GHz
  • 6 GB GDDR5 (graphics DDR) memory operating at 1.85 GHz
  • PCIe x16 host interface
  • 225W (TDP) power consumption
  • Double Precision floating point performance (peak): 665 Gflops
  • Single Precision floating point performance (peak): 1.33 Tflops


Operating environment

This card is supported in the following environment:
  • Temperature: 10° C to 35° C (50° F to 95° F)
  • Operating power: 270 W
  • Dimensions: 24.5 cm x 5.9 cm x 45.7 cm (9.7 in. x 2.3 in. x 18.0 in.)


Supported servers

The IBM BladeCenter GPU Expansion Blade and GPU Expansion Blade II are supported attached to the IBM BladeCenter servers listed in Table 2.

TTable 2. Supported servers

Expansion card

Part number

HS12 (8028)

HS21 (8853)

HS21 XM (7995)

HS22 (7870)

HS22V (7871)

HS23 (7875)

HX5 (7872)

HX5 (7873)

JS12 (7998-60X)

JS23/43 7778

LS22 (7901)

LS42 (7902)

PS700/1/2 (8406)

PS703/4 (7891)

GPU Expansion Blade with NVIDIA Tesla M2070

46M6740*

N

N

N

Y

N

N

N

N

N

N

N

N

N

N

GPU Expansion Blade with NVIDIA Tesla M2070Q

46M6772

N

N

N

Y

N

N

N

N

N

N

N

N

N

N

GPU Expansion Blade with NVIDIA Tesla M2075

46M6771

N

N

N

Y

N

N

N

N

N

N

N

N

N

N

GPU Expansion Blade II with NVIDIA Tesla M2070Q

68Y7479

N

N

N

Y

N

Y

N

N

N

N

N

N

N

N

GPU Expansion Blade II with NVIDIA Tesla M2075

68Y7478

N

N

N

Y

N

Y

N

N

N

N

N

N

N

N

GPU Expansion Blade II with NVIDIA Tesla M2090

00D6881

N

N

N

Y

N

Y

N

N

N

N

N

N

N

N
* Special bid or Configure-to-Order (CTO) only

The expansion blade attaches to the blade server unit through the CFFh connector on the system board of the server blade. The top cover of the blade server needs to be removed and the expansion blade attaches to it, forming a double-width blade server unit. Additional expansion blades can be attached, for a total of four. A supported CFFh adapter can be attached to the top-most expansion blade if the server unit is installed in a BladeCenter H or BladeCenter HT chassis.


Supported BladeCenter chassis

Table 3 lists the supported chassis in which the IBM BladeCenter GPU Expansion Blades can be installed.

Table 3. Supported BladeCenter chassis

Description

BladeCenter S

BladeCenter E

BladeCenter H

BladeCenter T

BladeCenter HT

IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M2070

N

Y*

Y

N

Y

IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M2070Q

N

Y*

Y

N

Y

IBM BladeCenter GPU Expansion Blade with NVIDIA Tesla M2075

N

Y*

Y

N

Y

IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M2070Q

N

Y*

Y

N

Y

IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M2075

N

Y*

Y

N

Y

IBM BladeCenter GPU Expansion Blade II with NVIDIA Tesla M2090

N

Y*

Y

N

Y
* Supported in BladeCenter E models 8677-4xx and 3xx only; Chassis requires 2,320 W AC Power Supply options, 46M0508


Supported CFFh expansion cards

Table 4 lists the CFFh expansion cards that are supported or planned to be supported in the IBM BladeCenter GPU Expansion Blades. The use of these CFFh expansion cards requires that the expansion blade be installed in a BladeCenter H or BladeCenter HT chassis.

Table 4. Supported CFFh expansion cards
DescriptionPart numberFeature codeGPU Expansion BladeGPU Expansion Blade II
2/4 Port Ethernet Expansion Card (CFFh) 44W44795476
Yes
Yes
2-Port 40 Gb InfiniBand Expansion Card (CFFh)46M60010056
Yes
Yes
Broadcom 10 Gb Gen 2 2-port Ethernet Expansion Card (CFFh) 46M61680099
Yes
Yes
Broadcom 10 Gb Gen 2 4-port Ethernet Expansion Card (CFFh) 46M61640098
Yes
Yes
Brocade 2-port 10GbE Converged Network Adapter81Y16505437
No
Yes
Emulex Virtual Fabric Adapter (CFFh) 49Y42355755
Yes
Yes
Emulex Virtual Fabric Adapter Advanced (CFFh)49Y42752435
Yes
Yes
Emulex 10GbE Virtual Fabric Adapter II (CFFh)90Y3550A1XG
Yes
Yes
Emulex 10GbE Virtual Fabric Adapter Advanced II (CFFh)90Y3566A1XH
Yes
Yes
InfiniBand 4X DDR Expansion Card (CFFh) 43W44232991
No
No
Mellanox 2-port 10Gb E'net Expansion Card (CFFh)90Y3570A1NW
Yes
Yes
QLogic 2-port 10 Gb Converged Network Adapter (CFFh)42C18303592
Yes
Yes
QLogic Ethernet and 8 Gb Fibre Channel Expansion Card (CFFh)44X19405485
Yes
Yes

Consult IBM ServerProven to see the latest list of CFFh expansion cards that are supported in the Expansion Blades:
http://ibm.com/servers/eserver/serverproven/compat/us/


Operating system support

The following operating systems have been tested for compatibility with the IBM BladeCenter GPU Expansion Blade.
  • Windows Server 2008 HPC Edition (64-bit)
  • Windows HPC Server 2008 (64-bit)
  • Windows 2008 R2 (64-bit)
  • Red Hat Enterprise Linux 5 (64-bit)
  • SUSE Linux Enterprise Server Edition 11 for x86_64 (64-bit)


Related publications
For more information, refer to these documents:

Special Notices

This material has not been submitted to any formal IBM test and is published AS IS. It has not been the subject of rigorous review. IBM assumes no responsibility for its accuracy or completeness. The use of this information or the implementation of any of these techniques is a client responsibility and depends upon the client's ability to evaluate and integrate them into the client's operational environment. publib-b.boulder.ibm.com

Profile

Publish Date
08 November 2010

Last Update
06 March 2012


Rating:
(based on 2 reviews)


Author(s)

IBM Form Number
TIPS0798