ENCLOSURE 2 • l~ Product Guide 5 ......

SGrOrtgtn'" 3000 Sertes Modular, Hlgh-Performance Servers SGIOrigin 3000 Series Product Stanford and SGI .Q.y..~IYi~w. Announce New ~l,;?!.Q[igln.~2..Q.Q Partnership in §.l,;?LQ.r:isln Biomedical ~ Supercomputing SGI Origin 3400 [more] SGI Origin 3800 SARA Supercomputer System Facility Installation [more] Software Modular NUMAflex Supercomputing [view] Bricks ill Partijioning SAIC Upgrades FBI's The SGI Origin 3000 series takes system modularity to Datasheet & National Instant new heights, as NUMAflex™ allows you to scale CPU, Criminal Background White Papers storage, and I/O components independently within each Check System with Server system. Complete multidimensional flexibility allows SGI Origin Technology Solutions organizations to deploy. service, and expand system [more] components in every possible dimension to meet any Related News ShareAPhoto Unveils business demand. The only limitation is your imagination. Revolutionary Online Contact Us Digital Video Solution Using SGI Origin 3000 Configure Series Server [more] This System SGIOrigin Origin 3000:S ~ll.if.S 3200 SGIOrigin dem 3400 p';:2l;? i:t ~{!q(lj(~~ M'l\;iUlilM\;j SGIOrigin f·;ilS~".~ pi<~.hi 3800

Related Sites SGIOnyx 3000 Series Storage Networking Streaming Media

privacy policy IQyestions/CQmments • Copyright © 1993-2001 , Inc. All rights reserved. ITrademark InfonDation °SGIOrigin Product Overview 3000 Series SGITM Origin™ 3000 Series of Servers Product Overview The SGITM Origin™ 3000 Series of servers sets the standard for high • S.G.LQ.r.igin.~2..QQ performance in today's market place by delivering flexibility, resiliency, • S.g.LQrisin investment protection, and performance in a new and innovative 3200c package. SGl's third generation NUMA architecture addresses the • SGI Origin 3400 challenges and cost of true scalability though a new approach in modular • SGI Origin 3800 computing. We call this approach NUMAflex, and it will revolutionize the System high performance server market by making it possible to scale in multiple Software dimensions as your business needs grow and change over time. Acquire only the computation, data processing, or communications power you NUMAflex need, and expand and redeploy systems with ease as new technologies • Bricks emerge. • Partitioning Datasheet & NUMAflex™, Setting the Standard in Modular Computing White Papers SGI realizes that a growing number of competitive business environments have increasingly complex problems that need to be resolved in a short Server period of time. But not every business has the same problems and needs Solutions or the resources to frequently replace computing systems. Providing the Related News ability to build a system that meets the requirements for a specific problem, NUMAflex is an innovative approach to computing, combining Contact Us the third generation of SGI NUMA architecture and a modular hardware design that isolates the various components of today's servers. These Dconfigure modular components, or "bricks", isolate the CPU and memory, 1/0, This System interconnect fabric, and storage into separate bricks. Scalability is no SGIOrigin longer measured by how many systems you can link together, but by the 3200 ability to scale the various components of your system to meet the demands of your applications. SGIOrigin 3400 Flexibility to Design Your Own System SGIOrigin You no longer have to pay for the wasted potential common in yesterday'S 3800 server architectures. The SGI NUMAflex snap-together concept allows you to configure and reconfigure systems brick by brick to meet the exact 8Related Sites demands of your business applications. Complete multidimensional flexibility allows you to deploy, upgrade, service, expand, and redeploy SGIOnyx systems in every possible dimension to meet your current and future 3000 Series business needs. Storage Systems Ready for Tomorrow's Technology Networking NUMAflex offers the most powerful investment protection in the industry. .§tr~~m.i.!'Jg Keeping you on the leading edge, the SGI Origin 3000 series of servers Media can be upgraded with new technologies as they emerge, without having to replace the existing infrastructure. For example, upgrading to new 1/0 technology is as simple as adding a new brick to your system, instead of resorting to a "forklift" upgrade as required by other architectures.

And for those who are not sure about the they want to use in the future, SGI is in development of an Intel® 1A/64 CPU brick capable of running LinuX®. SGI NUMAflex will allow you to upgrade your CPU bricks while maintaining your investment in the infrastructure of your existing system.

Minimize Downtime, Increased Serviceability SGI NUMAflex also provides increased resiliency through its modular design. Many hardware components can be removed and serviced as needed without interrupting the overall system performance. And for maximum fault containment, IRIX® 6.5 on the SGI Origin 3000 series supports partitioning, allowing you the flexibility to divide a single large SSI (single system image) server into smaller SSI systems.

High Performance Scalability without Roadblocks Building on the performance of the award winning SGI Origin 2000 series and its second generation NUMA arChitecture, the SGI Origin 3000 series achieves significant performance gains. Through its modular design and improvements in SGI NUMA 3 architecture, the system structure allows for greater system and memory bandwidth with reduced latencies. And true scalability means that performance will scale proportionately as your system grows.

SGITM Origin™ 3200 This affordable, routerless system scales from two to eight processors in a single shared-memory image and can be clustered (SGIT.M Ori.Q.inT.M 3200C) to leverage the power of hundreds of CPUs using a variety of low-latency, high-bandwidth interconnects. A single point of administration is achieved to manage the cluster through the SGI IRISconsole™ management system.

SGITM Origin™ 3400 This server includes two six-port routers to accommodate scaling from 4 to 32 processors in a single shared-memory image. For maximum flexibility, the processor and I-bricks can be configured within the same rack, with the option of adding a second rack for additional I-bricks in application areas demanding higher levels of 1/0.

SGJTM Origin™ 3800 With the largest single-kernel, shared-memory image available, SGI Origin 3800 easily scales with snap-together modularity to 512 processors and a terabyte of memory. Built-in Metarouters allow clustering to tens of thousands of CPUs. Customers interested in solving very large problems will find SGI Origin 3800 the most powerful and flexible shared-memory system platform on the market. If workloads require a mixture of shared-memory and message-passing development environments, it is flexible enough to be configured with software as a single 512-processor shared-memory server or be divided into several partitions, each running a separate OS without having to recable or reconfigure the hardware. Partitioned system benefits from the low-latency and high-bandwidth NUMAlink™ Interconnect between system partitions while also providing the advantage of higher availability that a cluster delivers.

privacy poli~ Iquestions/comments C.QP''y!.!gbt~.1.~;}':;?9QL§'lllY.9.!l_GmlbjQ§~J!lg, All rights reserved. IIm9.,-mllm.lofQ!.milliQ.o. 03000

D§9.LQ.r.iQ.i.p' -SGI™ Origin™ 3200 3000 Series With the revolutionary SGITM NUMAflex™ Product computing model in the underlying system SGI Origin 3200 .Q.y..~.lYi~w. structure, you decide how much CPU, 1/0, • SGI Origin 3200 memory, and disk infrastructure to add to • ~.~LQ_rjg!n SGI Origin 3200. Every system component 3200C can be upgraded, maintained, or redeployed • SGI Origin 3400 independently, so the 3200 system can • SGI Origin 3800 evolve as quickly as your computing needs. System Software Built on the reliable SGITM NUMA architecture and IRIX® 6.5 operating NUMAflex system, SGI Origin 3200 servers work with your existing application • Bricks software and are fUlly compatible with other IRIX OS-based workstations • Partitioning and servers. The applications you use every day transition effortlessly Datasheet & and perform better than ever. With the same familiar tools and operating White Papers system, you can integrate the series with no retraining. The SGI Origin 3200 server protects your investments thoroughly and ensures the Server availability of a wide range of open systems software into the future. Solutions Related News 2-8 using 2 or Processors: Contact Us 4-processor C-bricks System 11.2GB/sec Dconfigure bandwidth: This System Maximum 16GB SGIOrigin memory: 3200 Router type: none SGIOrigin 3400 Base 110: I-brick SGIOrigin Additional 1/0: I-brick, P-brick, 3800 X-brick, D-brick

8Related Sites SGIOnyx 3000 Series Storage Networking Streaming Media

privacy policy Iquestions/comments Copyright © 1993-2001 Silicon GraPhics, Inc. All rights reserved. ITrademark Information M §9.LQ-r.!Q.i.r.l SGF Origin™ 3400 3000 Series With the revolutionary SGITM NUMAflexIM Product computing model in the underlying system .Q.Y..~fY.!~l!Y. structure, you decide how much CPU, I/O, S_G.!.Qr.igi'{1.~:?QQ memory, and disk infrastructure to add to the S.~.!.Qr.igin SGI Origin 3400 system. Every system ~ component can be upgraded, maintained, or SGI Origin 3400 redeployed independently, so the SGI Origin ~9.J..Qrjgj!J.~ 3400 system can evolve as quickly as your System computing needs. Software With the industry's most advanced NUMA NUMAflex architecture from SGI, you can configure your Bricks SGI Origin 3400 system up to a single Partitioning 32-processor shared-memory system, or use Datasheet & partitioning to divide it into 2 partitions and run Wl:!i1~.. P~.p..~.~ them as a tightly coupled cluster. Many application environments can improve availability by implementing a Server cluster of smaller partitions that can contain failures and leave other Solutions partitions unaffected. Utilizing the ultra low latency and high-bandwidth Related News NUMAlinkTM interconnect fabric as a communication vehicle, partitioning is an option that can deliver both high availability and high performance. Contact Us Built on the reliable SGITM NUMA architecture and IRIX® 6.5 operating Configure system, SGI Origin 3400 servers work with your existing application This System software and are fully compatible with other IRIX OS-based workstations SGIOrigin and servers. The applications you use every day transition effortlessly 3200 and perform better than ever. With the same familiar tools and operating system, you can integrate the series with no retraining. The SGI Origin SGIOrigin 3400 server protects your investments thoroughly and ensures the 3400 availability of a wide range of open systems software into the future. SGIOrigin 3800 4-32 using Processors: 4-processor C-bricks Related Sites System 44.8 GB/sec SGIOnyx bandwidth: 3000 Series Maximum 64GB Storage memory: Networking Router type: 6-port Streaming Base I/O: I-brick Media I-brick, P-brick, Additional 1/0: X-brick, O-brick

privacy policy Iquestions/comments GQP-'y'(i.9ht~J~~:~L~ll.lc.Q.n_9..OOlhj~ ...... lO!<, All rights reserved. 1I@l;te.mar.ls..lnf9f.mit~jQ.a SGI Orillin 3000 Series Technical Specifications

SGI Ort~n 3200 SCI Ortetn 3400 SCI Ortetn 3800 ProCesSllr> HI 4-32 16-512 ~Iem bandwidlh 11.2GB/sec 44.8GB/sec max. 716GBlsec max. Maldmwn memory 16GB 64GB ITB Router Iype None 6-port 8-port, Melarouter Base I/O One I-brick One I-brick One I-brtck Add/l1onal system 1/0 One P-brick ~temdisk 18GB 18GB 18GB Operal1"ll syslem IRIX 6.5 IRiX 6.5 IRiX 6.5

R-brtck ElII\ronmentallOperattlllll '6-port supports shared memory system •Temperalure +5to +35"C, altilude 5,000 MSl • +510 +30"C, allllude 10,000 MSl configurations up 10 32 CPUs Power Bay '8-port Supports shared memory syslem '!'oNer requtl1!lllel1s 220-240 VAC external Sllurce "Hwnldltv 10% to 90% noncondensing configurations up 10 128 CPUs •Power d1strtbul1on 48 VDC internally dlstrtbuted to all brtcks •Metarouter Supports shared memory syslem ElWlronmentallNonoperattnal configurations up 10 512 CPUs PCI Adapters •Temperalure -2010 +6O"C •I-port Fibre Channel optical •Humidity 10% 1095% noncondenslng C-brtck •I-port Fibre Channel copper ·Altilude 40,000 MSl · Processors 4 RI2000" or RI40ll0"~ CPUs •I-port ATMOC3 •Memory Up 108GB ECC SDRAM In 4 banks ·I·port ATMOCI2 Electrtcal and Power •Memory kils 512MB, 1GB, 2GB~ •I-port Gill"btt Ethernet optical · Vollage 200-230 VAC, stngle-phase and 3-phase, •Menwrv controller 5-port crossbar •I-port Gigabit E1hemet copper 50160 Hz •Memory bandwidth 3.2GB/sec total memory bandwidth '2-port Ultra SCSI diffurenl1al •Heal/power 4,500 Wmaximum per power bay, N+I Interleaving 4-way per C-brick '2-port Ultra2 SCSlllVDI (6 x 7SO Wsupplies), 15,100 BTU/hr '8-port d1gilal aud10 •Electrtcal servtl!ItVpe NEMA 6-30, 208 VAC (ijJ 30 amp I-brtck •Ports 2-ports USB, looBase-T, I-port IEEE 1394, XIO Adapters Software I·port sertal, I-port FIbre Channel •I-port FOOl dual attach 'System software IRIX 6.5 Advanced Server Environment, •Internal devices I system disk standard, CD-ROM drtve 'I-port HIPPl BOO sertal X/OPEN XPG4 BASE 95, IEEE POSIX 1003.2, •DIsk Inlerface Fibre Channel •Digital video and 1OO3.1b. IOO3.1c FIPS 151·2, UNIX ~tem '1/0 Interlace One 64-bil/66 MHz PCI bus, 2 skits; •Dtgital video with DVCPro 4.4,4.3 BSD extenSions, MIPS ASI, SVID one 64-biI133 MHz PCI bus, 3 slols •High-definition video Issue 3, XII R6, MoI11 Window Manager 1.2, •I-port GSN lhall bandwidth) IRIS Gl", OpenG/.' P-brtck •I-port GSN Ifull bandwidthI •Networldng TCP/IP, NFS V2N3, RSVp, DHcp, Bulk Data 'Inlerlace 64-biI/66 MHz PCI, 3.3 V, and Unlver>al 'VME 6U Service IBDSpro). NetVtsualvzer", SNMP •Number of buses 6 'VME 9U managemenl, SNMP MIB, NISIONC +, •Number 01 slots 12 12lbusl lull-Ieneth '4-port ATMOC3~ OS bypass with Schedule Transler ISTI •Total 1/0 bandwidlh 3.IGB/sec peak tolal protocol Mass Sto... •Server software XFS" 64-b1t journaied IUes'/Stem with X-brtck •HBA Interfaces Fibre Channel, Ultra SCSI, Ultra2 SCSI guaranteed rale lID, Clustered XFS ICXFSI, 'Inlerface XIO •RAID controller Fibre Channel, 128MB ; Networker, HPC Performance Co-Pilot" •Number of slols 4 2 controllers perSGl" TP9100 module svstem and network monltortng, S'/Stem ·Tolail/O bandwidth 1.6GB/sec peak 'Intemal loops Two standard per TP9100 MIB IProvislon). Sllftware dtstrtbul10n •Maldmum capacity I66TB )800, 656T8 RAID Wropell, Enlighten llSM TP9100 RAID rack; maximum of 9 D-brtck •RAID storage ·Clusler software MPi Toolkil, LSF, and IRIX Advanced •Interface 66 MHz/1GB Fibre Channel, SAN aware TP9100 modules Cluster EnvlronmenllACEI provide . Drive bays 12 hoI-plug, 3.5" power 1101220 V, centralized administration 10 support redundanl power supplies standard Devtee capactty clustered or partitioned server>, job •18GB, 36GB. 73GB RAID •Maldmum bandwidth 200MB/sec scheduling, accounl1ng, load balancing •Device capacity 18GB, 36GB, 73GB JBOD of bateh/lnteracllve Jobs, SIW dlstrtbul1on, Dtmenslons and Weteht5 and user, s'/Stem, and network management 'SGIOrigin3200 34" Hx 40" 0 x 24' W; 17U Internal •Compiler> ANSI C, C++, Fortran 77 and 90, AllA, G-br1ck usable space; 250 Ib max. Pascal, Power CAcceleralor WCAI, 'SGIOrtgin3400l3800 74' Hx SO' 0 x 30" W; 39U Inlernal 'InflnlteReality3 graphiCS Power Fortran 77 and 90 usable space; 970 Ib max. '1-2 graphics pipelines per G-brtck Samba environments lor PC '1/0 rack 74" Hx SO" 0 x 3D" W; 39U Internal 'Interoperabtlitv "Ftr>t pipe: Ior 2 Rasler Manager> usable space; I,OSO Ib max. 'Securttv Trusted IRIX" BI securtty, Commercial •Second pipe: 1,2, or 4 Rasler Manager> •RAlD/jBOO rack 71- Hx 32' 0 x 24' W; 38U Internal Securtty Pack ICSP) · 2~ display channels per usable space; 1,265 Ib max. •PartitionIng CU Interlace mkpart In IRIX~ Processor Data • MIPS RISC RI2000 al400 MHz, RI4000 al 500 MHz~ •Prtmary cache R120oo: 2-way set-associative 32KB instrucl1on/32KB dala cache RI4000: 2-way set-assoclallve 32KB Instrucl1on/32KB dala cache "Secondary cache R12000: 8MB, SDRAM RI4000: 8MB. DDR lull-speed SDRAM~ • Available Q2CYOI

Corporate Offk:e Nortfl America 11800) 800·7441 1600 AmphItheatre Pkwy. lattn America 116S01 933-4637 sgt Mountain View, CA 94043 Europe [441118.925.75.00 1650/960·1980 Japan 18113.5488./811 www.set.com Asia Pactfic 165) 771.0290

102000 SIlicon Graphlcs,lnc. All r~ts reservel1 SpeclfbUons subJect to C:han~ wtl~l nolkl!. Silicon Graphtcs, iRIX,. InflnlteReiIllV. OpenGL. and IRIS are teQlsterm lr1demarlcs. and SCI, lRiS FailSafe, Trusred lADe, Or1Qln, Performara Co-Ptlt, XFS, CXr:s. lnflnltl!RuUt,3, NUMAfb, NUMAlln!l. IRtSconsole.IR1S GL. Nel:Vlsualyzer. and Ifle SCi bQO are tr3demarks, d SllleDn Graph~ Inc. MIPS Is a Itilstered trademark, and RI2000 an:! RI4000 if! tradernilr1ts, 01 MIPS Technoloites. Inc.. used UIlder lkense bySlIlcon Inphlcs, Inc. UNIX Is i reilstered tra.dM1ari In the u.s. and other COllnlrles, Ilcensed excltSl~ly IhftllJliVt XlOpen Company Umlled. All oth!f Irademaris IT'entloned herein are the propert'f d their respecll~ OWI'I!I'1 2775 r'IOOJ j12056 • Oatasheet 5 (

SG rM Origin™ 3000 Series SGIIM Origin'" 3200. SGI'" Origin'" 3400. and SGI'" Origin'" 3800 Servers

F~atur"5 As Fl",xtllie as Yu!.lr lma~t!latiun ·True multidimensional scalability Building on the robust NUMA architecture that made award-winning SGI •Snap-together flexibility, serviceability. and resiliency Origin family servers the most modular and scalable in the industry. the ·Clustering to tens of thousands of processors SGI Origin 3000 series delivers flexibility, resiliency, and performance at •SGI Origin 3200-scales from two to eight breakthrough levels. Now taking modularity a step further, you can scale MIPS processors CPU. storage, and I/O components independently within each system. •SGI Origin 3400-scales from 4 to 32 MIPS processors Complete multidimensional fleXibility allows organizations to deploy, •SGI Origin 3800-scales from 16 to 512 MIPS processors upgrade, service, expand, and redeploy system components in every possible dimension to meet any business demand. The only limitation is your imagination.

Bl;t!J II '{(tur Wa'l SGI" NUMAflex" is a revolutionary snap-together server system concept that allows you to configure-and reconfigure-systems brick by brick to meet the exact demands of your business applications. Upgrade CPUs to keep apace of innovation. Isolate and service I/O interfaces on the fly. Pay only for the computation, data processing, or communications power you need. and expand and redeploy systems with ease as new technologies emerge.

P;!rj'.}O'l)

100i:!:;jrmmt Pr....tm:Ho!l Built on the reliable SGI NUMA architecture and IRIX<» 6.5 operating system. SGI Origin 3000 series servers work with your existing applica­ tion software and are fully compatible with other SGI IRIX OS-based workstations and servers. The applications you use every day transition effortlessly and perform better than ever. With the same familiar tools and operating system, you can integrate the series with no retraining. SGI Origin 3000 series servers protect your investments thoroughly and ensure the availability of a wide range of open systems software into the future...... /" . :· ..... f./ :

>~""":' .., . -'-...... ;'.-..'{,:" ~ -=, ....

The SGI Origin 3000 series gives you the flexibility to build ~.. a system that truly matches tomorrow's growth plans and today's business requirements-whether you need advanced

computation for weather simulation, massive storage for ,".0"'-': archiving bioinformatic data, or high-performance I/O for <)··t media streaming. To scale system performance to meet your needs. choose the bricks your applications require. .:.:.:...•. . .

..;- :: ~.~•.....:•..~: .. .,;. N· ... Build and aintain Your Ideal System­ One Brick at a Time

SCI Ort~i.n 3200 SCI On~i.n 3!~(jO SGI Or~tn 3800 This affordable. routerless This server includes two With the largest single-kernel, shared-memory image available, system scales from two to six-port routers to accom­ SGI Origin 3800 easily scales with snap-together modularity to eight processors ina si ngle modate scaling from 4 to 32 512 processors and a terabyte of memory. Built-in Metarouters shared-memory image and processors in a single allow clustering to tens of thousands of CPUs. making super­ can be clustered to leverage shared-memory image. For computing more accessible than ever. Customers interested in the power of hundreds of maximum flexibility. the very large memories to solve the largest problems will find SGI CPUs using a variety of processor and I-bricks can Origin 3800 the most powerful and flexible shared-memory low-latency, high-bandwidth be configured within the system platform. If workloads require a mixture of shared­ interconnects. Asingle same rack, with the option memory and message-passing development environments, point of administration is of adding a second rack SGI Origin 3800 is flexible enough to be configured with achieved to manage the for additional I-bricks in software as a single 512-processor shared-memory server cluster through the SGI application areas demand­ or be divided into several partitions, each running a separate IRISconsole" management ing higher levels of I/O. OS without having to recable or reconfigure the hardware. system. Apartitioned system benefits from the low-latency and high­ bandwidth NUMAlink- Interconnect between system partitions while also providing the advantages of higher availabili ty that a cluster delivers. \, Router Interconnect As the structural building block of the system, the R-brick replaces the system bus; it's a high-speed crossbar connectinil processors and memory and enabling each system component to be serviced or upgraded individually. Add infrastructure as you need it-from routerless deskside systems to an eight-port router multi rack configuration that delivers 512 processors in a single shared-memory environment. CPU Module The basic C-brick module contains four MIPS· CPUs and local memory. Asingle crossbar delivers 200% greater CPU-to-memory bandwidth than previous generations. Now with four CPUs in a C-brick, the system offers a two-fold increase in CPU density, improving memory latency by up to 50% and minimizing the use of valuable floor space. Baoe 1./0 Module The I-brick, standard in all systems, provides base I/O in a module and includes the system disk, CD-ROM, Ethernet, and four available PCI slots. As the system grows, customers have the option to partition the system for ilreater availability, using additional I-bricks as base I/O for each partition. pC! EXpaiMion For PCI expansion, a P-brick provides 12 hot-swappable PCI slots distributed over six 64-bit/66 MHz PCI buses. Total peak I/O bandwidth 1 exceeds 3GB per second. f / XiO Exrsan(;l!'on For high:performance I/O expansion, X-bricks deliver four XIO slots that support HIPPI, GSN, VME, and digital video. Graphic6 ExpanMon Tightly integrated InfiniteReality3 ilraphics add Iarlle-scale / capabilities for accelerated insight into complex data sets. ~ Di6k Storagr. D-bricks provide modular JBOD mass storage for data-intensive applications. D-bricks support up to 12 drives, have dual-power supplies standard, and support drive capacities of 18GB, 36GB, and 73GB. !t : SGI Origin 3000 Series Technical Specifications

SCI Ortgln 3200 SGI Ortgtn 3400 SGl Ortgtn 3800 Processors 2-8 4-32 16-512 System bandwidth 11.2GB/sec 44.8GB/sec malt 716GB/sec max. Maximum mem0"l 16GB 6/tGB ITB Router type None 6-port 8-port, Metarouter Base I/O One I-brick One I-brick One I-brick Addtl10nal system VO One P-brlck System disk 18GB 18GB 18GB Operati~ system IRIX 6.5 IRIX 6.5 IRIX6.5

R-brtck Power Bay EnvlronmentallOperatill\l1 06-port Supports shared memo"l svstem -PoNer requtl1!lTl!l1s 220-240 VAC external source 'Temperature +5 to +35'C, altttude 5,000 MSl • +5 to +30"C, altttude 10,000 MSl configurattons up to 32 CPUs 'Power dtstr1bUl1on 48 VDC internallv dlstr1buted to all br1cks ·8-port Supports shared memo"l system 'Humidity 10% to 90% noncondenslng configurations up to 128 CPUs PCI Adapters •Metarouter Supports shared memo "I svstem 'I-port Fibre Channel opttcal EnvlronmentallNonoperattll\l1 configurations up to 512 CPUs •I-port Fibre Channel copper ,Temperature -20 to +60"C •I-port ATMOO •Hum1rlity 10% to 95% noncondenslng C-br1ck 'I-port ATMOCI2 •Altttude 40,000 MSl -Processors 4 R12ooo" or RI4ooo"' CPUs 'I-port Gtgab1t Ethernet opttcal -Memo"l Up to 8GB ECC SDRAM In 4 banks 'I-port G~blt Ethernet copper Electr1ca1 and Power •Memo"l kits 512MB, 1GB, 2GB" -2-port Ultra SCSI differential , Voltage 200-230 VAC, single-phase and 3-phase, -Memo"l controller 5-port crossbar '2-port Ultra2 SCSlllVDI SO/60 Hz -Mem0"l bandwidth 3.2GB/sec total memo"l bandwidth .8-port digital audio 'Heat/power 4,500 W maximum per power bay, N+I Interleaving 4-waV per C-brick 16 x 7SO Wsupplies!. 15,100 BTU/hr XIO Adapters 'Electrical servtceItvJle NEMA 6-30, 208 VAC (ii130 amp l-br1ck 'I-port FOOl dual attach -Ports 2-ports USB, 100Base-T, I-port IEEE 1394, 'I-port HIPPI 800 serial Software I-port serial, I-port fibre Channel , Digital video •Sysll!m software IRIX 6.5 Advanced Serlll!r Environment, ·Internal devices Isvstem disk standard, CD-ROM dr111l! 'Oigital video with DVCPro X/OPEN XPGll BASE 95, IEEf POSIX 1003.2, · DIsk Interface Fibre Channel 'High-definitton video and 1003.lb, 1OO3.1c FIPS 151-2, UNIX Svstem -I/O interface One 64-bit/66 101Hz PCI bus, 2 slots; 'I-port GSN lhalf bandwidthI 4.4,4.3 BSD extensions, MIPS ABI, SVID one 6/t-bitfl3 101Hz PCI bus, 3 slots 'I-port GSN (full bandwidthl issue 3, XII R6, Mottf Window Manager 1.2, 'VME 6U IRIS Gl", OpenGlo 'VME 9U P-br1ck , Networking TCP/lP, NFS V2N3, RSVp, DHCp, Bulk Data +port ATMOC3' 'Interface 6/t-blt/66 MHz PCI, 3.3 V, and Universal Service IBOSprol, NelVisualvzer", SNMP , Number of buses 6 management, SNMP MIB, NISIONC +, -Number of slots 12 12/busl full-length Mass Storage OS bypass with Schedule Transfer ISTI -Total VO bandwidth 3. 1GB/sec peak total 'HBA interfaces Fibre Channel, Ultra SCSI, Ultra2 SCSI protocol ,RAID controller Fibre Channel, 128MB cache; 'Server software XFS" 6/t-lit journaled ftlesystem with X-br1ck 2 controllers per SGI" TP9loo module guaranteed rate I/O, Clustered XFS ICXFSI, -Interlace XIO 'Internal loops Two standan! per TP9I00 Networker, HPC Performance Co-Pilot" ·Number of slots 4 -Maximum capacity I66TB JBOD, 656TB RAID system and network montloring, System •Total I/O bandwidth 1.6GB/sec peak , RAIO storage TP9100 RAID rack; maximum of 9 MIB IProvislon), software dlstrlbutton TP9100 modules (Propell, Enlighten OSM D-brtck Devtce Capactty 'Cluster software MPI Toolkit, LSF, and IRIX Advanced -Interface 66 101Hz/1GB Fibre Channel, SAN aware '18GB, 36GB, 73GB RAID Cluster Environment IACEI provide -Drive baVS 12 hot-plug, 3.5" power II0n20 V, centralized admlnlstratton to support redundant power supplies standard clustered or parttttoned serlll!rs, job -Maximum bandw1rlth 200MB/sec DImensions and weIQhts scheduling, accounttng, load balancing -Device capacity 18GB, 36GB, 73GB IBOD 'SGIOrlgin3200 34' Hx 40" 0 x 24' W; 17U 1nternal of batchlinteracttve jobs, SIW distribution, usable space; 250 Ib max. and user, system. and network management . SGI Orlgtn3400fl800 74' Hx SO' 0 x 30' W; 39U Internal C-brtck usable space; 970 Ib max. 'Compilers ANSI C. C+ +, Fortran 77 and 90, ADA, Pasca~ 'Infin1teReallty3 graphics '1/0 rack 74- Hx SO' 0 x 30' W; 39U Internal Power CAccelerator IPCAI, -1-2 graphics pipelines per G-brick usable space; I,OSO Ib max. Power Fortran 77 and 90 •First pipe: lor 2 Raster Managers -RAID/IBOD rack 71- Hx 32' 0 x24' W; 38U Internal 'Interoperability Samba environments tor PC 'Second pipe: 1.2, or 4 Raster Managers usable space; 1,265 Ib malt •Security Trusted IRIX" BI security, Commercial -2-8 display channels per graphics pipeline Secur1tv Pack ICSP) CU Interface mkpart In IRIX' Processor Data •Parttttoning -Microprocessor MIPS RISC R12000 at 400 101Hz, R14000 at 500 101Hz' -Prima"l cache R12000: 2-way set-associative 32KB instndionfl2KB data cache R14000: 2-way set-associative 32KB instruclionfl2KB data cache -Secondarv cache RI2000: 8MB, SDRAM R14000: 8MB, DDR full-speed SDRAM" • Availabl, Q2CYOI

Corporate Otftce North America 118001 800-7441 1600 AmphItheatre Pkwy. lattn America 116501 933-4637 Mountain View, CA 94043 Europe 1441118.925.75.00 [6501 960·1980 lapan 1811 3.5488.1811 www.sgt.com Asia pacinc 1651 771.0290

02000 SlUcon GraphIcs, Inc. AU rlQhls l"e5l!'rvtd. Spectflcollons SlJb,leCI to ch.ln~ \fIlIlthout notte. Silicon GraphiCs, IRIX. InfintteRuUty. OpenGl, and IRIS are rtilstered trademar1ls, and SCI. IRIS FailSafe, Trusted IRrx. Ortall1, Performance ~-Plb~ XFS. CXFS. InflnlteReallfyJ, NUMAflex, HUMAllnk, IRISconsoIe, IRIS Gl, NetVlsuil./yler, and the SGIIoQo are lrademal1:s, rJ Silicon GraphiC$, Inc. MIPS \5 a re~lsrerec! rridemar1l,.oo RI2000 ana Rl4000 m' lridemalts. of MIPS TechnoloQt~ Inc.• used under lanR by SIlicon Graphics. Inc. UNO!. IS a ~stered Imlemarllin the u.s. and other countries, hCfn2d exc:lu$l~I¥ th~ XlOpen Company Umlted. All ether traderrwks menlloned herein are the property d their rnpectlve OWrJefl, m, rl/OO] 112048 SGITM 3000 Family Reference Guide SGITM 3000 Family Reference Guide

Introduced in July 2000. the SCI 3000 family of high-performance servers and visualization systems is the latest addition to the line ofdistributed shared-memory systems from SCI. Known as the SCI™ Origin™ 3000 line ofservers and the SCJTM Onyx(R) 3000 line ofvisualization systems. all members of the SCI 3000 family are designed with the SCITM NUMA 3 architecture, thethird generation in a line ofNon-Uniform Memory Access (NUMA) architectures from SCI.

Starting from a minimum configuration oftwo 64-bit MIPS(R) RISe . the NUMA 3 architecture of the SCI 3000 family scales to support up to 512 processors in a single memory image, single operating system kernel, cache-coherent multiprocessor system. In addition to being a highly scalable server platform, the SCI InfiniteReality3™ graphics subsystem is tightly integrated into the architecture to provide customers with a highly scalable visualization system.

There are a total ofsix models in the SCI 3000 family--three configurations are specified as servers and another three as visualizationsystems. Each model is defined by the maximum number of processors supported in the system. SCI 3200 series--supports a maximum of8 processors SCI™ Origin™ 3200 SCJTM Onyx(R) 3200 SCI 3400 series--supports a maximum of32 processors SCI™ Origin™ 3400 • SCI Onyx 3400 SCI 3800 series--supports a maximum of 512 processors • SCI™ Origin™ 3800 scrM Onyx(R) 3800

1 SGI 3000 Family Reference Guide

SGI 3000 Family Product Overview

With the announcement of the SCI 3000 family, SCI introduces NUMAflex™ as the desired methodology for creating high-performance systems to meet the vast information demands of the future. NUMAflex offers the ability to configure a modular and scalable system to meet the demands ofyour application environment. The concept of NUMAflex delivers a proven SCI NUMA architecture and a bevy ofbenefits for the customer. namely maximum flexibility. resilience. and investment protection: • Delivers precise mix of capabilities. saving money and space • Virtually unlimited expansion takes the guesswork out ofplanning • Reconfigures to meet changing requirements without penalty Breakthrough in availability. service. administration. and logistics Incorporate new technologies as they're introduced (e.g., IA-64. I/O) Independent component upgrades for maximum investment protection Isolate components as necessary for easy serviceability • Run your machine in single system image (shared-memory) or duster mode Manage your large SSI system. partitioned system. or duster from a single point of administration using a management platform from SCI • Use NUMAIinkTM as a high-speed interconnect to run shared memory. message passing. or OpenMP™ jobs. and as a capability or throughput machine simply through software changes--no cabling changes are required Customers can buy. deploy. and redeploy building blocks to build the correct system and exact configuration for the job they wish to do Modular and reconfigurable system allows easy upgrades as needed when new technology is introduced • NUMA 3 modularity allows customers to remove and service failed components while the rest of the system continues to produce results Software enhancements offer increased control ofpartitioning and automatic restart ofsystem

2 NUMA 3 Architecture

The system architecture for the SGI 3000 family is a thirdgeneration NUMA architecture from SGI known as NUMA 3. In the SGI NUMA 3 architecture. all processors and memory are tied together into a single logical system through the use ofspecial crossbar switches developed by SGI. This combination ofprocessors. memory. and crossbar switches constitute the interconnect fabric caUed NUMAlink.

One component of the SGI NUMA 3 architecture is the Bedrock chip. This ASIC is an 8-input by 6-output crossbar that acts as the memory controUer between processors and memory in the system. both local and remote. The Bedrock chip has a total aggregate peak bandwidth of3.2GB/second. The chip also has a channel thatconnects processors to system I/O. which allows every processor in a system direct access to every I/O slot in the system.

Another component ofthe SGI NUMA 3 architecture is the router chip. This 8-port crossbar ASIC is found in a router node and. through the use of highly specialized cables, provides the high-bandwidth. extremely low latency interconnectknown as NUMAlink. The router node channels information between all the compute nodes in the system. connecting all Bedrock switches together to create a single contiguous memory in the system of up to 1 terabyte.

These crossbar switches. together with the NUMAlink interconnect fabric. deliver an extremely low-latency and highbandwidtharchitecture. notjust anothercluster solution tightly integrated in a single chassis and passed off as an SMP system. From the smallest 4-processorsystem to the largest 512-processorsupercomputer. thememory latency ratio between remote and local memory in the SGI NUMA 3 architecture is only 2:1 and sti)) less than 600 nanoseconds round-trip in the largest configuration.Not all NUMA implementations are considered equal. Other NUMA architectures available in the industry exhibit remote to local memory latencies of 3:1 or worse. proving how difficult it is to design an effective NUMA implementation. While other system vendors introduce NUMA architectures for the first time. SGI is delivering its third generation and has nearly 4 years ofexperience delivering advanced NUMA solutions.

The SGI NUMA 3 architecture proVides the flexibility necessary to build a series of highly scalable servers and high-end visualization systems. The SGI 3000 family consists ofmodular systems built by using SGI NUMA building blocks installed in industry-standard 19-inch racks. Two rack sizes are offered as a starting point from which the customer can configure a system. For smaUer configurations or situations where space is limited. a 17U deskside rack is recommended. For larger systems where scalability is required, a

3 SGI 3000 Family Reference Guide

39U full-height rack can be used to build systems capable of supporting thousands of processors. A 16-processor block diagram of the SGI NUMA 3 architecture is illustrated in Figure 1. 3.2 GBlIeC full-duplex (1.6 GBlIeC each dll1lCllon) NUMAllnk3 Channels (~)

Xtown2 Ch8nnel Xt0wn2 Ch8nnel 2.4 GB/sec full-duplex 2.4 GB/sec full-duplex (1.2 GB/sec each direction) (1.2 GB/sec each direction)

(e brick) (e brick) .------f'-'-'~ : TorusIHypen:ublI :,.-i $ : :~ $'~~~: ,::'Io"YIo* I I I I ,I ,I I~- U$.;.., ::x~ $ ~ I: ~: ~_:~~------

Xt_n2 Ch8nnel NUMAllnk3 Xt0wn2 Ch8nnel 2.4 GB/sec full-duplex Channel. 2.4 GB/sec full-duplex (1.2 GBlsec each direction) (1.2 GB/sec each direction) 3.2 GB/sec full-duplex (1.6 GB/sec each direction) Figure 1. SGI 3000 Family 16-Processor Configuration

4 SGI NUMA 3 Building Blocks-Bricks

All system configurations are generated by combining SGI NUMA building blocks called "bricks.· Each brick provides a specific function and can be added independently ofother functional bricks in the system to achieve a design specifically addressing your application requirements. As bricks are added to a system, the bandwidth and performance scale in a manner that is almost linear.

The following is a list ofbricks used to build the SGI 3000 family ofsystems: • C-brick CPU module (processors and memory) • R-brick Router interconnect • I-brick Base I/O module P-brick PCI expansion X-brick XIO expansion G-brick Graphics expansion • D-brick ]BOD disk storage • Power bay N+I redundant power

5 SGr 3000 Family Reference Guide

C-brick Features: Two or four processors • Four memory banks • Bedrock memory controller with 604GB/second aggregate bandwidth 512MB to 8GB memory • 512MB, 1GB. 2GB memory options

The C-brick is a 3U rack-mountable enclosure that houses processors and memory in the system. Scaling the compute performance ofthe system is as simple as adding more C-bricks. Each C-brick contains two or four MIPS 64-bit processors, and CPU to memory bandwidth is controlled by the Bedrock switch. Two MIPS processors share a single 1.6GB/second channel to the Bedrock. and the Bedrock has two such channels. so an aggregate of up to 302GB/second ofCPU to memory bandwidth is available for local memory addressing per C-brick throughout the system. Refer to Figure 2, which illustrates two C-bricks in an 8-processor configuration.

Each C-brick includes one network interface (NI) and one I/O interface (II). These interfaces are used to attach NUMAlink interconnect cables. which in turn integrate each C-brick into the interconnect fabric. The NI channel connects the C-brick to a router in the system, and the II channel connects to a single I/O brick.

C-bricks within SGr NUMA 3 will support a mixture of processor speeds in the system. where (C.brlck) (C.brlck) each processor runs at its own native speed. This feature allows you to upgrade as necessary by adding newer, faster processors without replacing all existing processors and benefit from XIown2 Channa' XIown2 Channa' increased performance with the 2.4GB/aacfulkluplex 2.4 GB/88C fuU-duplex latest technology. (1.2GB/aac each dl18dlon) (1.2 GBIsec each dlrllCtiCln) Figure 2. SGI 3000 series 8-processor configuration

6 R-brick

The R-brick is the foundation of the SCI NUMA 3 architecture used in the SCI 3000 family. The R-brick is a 2U rack-mount enclosure that acts as a central hub connecting all C-bricks in the SCI NUMA 3 interconnect fabric. Each R-brick provides eight NUMAlink channels, one for every portofthe router crossbar. Four channels are used to connect to four C-bricks. so up to 16 processors can be interconnected to a single R-brick. The remaining four channels are responsible for communication to other Router nodes in the system. Each router node can communicate with similar router nodes, each supporting 16 processors. up to a maximum configuration of 128 compute nodes or 512 processors.

There are three versions of the R-brick: • 6-port router-standard in SCI Origin 3400 and SCI Onyx 3400. can be configured to support systems up to 32 processors 8-port Router-standard in SCI Origin 3800 and SCI Onyx 3800. can be configured to support up to 128 processors • Metarouter-for the largest configurations. supporting up to 512 processors in single shared-memory system (reqUired on systems over 128 processors)

The R-brick. like all other bricks in the SCI NUMA 3 architecture. can be independently upgraded at any point in the future. This is critical for yoursystem to support additional bandWidth and functionality required for advanced bandwidth needs of the future.

7 SGI 3000 Family Reference Guide

I-brick

The I-brick contains the base I/O for all members ofthe SGI 3000 family. One I-brick comes standard in all entry-level system configurations. The SGI 3000 series supports partitioning to improve system resiliency while maximizing access to your data. A number ofthese 4U rack-mount enclosures can be added to a system configuration to provide multiple base I/O features and multiple copies of the IRIXlR) operating system. Implementing a configuration that consists ofseveral I-bricks assists customers interested in a very flexible. resilient server with the highest level of availability. Features: • One 18GB system disk. optional 2nd drive bay for OS mirroring CD-ROM 5 PCI expansion slots Two Xtown2 1.2GB/second ports (to connect to C-brick) Two external USB ports One 10/100Base-T Ethernet port • One external IEEE 1394 port

8 P-brick

The P-brick is a PCl-based I/O expansion subsystem that is available for customers wanting server configurations with PCl I/O beyond the base I/O offered in the I-brick. Each P-brick is 4U high. comes standard in all SGI 3800 configurations, and is optional on the SGI 3200 and SGI 3400 systems. The P-brick does not merely add PCI slots to the system, itadds incremental I/O to the system. Every P-brick contains three Xbridge cWps. each supporting two separate PCI buses. for a total ofsix PCI buses. Aggregate incremental I/O that is provided by a single P-brick is more than 3GB/second. Two 110 (IO interface connectors in the rear ofthe unit allow the unit to be dual-hosted to two separate C-bricks. increasing the aVailability of the I/O in case ofa failure. Features: Provides 12 hot-plug PCI expansion slots Provides six 64-bit/66- MHz PCI busses Comes with 12 PCI carriers for hot-plug PCI Supports 3.3V or Universal PCI cards

PCI Card Carrier

The PCI card carrier is a custom-deSigned carrier that provides the ability to hot-plug PCI cards in both an I-brick and a P-brick. This feature allows you to add or remove PCI adapters wWle the system is still in operation. so your ability to change I/O requirements is flexible and dynamic while not affecting application availability. Each carrier supports both full-length or half-length PCI adapters.

9 SGI 3000 Family Reference Guide

X-brick

The X-brick is an optional I/O expansion subsystem that supports the XIO interface. Designed to support high-bandwidth applications beyond what the pcr bus can offer. the X-brick is 4U high and supports a number ofSGI XIO adapter cards. Each X-brick is powered by a single Xtown2 crossbar. whose aggregate bandwidth is 204GB/second. shared between all four XIO slots. Two I/O (II) interface connectors in the rear of the unit allow the unit to be dual-hosted to two separate C-bricks. increasing the aVailability ofthe I/O in case of a failure. Features: • Provides four expansion slots for standard XIO cards • Provides two Xtown2 ports (I.2 GB/second in each direction) Supports many existing XIO cards

10 G-brick

The C-brick is an optional ~---Klown2 expansion module that houses advanced SCI graphics subsystems. referred to as graphics pipes. The C-brick is optional onall SCI Origin 3000 setvers and is part ofthe standard base configurations Geometry Engine(Rj of the SCI Onyx 3000 series of ~~~~~4-- RMfTM systems. The C-brick supports one (2RM) @i44iiAA~4;l- RMfTM or two graphics pipes. either ~~~~~- RMfTM InfiniteReality2™ or InfiniteReality3 ~SiilSiil#ii;i+-- RMfTM ~*""~**-':&-- versions. and has a maximum DG-5 configuration ofone 4RM pipe and (4RM) one 2RM pipeline. Customers with existingOnyx2(R) InfiniteReality(R) class graphics pipelines can simply migrate to the C-brick by ordering an upgrade kit. The C-brick measures 18Uand requires both one C-brick and one I/O brick for each pipeline. In the SCI Onyx 3400 and 3800 systems, up to two C-bricks can be configured in a graphics rack, each with two pipes, supporting solutions up to 8 pipes and 16 pipes per system. respectively.

The C-brick with graphics pipeline isn'tjustan adapter that plugs into an I/O slot in the machine. but a tightly integrated graphics subsystem that has a special high-bandwidth I/O interface to deliver the extreme levels of performance necessary to generate sophisticated visualization solutions.

11 SGI 3000 Family Reference Guide

D·brlck

The D-brick provides JBOD storage capacity for the SGI 3000 family. The D-brick can be integrated into single-rack SGI 3200 and SGI 3400 systems or into the I/O rack ofSGI 3800 systems. Features: Maximum 12 dual-ported Fibre Channel disk drives Hot-pluggable drive carriers Mounts in standard 19-inch rack, 4Uhigh Single-phase input power is switchable 200 to 230 VAC Industry-standard SCA-2 interface connectors Two hot-pluggable 400 W power supplies and fans that offer full redundancy 18.2GB, 36.4GB, and 72.8GB drive capacities, all 10,000 RPM

12 PCI Cards

Maiketl....gQ*d~ < ."- ,,< ,' .....

PCI-SCSI-LVD-2P Ultra2 SCSI (80MB/sec) Single-ended 2 port card PCI-FC-IPOPT-A Fiber Channell-Port card with Fiber optic cable PCI-FC-IPCOP-A Fibre Channell-Port card with copper cable PCI-GIGENET-C Gigabit Ethernet (100/1000) copper PCI-GIGENET-OR Gigabit Ethernet (100/1000) fiber optic PCI-AUD-D1000 Digital Audio PCI-ATMOC3-IP ATMOC3card PCI-ATMOCI2-IP ATM OCl2 card

XIO Cards

MarketlngC04e ,I <../>/ .> •• <> •• > ,.< ·····i··,·· XT-HIPPI-800-SER Single Port Serial HIPPI card XT-DIVO Digital XT-DIVO-DVC Digital Video card, supports DVC XT-GSN-C-lXIO GSN Adapter single XIO slot. copper XT-GSN-C-2XIO GSN Adapter dual XIO slot, copper XT-HD High Definition video card XT-FDDI-D FDDI Dual Attach single port card XT-VME-6U VME card. supports 6U XT-VME-9U VME card, supports 9U XT-ATM-DC3C-4P Four-port ATM OC3 card

13 SGI 3000 Family Reference Guide

Power Bay

The power bay delivers power to each of the bricks that are used to configure a system. Each power bay is 3U high and houses up to six hot-swap distributed power supplies (DPS). System configurations come standard with a predetermined number ofDPSs that OPSI provide N+ I redundancy. Smaller OPS2 Of'S 3 configurations come standard with one Of'S. OPS5 power bay, while larger or more complex Of'S 6 configurations will have two. Features: Provides N+1 power • Holds a maximum of6 hot-swap DPSs • Each DPS is rated at 950 W Has eight 48 VDC output connections

Power Distribution Unit (PDU)

Both single-phase and three-phase power distribution are available.

POwers.

Slngll>-phasc PDU Three-phase PDU HU-SN-PDU-IP-NAJ single phase for domestic, Mexico. Canada. and Japan HU-SN-PDU-IP-INTL single phase for Europe and elsewhere HU-SN-PDU-3P-NAJ three phase for domestic, Mexico. Canada and Japan HU-SN-PDU-3P-INTL three phase for Europe and elsewhere

14 Short Rack

The short rack can be configured multiple ways by combining any of the standard 19-inch rack-mounted bricks up to a total of 17U. The short rack was designed for customers with space limitations orfor solutions that are satisfied by scalability to 8 processors. A minimum server configuration would require one power bay, one C-brick. and one I-brick. Features: 17U ofconfigurable space 19-inch EIA standard mounting rails • Mounted on casters 36 in. high x 25.5 in. wide x 36 in. deep One 220 VAC 10 A power distribution strip

Tall Rack

The 19-inch industry-standard tall rack is the base rack for all highly scalable server configurations. It provides maximum flexibility. allOWing customers to design a server that addresses a specific application. An endless combination ofbricks can be used to configure an ideal system. Processors, I/O bandwidth, and storage can be added independently to provide a level of granularity and flexibility not offered by any other system architecture. Features: • 39 U ofconfigurable space • 19-inch EIA standard mounting rails Mounted on casters • 74 in. high x 30 in. wide x 50 in. deep Can use any of the standard rack PDUs

15 SGI 3000 Family Reference Guide

SGI Origin 3200 Server

L2 contraDer

I-brick

PDS P-brick

C-brick

~-- C-brick

:..------Power bay

The SCI Origin 3200 server comes in one of four easy to order preconfigured bundles. Designed for customers with limited space requirements or modest scalability needs. SCI Origin 3200 is integrated into a short rack and uses the same bricks as all other models in the SCI 3000 family. A maximum of two C-bricks are supported. connected point-to-point with the NUMAlink Interconnect. SCI Origin 3200 models are routerless configurations and do not come with an R-brick.

16 SCI Origin 3200 Technical Specifications

A single 17U deskside rack configuration with no additional racks for processors orI/O:

' , .... , .. ,., ...... '. '."." .··.\i> » ,~, 1> ...... ,... . , , " .., ." .... ." ...' ....' ...... Memory size 512MB: (one C-brick with one 16GB: (two C-bricks. each 512MB bank installed) with four 2GB banks installed)

Processors One C-brick (2 processors) Two C-bricks (8 processors) Input/output One I-brick One I-brick and one optional I/O brick R-bricks None None Power bays One power bay One power bay

SGI Origin 3200 pre-configured bundles:

Standard Pro~essor ..~ .E·rocess()J" .. . Bundles' Base Configurations IU.~.lll..,.:~.a4d4a. fOS.ldbil.i.Jti.. e~i I 2 processors One 2P C-brick Add (1) 2P PIMM or Add (1) 4P C-brick 4 processors One 4P C-brick Add (1) 4P C-brick 4 processors Two 2P C-bricks Add (1-2) 2P PIMMS 8 processors Two 4P C-bricks N/A

17 SGr 3000 Family Reference Guide

SGI Onyx 3200 System

L1 system conIrol1or

D.lJrick (optional)

CIHlOMlDVD

l-brick

Although offering the same easy-to-order four bundles as SGI Origin 3200. the SGI Onyx 3200 bundles are all based on the taller 39U rack. Due to the size ofthe G-brick (necessary for graphics pipes), customers cannot upgrade SGI Origin 3200 servers with a graphics subsystem. All SGI Onyx 3200 system configurations are single-rack solutions. a benefit to customers interestedin minimizingspace. The base configuration comes with a single G-brick, to which one or two graphics pipes can be added. Additional G-bricks cannot be added. The G-brick also supports the graphics pipes from legacy Onyx2 systems, for customers wanting to migrate their existing graphics to a newer CPU architecture. As with other models of the SGI Onyx 3000 series, InfinlteReality2 or InfinlteReality3 graphics pipes are supported.

18 SCI Onyx 3200 Technical Specifications

A single 39U rack configuration with no additional racks for processors or liD:

...... ~...... ••• /» ii ...... ~ ••••...... ········r De&criPtl°n · ...... ·····i»> Memory size 512MB: (one C-brick with one 16GB: (two C-bricks. each 512MB bank installed) with four 2GB banks installed)

Processors One C-brick (2 processors) Two C-bricks (8 processors) Graphics module One G-brick (pipe not included) One G-brick Graphics pipe One InfiniteReality2 or IR3 pipe. Two graphics pipes one RM. one DG. one monitor Graphics options Keyboard. mouse. audio card. Two sets ofoptions. one for USB extender each graphics pipe Input/output One I-brick One I-brick and one optional liD brick R-bricks None None Power bays One power bay One power bay

SGI Onyx 3200 pre-configured bundles: - ... Staildard Pr~cessor w. .. Bundles Base Configurations ...... 2 processors One 2P C-brick Add (1) 2P PIMM or (1) 4P - C-brick 4 Processors One 4P C-brick Add (I) 4P C brick 4 Processors Two 2P C-bricks Add (1-2) 2P PIMMS 8 Processors Two 4P C-bricks N/A

19