Tsev > Exhibition > Ntsiab lus

Silicon Graphics High-end server lag luam

Mar 14, 2019

SGI txuas ntxiv mus txhawb nws txoj kab ntawm servers (nrog rau ib co supercomputers) raws li tus qauv SN. SN, rau Scalable ntawm, yog ib lub tshuab tsim los ntawm SGI hauv nruab nrab xyoo 1990 uas siv cache-coherent non-uniform nco nkag (cc-NUMA). Hauv SN, qhov chaw ua haujlwm, kev nco, thiab lub npav-thiab cov cim-tswj tau muab tso ua ke rau hauv ib qho chaw hu ua "node", feem ntau ntawm ib lub rooj tsav xwm ib zaug. Nodes txuas nrog los ntawm kev sib txuas lus high-speed hu ua NUMAlink (ua lag luam thaum CrayLink). Yog tsis muaj tsheb npav, thiab siv tau ntawm cov txheej txheem, kev nco, thiab I / O khoom siv los ntawm kev hloov cov ntaub ntawv thiab cov chaw khiav haujlwm.


Ua tsaug rau kev sib koom ua ke ntawm kev faib tawm, SN systems teev ob peb zaug ntawm ib zaug: raws li CPU suav nce ntxiv, yog li qhov cim xeeb, I / O peev, thiab qhov system kab kev ywj pheej. Qhov no tso cai rau lub cim xeeb ntawm tag nrho cov nodes yuav tsum tau nkag mus hauv ib qho duab OS uas yog siv cov txheej txheem sib koom ua ke-kev sib hloov synchronization. Qhov no ua rau SN cov kab ke kom yooj yim dua rau qhov kev zov me nyuam thiab ua kom nce siab dua mus rau qhov siab tshaj qhov tsis muaj cache-coherent systems zoo li cov pawg ua ke los yog massively rau hauv cov computers uas yuav tsum tau siv daim ntawv sau thov kom sau (los yog rov sau) los ua kom meej meej xov-kev sib txuas lus ntawm lawv cov ntshav.


Thawj SN uas, hu ua SN-0, raug tso tawm hauv 1996 nyob rau hauv lub npe khoom npe Oriental 2000. Raws li MIPS R10000 processor, nws scaled ntawm 2 mus rau 128 processors thiab ib tug me me version, lub hauv paus chiv keeb 200 (SN-00) scaled los ntawm 1 mus rau 4. Tom qab kev tshaj tawm muaj peev xwm ua kom zoo li muaj li 512 processors.


Qhov thib ob lub hauv kev ua qauv, thawj coj hu ua SN-1 tab sis tom qab SN-MIPS, raug tso tawm hauv Lub Xya Hli 2000, ua lub Keeb Kwm 3000. Nws tau ua tiav ntawm 4 txog 512 processors, thiab 1,024-processor configurations tau xa los ntawm kev txiav txim tshwj xeeb rau qee tus neeg muas zaub. Ib qho me, tsawg dua qhov ntsuas tau ua raws, hu ua Keeb Kwm 300.


Thaum lub Kaum Ib Hlis Ntuj xyoo 2002, SGI tau tshaj tawm lub rebuilt ntawm nws qhov SN, raws li lub npe keeb kwm 3900. Nws quadrupled qhov chaw ntawm cheeb tsam ntawm SN-MIPS, los ntawm 32 txog 128 processors ib khib thaum tsiv mus rau "rog ntoo" interconnect topology .


Nyob rau lub Ib Hlis 2003, SGI tshaj tawm ib yam txawv ntawm SN platform hu ua Altix 3000 (hauv kev hu ua SN-IA). Nws siv Intel Itanium 2 processors thiab khiav lub tshuab khiav ua haujlwm Linux. Thaum lub sij hawm nws tau tso tawm, nws yog lub ntiaj teb feem ntau scalable Linux raws computer, uas txhawb txog 64 processors nyob rau hauv ib qho system ntawm [57] Nodes yuav txuas nrog kev siv tib lub tshuab NUMAlink los tsim cov lus SGI los ua "superclusters".


Nyob rau lub Ob Hlis 2004, SGI tshaj tawm kev them nyiaj yug rau 128 processor nodes los ua raws li 256 thiab 512 processor versions xyoo.


Nyob rau hauv lub Plaub Hlis 2004, SGI tshaj tawm tias nws txoj kev lag luam ntawm Alias software lag luam rau thaj tsam li ntawm $ 57 lab.


Thaum Lub Kaum Hli Ntuj xyoo 2004, SGI tau ua tus Supercomputer Columbia, uas ua rau lub ntiaj teb cov ntaub ntawv ceev ceev, rau NASA Ames Research Center. Nws yog ib pawg ntawm 20 Altix supercomputers txhua nrog 512 Intel Itanium 2 processors khiav Linux, thiab tau nce ceev ceev ntawm 42.7 trillion floating-point operations thib ob (teraflops), yooj yim topping Nyij Pooj lub ntiaj teb Simulator cov ntaub ntawv ntawm 35.86 teraflops. (Ib lub lim tiam tom qab, IBM tau kho dua tshiab xoob xoob xoob / L hauv lub sijhawm ntawm 70.7 teraflops).


Nyob rau hauv Lub Xya Hli Ntuj 2006, SGI tau tshaj tawm SGI Altix 4700 nrog 1,024 processors thiab 4 TB ntawm kev nco uas khiav ib qho duab hauv Linux.