I-Cisco Performance Tuning ye-UCS M8 Platforms
Inhloso yedokhumenti nobubanzi
I-Basic Input-and-Output System (BIOS) ihlola futhi iqalise izingxenye zehadiwe zesistimu futhi iqalise isistimu yokusebenza kusukela kudivayisi yokugcina. Isistimu yokubala evamile inezilungiselelo ezimbalwa ze-BIOS ezilawula ukuziphatha kwesistimu. Ezinye zalezi zilungiselelo zihlobene ngokuqondile nokusebenza kwesistimu.
Lo mbhalo uchaza izilungiselelo ze-BIOS ezivumelekile kumaseva e-Cisco Unified Computing System™ (Cisco UCS®) M8 ane-AMD EPYC™ 4th Gen kanye ne-5th Gen processors. Ichaza indlela yokuthuthukisa izilungiselelo ze-BIOS ukuze kuhlangatshezwane nezidingo zokusebenza kahle nokusebenza kahle kwamandla kuma-Cisco UCS X215c M8 Compute Nodes, Cisco UCS C245 M8 Rack Servers, kanye ne-Cisco UCS C225 M8 Rack Servers.
Lo mbhalo futhi uxoxa ngezilungiselelo ze-BIOS ezingakhethelwa izinhlobo ezihlukahlukene zomthwalo wokusebenza kumaseva e-Cisco UCS M8 anamaphrosesa we-AMD EPYC 4th Gen kanye no-5th Gen. Ukuqonda izinketho ze-BIOS kuzokusiza ukhethe amanani afanelekile ukuze uzuze ukusebenza kwesistimu okuhle kakhulu.
Lo mbhalo awuxoxi ngezinketho ze-BIOS zokukhishwa kwe-firmware ethile yamaseva e-Cisco UCS M8 okusekelwe ku-AMD EPYC 4th kanye ne-5th Gen processors. Izilungiselelo eziboniswe lapha zijwayelekile.
Yini ozoyifunda
Inqubo yokusetha izinketho zokusebenza kusistimu yakho ye-BIOS ingaba yinkimbinkimbi futhi idide, futhi ezinye zezinketho ongazikhetha azicacile. Ngezinketho eziningi, kufanele ukhethe phakathi kokuthuthukisa iseva ukuze wonge amandla noma ukuze usebenze. Lo mbhalo uhlinzeka ngeziqondiso ezijwayelekile kanye neziphakamiso zokukusiza ukuthi uzuze ukusebenza kahle kumaseva akho e-Cisco UCS M8 asebenzisa i-4th Gen kanye ne-5th Gen AMD EPYC yomndeni yama-CPU.
Amaphrosesa we-AMD EPYC 9004 Series
Amaphrosesa we-AMD EPYC 9004 Series akhiwe ngama-cores we-Zen 4 kanye ne-AMD Infinity architecture. Amaphrosesa we-AMD EPYC 9004 Series ahlanganisa amakhompiyutha, izilawuli zenkumbulo, abalawuli be-I/O, Ukuthembeka, Ukutholakala, kanye Nokusebenza (i-RAS), nezici zokuphepha zibe Uhlelo oludidiyelwe ku-Chip (SoC). I-AMD EPYC 9004 Series Processor igcina ukwakheka okuqinisekisiwe kwe-Multi-Chip Module (MCM) Chiplet yamaphrosesa aphumelele we-AMD EPYC ngenkathi yenza ukuthuthuka okwengeziwe ezingxenyeni ze-SoC. I-SoC ihlanganisa i-Core Complex Dies (CCDs), equkethe i-Core Complexes (CCXs), equkethe ama-cores asekelwe ku-Zen 4-4.
Amaphrosesa we-AMD EPYC 9004 Series asekelwe ku-Zen 4 compute core entsha. I-Zen 4 core yakhiwe kusetshenziswa inqubo ye-5nm futhi iklanyelwe ukunikeza Imiyalo ngomjikelezo ngamunye (IPC) ukuphakamisa kanye nokuthuthukiswa kwefrequency ngaphezu kwama-Zen cores esizukulwane sangaphambili. Ingqikithi ngayinye inenqolobane enkulu ye-L2 nokusebenza okuthuthukisiwe kwenqolobane esizukulwaneni sangaphambilini.
Umongo ngamunye usekela i-Multithreading ehambisanayo (i-SMT), eyenza imicu emibili yehadiwe ehlukene isebenze ngokuzimela, yabelane ngenqolobane ehambisanayo ye-L2.
I-Core Complex (CCX) yilapho ama-cores afinyelela ku-Zen 4-based abelana nge-L3 noma i-Last Level Cache (LLC). Ukunika amandla i-Multithreading Ehambisanayo (SMT) kuvumela i-CCX eyodwa ukuthi isekele imicu yehadiwe efika ku-16 ngesikhathi esisodwa.
Amaphrosesa we-AMD EPYC 9004 Series afaka ubuchwepheshe be-AMD 3D V-Cache die-stacking eyenza ama-9700 Series processors azuze ukuhlanganiswa kwe-chiplet okusebenza kahle kakhulu. I-AMD 3D Chiplet ye-architecture inqwabelanisa amathayela enqolobane ye-L3 ngokuqondile ukuze inikeze kufika ku-96MB ye-L3 inqolobane ngokufa (futhi kufikela ku-1 GB L3 Cache isokhethi ngalinye) kuyilapho isahlinzeka ngokuhambisana kwesokhethi nawo wonke amamodeli wokucubungula we-AMD EPYC 9004 Series.
Amaphrosesa we-AMD EPYC 9004 Series anobuchwepheshe be-AMD 3D V-Cache asebenzisa ukupakishwa okunengqondo okuhamba phambili embonini okusekelwe kunqubo ye-chip-on-wafer ye-copper-to-copper hybrid bonding "bumpless" ukuze unike amandla ngaphezu kuka-200X ukuminyana koxhumano lobuchwepheshe bamanje be-2D (kanye nangaphezu kuka-15X okunye ukuminyana kwe-3D) ihumusha ukubambezeleka okuphansi, umkhawulokudonsa ophezulu, namandla amakhulu nokusebenza kahle kwe-thermal.
Ama-CCD axhuma kumemori, i-I/O, kanye nokunye nge-I/O Die (IOD) ebuyekeziwe. Le ndwangu ye-AMD Infinity emaphakathi inikeza indlela yedatha nokwesekwa kokulawula ukuze kuxhunywe ama-CCX, inkumbulo, kanye ne-I/O. I-CCD ngayinye ixhuma ku-IOD ngesixhumanisi esizinikezele se-Global Memory Interconnect (GMI) esinesivinini esikhulu. I-IOD isiza ukugcina ukubumbana kwenqolobane futhi ngaphezu kwalokho inikeza isixhumi esibonakalayo ukuze inwebe indwangu yedatha kumcubunguli wesibili ongaba khona nge-xGMI, noma izixhumanisi ze-G. Amaphrosesa we-AMD EPYC 9004 Series asekela kufika ku-4 xGMI (noma izixhumanisi ze-G) ngesivinini esifika ku-32Gbps.
I-IOD idalula iziteshi zememori ze-DDR5, i-PCIe Gen5, i-CXL 1.1+, nezixhumanisi ze-Infinity Fabric. I-IOD ihlinzeka ngezilawuli zeMemori Ezihlanganisiwe (UMCs) eziyishumi nambili ezisekela inkumbulo ye-DDR5.
I-UMC ngayinye ingakwazi ukusekela amamojula angu-2 eMemori Ekabili Ekulayini (ama-DIMM) esiteshini (i-DPC) ngesilinganiso esiphezulu sama-DIMM angu-24 isokhethi ngalinye. Amaphrosesa we-4th Gen AMD EPYC angakwazi ukusekela inkumbulo efika ku-6TB ye-DDR5 isokhethi ngalinye. Ukuba neziteshi zememori ezengeziwe futhi ezisheshayo uma kuqhathaniswa nezizukulwane ezedlule zamaphrosesa we-AMD EPYC kunikeza umkhawulokudonsa wememori owengeziwe ukuze uphakele amaphrosesa wokubala okuphezulu. Ukuphuma kwenkumbulo eziteshini ezingu-2, 4, 6, 8, 10, kanye ne-12 kusiza ukulungiselelwa ukuhlukahluka komsebenzi kanye nokulungiselelwa kwememori.
Iphrosesa ngayinye ingaba nesethi yezixhumanisi ezingu-4 ze-P kanye nezixhumanisi ezingu-4 ze-G. Idizayini ye-OEM motherboard ingasebenzisa isixhumanisi se-G ukuze ixhume kuphrosesa yesibili ye-4th Gen AMD EPYC noma inikeze imizila eyengeziwe ye-PCIe Gen5. Amaphrosesa we-4th Gen AMD EPYC asekela amasethi afika kwayisishiyagalombili emizila ye-x16-bit I/O, okungukuthi, imizila engu-128 ye-PCIe Gen5 enesivinini esikhulu kumapulatifomu anesokhethi eyodwa kanye nemizila efika kweyi-160 kumapulatifomu anesokhethi ekabili.
Amaphrosesa we-AMD EPYC 9004 Series 4th Gen akhiwe ngokucaciswa okusohlwini lweThebula 1.
Ithebula 1. Ukucaciswa kwephrosesa ye-AMD EPYC 9004 Series 4th Gen
| Into | Ukucaciswa |
| Cores inqubo ubuchwepheshe | 5-nanometer (nm) Zen 4 |
| Inombolo enkulu yama-cores | 128 |
| Isivinini esiphezulu sememori | 4800 Mega-Transfer ngomzuzwana (MT/s) |
| Ubuningi beziteshi zememori | 12 isokhethi ngalinye |
| Umthamo omkhulu wenkumbulo | 6 TB isokhethi ngalinye |
| I-PCI | Imizila engu-128 (ubuningi) yesokhethi engu-1
Imizila engu-160 (ubuningi) ye-PCIe ye-2-socket Gen 5 |
Ukuze uthole ulwazi olwengeziwe mayelana ne-Microarchitecture ye-AMD EPYC 9004 Series processors, bheka Kuphelileview ye-AMD EPYC 9004 Series Processors Microarchitecture.
Amaphrosesa we-AMD EPYC 9005 Series
Systems based on 5th Gen AMD EPYC processors can support IT initiatives from data-center consolidation and modernization to increasingly demanding enterprise application needs. These systems can enable expanding AI within the enterprise while supporting business imperatives to improve energy efficiency and rein in data-center sprawl through high-density support for virtualization and cloud environments. Modernizing IT infrastructure is key to freeing up space and energy to accommodate AI and other innovative business initiatives within existing data-center footprints.
Abaprosesa be-AMD EPYC bazuze ngokuqhubekayo okuzuziwe okunamadijithi amabili ekusebenzeni komjikelezo wewashi ngalinye (IPC) ngesizukulwane esisha ngasinye, kanti umnyombo wakamuva we-Zen 5 kuma-processor we-5th Gen AMD EPYC uletha ukukhushulwa okubalulekile kwe-ML, i-HPC, kanye nomsebenzi webhizinisi. I-Zen 5c yethu ethuthukisiwe eyenziwe kahle inika amandla ama-CPU ngesibalo esiphezulu somsuka kunoma yimaphi amaphrosesa e-x86-architecture, alethe ukuminyana okuphezulu kakhulu okuyisisekelo kokusebenza okubonakalayo kanye nokusebenza kwamafu.
Amaphrosesa we-5th Gen AMD EPYC akuvumela ukuthi uhlukanise futhi ubhekane nendawo yonke eqhubekayo ebanzi yezidingo zomthwalo womsebenzi. I-hybrid yethu, i-multichip architecture isenza sikwazi ukuhlukanisa izindlela zokusungula futhi silethe imikhiqizo emisha eqhubekayo, esebenza kahle kakhulu. Ama-cores we-Zen 5 kanye ne-Zen 5c amele enye inqubekelaphambili ebalulekile evela esizukulwaneni sakamuva kakhulu, ngokusekelwa okusha kwezinhlelo zokusebenza eziyinkimbinkimbi kakhulu zokufunda umshini kanye ne-inferencing.
Kumaphrosesa we-5th Gen AMD EPYC, sisebenzisa ama-cores amabili ahlukene ukuze sibhekane nezinhlobonhlobo zezidingo zomthwalo womsebenzi ngokushintsha uhlobo nenani lama-cores nendlela esiwapakisha ngayo.
Zen 5 core
Lo mgogodla ulungiselelwe ukusebenza okuphezulu. Kufike ama-cores ayisishiyagalombili ahlanganiswe ukuze kwakhiwe i-core complex (CCX) ehlanganisa inqolobane ye-L3 eyabelwe engu-32-MB. Le ngxubevange yenzelwe i-die (CCD), efika ku-16 engalungiselelwa ibe iphrosesa ye-EPYC 9005 kufika kuma-cores angu-128 ku-SP5 form factor. Uma kuqhathaniswa nesizukulwane sangaphambilini, ama-processor we-5th Gen AMD EPYC, anikwa amandla yi-Zen 5 core ethuthukisiwe, kanye nenkumbulo esheshayo nokunye ukuthuthukiswa kwe-CPU okubalulekile, ahlinzeka ngamaphesenti angu-20 inani elikhulu kanye nokusebenza kwamaphuzu antantayo okungamaphesenti angu-34 kumaphrosesa angu-64-core asebenza ngaphakathi kwebanga elifanayo le-360W TDP 9xx5-050-070, 070, 9x5-070, 070, 070, 9
Zen 5c core
Lo mgogodla ulungiselelwe ukuminyana nokusebenza kahle. Inomqondo ofanayo wokudlulisa irejista njenge-Zen 5 core, kodwa ukwakheka kwayo okubonakalayo kuthatha indawo encane futhi yakhelwe ukuletha ukusebenza okwengeziwe nge-watt ngayinye. I-Zen 5c core complex ihlanganisa kufika ku-16 cores kanye ne-cache engu-32-MB L3 eyabiwe. Kufikela ku-12 kulawa ma-CCD angahlanganiswa ne-I/O CCD ukuletha ama-CPU anama-cores afika kwangu-192 ku-SP5 form factor.
Amaphrosesa we-AMD EPYC 9005 Series 5th Gen akhiwe ngokucaciswa okusohlwini lweThebula 2.
Ithebula 2. Ukucaciswa kwephrosesa ye-AMD EPYC 9005 Series yesi-5
| Into | Ukucaciswa |
| Cores inqubo ubuchwepheshe | 4-nanometer (nm) Zen 5 kanye ne-3-nanometer Zen 5c |
| Inombolo enkulu yama-cores | 192 |
| Ubuningi benqolobane ye-L3 | 512 MB |
| Isivinini esiphezulu sememori | 6000 Mega-Transfer ngomzuzwana (MT/s) |
| Ubuningi beziteshi zememori | 12 isokhethi ngalinye |
| Umthamo omkhulu wenkumbulo | 6 TB isokhethi ngalinye |
| I-PCI | Imizila eyi-128 (ubuningi.) yemizila eyi-1-socket 160 (ubukhulu.) ye-2-socket PCIe Gen 5 |
Qaphela: Izingxenyekazi ze-Cisco UCS M8 zisekela kuphela ama-cores angu-160 400W TDP we-Zen 5c processors.
Ukuze uthole ulwazi olwengeziwe mayelana ne-AMD EPYC 9005 Series 5th Gen processors microarchitecture, bheka Kuphelileview ye-AMD EPYC 9005 Series Processors Microarchitecture.
I-topology ye-Non-Uniform Memory Access (NUMA).
I-AMD EPYC 9004 kanye ne-9005 Series processors isebenzisa ukwakheka kwe-Non-Uniform Memory Access (NUMA) lapho ukubambezeleka okuhlukile kungase kube khona kuye ngokuba seduze kwe-processor core kumemori nezilawuli ze-I/O. Ukusebenzisa izinsiza ngaphakathi kwendawo efanayo ye-NUMA kunikeza ukusebenza okuhle okufanayo, kuyilapho ukusebenzisa izinsiza kumanodi ahlukahlukene kukhulisa ukubambezeleka.
Umsebenzisi angalungisa ukulungiselelwa kwe-BIOS ye-NUMA Nodes Per Socket (NPS) ukuze athuthukise le topology ye-NUMA endaweni yawo ethile yokusebenza kanye nomthwalo wokusebenza. Okwesiboneloample, ukusetha i-NPS=4 ihlukanisa iphrosesa ibe amaquadrants, lapho isinenja ngasinye sinama-CCD angu-3, ama-UMC angu-3, kanye nehabhu le-I/O elingu-1. Ibanga eliseduze le-I/O le-processor-memory liphakathi kwama-cores, inkumbulo, nama-peripherals e-I/O ngaphakathi kwequadrant efanayo. Ibanga elikude kakhulu liphakathi komgogodla nesilawuli senkumbulo noma ihabhu ye-I/O kumaquadrants anqamulanayo (noma enye iphrosesa ekucushweni kwe-2P). Indawo yama-cores, inkumbulo, nehabhu/amadivayisi e-IO kusistimu esekelwe ku-NUMA iyisici esibalulekile lapho ushuna ukusebenza.
Kumaphrosesa we-4th Gen EPYC, ukulungiselelwa kokuxhumanisa kwe-Infinity Fabric kunciphise umehluko wokubambezeleka nakakhulu. Kusetshenziswa amaphrosesa we-EPYC 9004 Series, kuzinhlelo zokusebenza ezidinga ukuminyanisa iphesenti elilodwa noma amabili wokugcina ukubambezeleka ngaphandle kwereferensi yememori, ukudala ukuhambisana phakathi kwebanga lememori kanye ne-CPU efayo (Zen 4 noma i-Zen 4c) kungathuthukisa ukusebenza. Umfanekiso 1 ubonisa ukuthi lokhu kusebenza kanjani. Uma uhlukanisa i-I/O die ibe amaquadrants amane ngokucushwa kwe-NPS=4, uzobona ukuthi ama-DIMM ayisithupha adla abe izilawuli zememori ezintathu, ezixhunywe eduze nge-Infinity Fabric (GMI) kusethi yokufa kwe-Zen 4 CPU, noma kufika kuma-CPU cores angama-24.
Umfanekiso:1
I-AMD EPYC 4th Gen 4th Gen processor block diagram enezizinda ze-NUMA
Ku-5th Gen EPYC processors, ukuthuthukiswa okwenziwe ku-AMD Infinity Fabric interconnects kuye kwanciphisa umehluko we-latency nakakhulu. Kusetshenziswa amaphrosesa we-EPYC 9005 Series, kuzinhlelo zokusebenza ezidinga ukuminyanisa iphesenti elilodwa noma amabili wokugcina ukubambezeleka ngaphandle kwereferensi yememori, ukudala ukuhambisana phakathi kwebanga lememori kanye ne-CPU efayo (Zen 5 noma i-Zen 5c), kungathuthukisa ukusebenza. Umfanekiso wesi-2 ubonisa ukuthi lokhu kusebenza kanjani. Uma uhlukanisa i-I/O die ibe amaquadrants amane ngokucushwa kwe-NPS=4, uzobona ukuthi ama-DIMM ayisithupha adla abe izilawuli zenkumbulo ezintathu, ezixhunywe eduze nge-Infinity Fabric (GMI) kusethi ye-Zen 5 CPU efika kwezine iyafa noma kufika ku-Zen 5c CPU emithathu iyafa.
Umfanekiso:2
I-AMD EPYC 5th Gen 4th Gen processor block diagram enezizinda ze-NUMA
I-NPS1
Ukusetha kwe-NPS=1 kukhombisa inodi eyodwa ye-NUMA isokhethi ngalinye. Lesi silungiselelo silungiselela zonke iziteshi zememori ku-processor zibe inodi eyodwa ye-NUMA. Wonke ama-processor cores, yonke inkumbulo enamathiselwe, nawo wonke amadivayisi we-PCIe axhunywe ku-SoC akuleyo nodi eyodwa ye-NUMA. Imemori ixhunywe kuwo wonke amashaneli ememori kusiphrosesa ibe yindawo eyodwa yekheli.
I-NPS2
Ukulungiselelwa kwe-NPS=2 kumisa iphrosesa ngayinye ibe yizizinda ezimbili ze-NUMA ezihlanganisa ingxenye yamacores nengxenye yamashaneli ememori ibe isizinda esisodwa se-NUMA, kanye namashaneli asele namashaneli enkumbulo abe yisizinda sesibili se-NUMA. Inkumbulo ixhunywe phakathi kwamashaneli enkumbulo ayisithupha esizindeni ngasinye se-NUMA. Amadivayisi we-PCIe azoba sendaweni eyodwa kwezimbili ze-NUMA kuye ngokuthi ingxenye ene-PCIe root complex yaleyo divayisi.
I-NPS4
Ukulungiselelwa kwe-NPS=4 kwehlukanisa iphrosesa ibe yizinodi ezine ze-NUMA isokhethi ngalinye nequadrant ngayinye enengqondo emiswe njengesizinda sayo se-NUMA. Inkumbulo ixhunywe kumashaneli enkumbulo ahlotshaniswa nequadrant ngayinye. Amadivayisi e-PCIe azoba sendaweni kwesinye sezizinda ezine zephrosesa ye-NUMA, kuye ngequadrant ye-IOD ene-PCIe ehambisanayo yempande yaleyo divayisi. Wonke amashaneli ememori amabili axhunyiwe. Lokhu kunconyelwe i-HPC neminye imisebenzi ehambisana kakhulu. Kufanele usebenzise i-NPS4 lapho uqalisa izinhlelo ze-Windows nge-CPU SMT enikwe amandla amaphrosesa e-AMD EPYC anama-cores angaphezu kuka-64, ngoba i-Windows ikhawulela usayizi weqembu le-CPU kubukhulu obungama-core cores angu-64.
Qaphela: Kuzinhlelo ze-Windows, qinisekisa ukuthi inani lamaphrosesa anengqondo ngenodi ngayinye ye-NUMA <=64 ngokusebenzisa i-NPS2 noma i-NPS4 esikhundleni se-NPS1 ezenzakalelayo.
I-NPS0 (ayinconyiwe)
Ukulungiselelwa kwe-NPS=0 kukhombisa isizinda esisodwa se-NUMA salo lonke uhlelo (kuwo womabili amasokhethi ekucushweni okuyizisekelo ezimbili). Lesi silungiselelo silungiselela zonke iziteshi zememori ohlelweni zibe inodi eyodwa ye-NUMA. Imemori ixhunywe kuzo zonke iziteshi zememori ohlelweni ingene endaweni eyodwa yekheli. Wonke ama-processor cores kuwo wonke amasokhethi, yonke inkumbulo enamathiselwe, nawo wonke amadivayisi we-PCIe axhunywe kunoma iyiphi iphrosesa akuleso sizinda esisodwa se-NUMA.
Inqolobane ye-Layer 3 njengesizinda se-NUMA
Ngokungeziwe kuzilungiselelo ze-NPS, enye inketho ye-BIOS yokushintsha ukucushwa kwe-NUMA iyatholakala. Ngenketho Yenqolobane Yesendlalelo 3 njenge-NUMA (L3CAN), inqolobane ngayinye Yesendlalelo 3 (eyodwa nge-CCD) ivezwa njengenodi yayo ye-NUMA. Okwesiboneloample, iphrosesa eyodwa enama-CCD angu-8 ingaba namanodi angu-NUMA: eyodwa ku-CCD ngayinye. Kulokhu, uhlelo olunamasokhethi amabili luzoba nengqikithi yamanodi angu-16 e-NUMA.
Izilungiselelo zephrosesa
Lesi sigaba sichaza izinketho zephrosesa ongazilungiselela.
Imodi ye-CPU SMT
Ungasetha inketho ye-CPU Simultaneous Multithreading (CPU SMT) ukuze unike amandla noma ukhubaze ama-core processor cores kumaphrosesa asekela inketho yemodi ye-AMD CPU SMT. Uma imodi ye-CPU SMT isethelwe kokuthi Okuzenzakalelayo (kunikwe amandla), umongo ngamunye wephrosesa ophathekayo usebenza njengamacore processor amabili anengqondo futhi uvumela izinhlelo zokusebenza zesofthiwe enemicu eminingi ukuthi zicubungule imicu ngokuhambisana ngaphakathi kwephrosesa ngayinye.
Eminye imithwalo yemisebenzi, ehlanganisa eminingi ye-HPC, ibona umphumela omaphakathi noma ngisho nokusebenza kabi lapho i-CPU SMT inikwe amandla. Ezinye izinhlelo zokusebenza, futhi hhayi nje umongo womzimba, zinikezwe ilayisense ngochungechunge lwehadiwe njengoba lunikwe amandla. Ngalezo zizathu, ukukhubaza i-CPU SMT kuphrosesa yakho ye-EPYC 9004 Series kungase kufiseke. Ngaphezu kwalokho, amanye amasistimu wokusebenza awanakho usekelo lwe-x2APIC ngaphakathi kwephrosesa ye-EPYC 9004 Series enikwe amandla, edingekayo ukuze isekele ukudlula imicu engu-255. Uma usebenzisa isistimu yokusebenza engakusekeli ukuqaliswa kwe-x2APIC ye-AMD, futhi unamaphrosesa amabili angu-64-core afakiwe, uzodinga ukukhubaza i-CPU SMT. Ithebula 3 lifingqa izilungiselelo.
Kufanele uhlole inketho ye-CPU hyperthreading kokubili inikwe amandla futhi ikhutshaziwe endaweni yakho ethile. Uma usebenzisa uhlelo lokusebenza olunochungechunge olulodwa, kufanele ukhubaze i-hyperthreading.
Ithebula 3. Izilungiselelo ze-CPU SMT
| Ukusetha | Izinketho |
| Ukulawula i-CPU SMT | ● Okuzenzakalelayo: isebenzisa izintambo ezimbili zehadiwe ngomongo ngamunye
● Khubaza: isebenzisa intambo eyodwa yehadiwe ngomongo ngamunye ● Nika amandla: isebenzisa intambo yehadiwe ephindwe kabili ngomongo ngamunye |
Vikela Imodi Yomshini Obonakalayo (SVM).
Imodi evikelekile yomshini obonakalayo (i-SVM) inika amandla izici zephrosesa futhi ivumela inkundla ukuthi isebenzise amasistimu wokusebenza amaningi nezinhlelo zokusebenza kuma-partitions azimele. Imodi ye-AMD SVM ingasethelwa kunoma yiliphi lamanani alandelayo:
- Ikhutshaziwe: iphrosesa ayikuvumeli ukwenziwa kwezinto ezibonakalayo.
- Inikwe amandla: iphrosesa ivumela amasistimu wokusebenza amaningi kuma-partitions azimele.
Uma isimo sohlelo lwakho lokusebenza singadingi ukwenziwa kwezinto ezibonakalayo, vala i-AMD ubuchwepheshe bokwenza izinto. Ngemuva kokuthi i-virtualization ivaliwe, vala futhi inketho ye-AMD IOMMU, engabangela umehluko ekubambezelekeni kokufinyelela kwememori. Ithebula lesi-4 lifingqa izilungiselelo.
Ithebula 4. Izilungiselelo zenketho ye-virtualization
| Ukusetha | Izinketho |
| I-SVM | ● Inikwe amandla
● Ikhutshaziwe |
I-DF C-states
Ngokufana nama-CPU cores, i-AMD Infinity Fabric ingangena ezifundeni zamandla aphansi ngenkathi ingenzi lutho. Kodwa-ke, kuzoba nokubambezeleka lapho ushintsha ubuyela kumodi yamandla agcwele, okubangela i-latency jitter. Emsebenzini wokubambezeleka okuphansi noma one-I/O eqhumayo, ungakhubaza isici se-Data Fabric (DF) C-states ukuze uzuze ukusebenza okwengeziwe, ngokuhwebelana kokusetshenziswa kwamandla okuphezulu. Ithebula lesi-5 lifingqa izilungiselelo.
Ithebula 5. I-DF C-states
| Ukusetha | Izinketho |
| I-DF C-states | ● Okuzenzakalelayo/Kunikwe amandla: kuvumela i-AMD Infinity Fabric ukuthi ingene kusimo samandla aphansi
● Ikhutshaziwe: ivimbela i-AMD Infinity Fabric ukuthi ingene esimweni samandla aphansi |
Inqolobane ye-ACPI SRAT L3 njengesizinda se-NUMA
Uma Inqolobane ye-ACPI SRAT L3 njengesilungiselelo sesizinda se-NUMMA sivuliwe, inqolobane ngayinye Yesendlalelo-3 ivezwa njengenodi ye-NUMA. Ngesilungiselelo se-Layer 3 Cache njenge-NUMA Domain (L3CAN), inqolobane ngayinye ye-Layer-3 (eyodwa nge-CCD) ivezwa njengenodi yayo ye-NUMA. Okwesiboneloample, iphrosesa eyodwa enama-CCD angu-8 ingaba namanodi angu-NUMA: eyodwa ku-CCD ngayinye. Uhlelo olukabili lweprosesa lungaba nengqikithi yamanodi angu-16 e-NUMA.
Lesi silungiselelo singathuthukisa ukusebenza kwemithwalo elungiselelwe kakhulu ye-NUMA uma imithwalo yomsebenzi noma izingxenye zomsebenzi zingaphinwa kumacores ku-CCX futhi uma zingazuza ngokwabelana ngenqolobane ye-Layer-3. Uma lesi silungiselelo sikhutshaziwe, izizinda ze-NUMA zibonwa ngokuya ngesethingi yepharamitha ye-NUMA NPS.
Amanye amasistimu okusebenza nama-hypervisors awawenzi ukuhlelwa kolwazi Lwe-Layer 3, futhi eminye imithwalo yomsebenzi iyazuza ngokumenyezelwa kwe-Layer 3 njengesizinda se-NUMA. Ithebula lesi-6 lifingqa izilungiselelo.
Ithebula 6. I-ACPI SRAT Cache 3 Cache njengezilungiselelo zesizinda se-NUMA
| Ukusetha | Izinketho |
| Inqolobane ye-ACPI SRAT L3 Njengesizinda se-NUMA | ● Okuzenzakalelayo (kukhutshaziwe)
● Khubaza: ayibiki inqolobane ngayinye ye-Layer-3 njengesizinda se-NUMA ku-OS ● Nika amandla: ibika inqolobane ngayinye Yesendlalelo-3 njengesizinda se-NUMA ku-OS |
I-Algorithm Performance Boost Disable (APBDIS)
Ikuvumela ukuthi ukhethe inani lokukhubaza i-Algorithm Performance Boost (APB) le-SMU. Esimeni esimisiwe, i-AMD Infinity Fabric ikhetha phakathi kwewashi lendwangu elinamandla aphelele namandla aphansi kanye newashi lenkumbulo, ngokusekelwe ekusetshenzisweni kwendwangu nokukhumbula. Nokho, kwezinye izimo ezifaka umkhawulokudonsa ophansi kodwa ithrafikhi ezwelayo
(nokuhlola ukubambezeleka kwenkumbulo), Ushintsho olusuka emandleni aphansi luye emandleni aphelele lungaphazamisa ukubambezeleka. Ukusetha i-APBDIS kokuthi 1 (ukuze ukhubaze i-Algorithm Performance Boost [APB]) futhi ucacise i-Infinity Fabric P-state engu-0 kuzophoqa i-Infinity Fabric nezilawuli zenkumbulo zibe yimodi yamandla agcwele, kususe noma iyiphi i-jitter enjalo yokubambezeleka. Amaphrosesa athile e-CPU nezinketho zenani lenkumbulo kuphumela esimeni lapho ukusetha i-Infinity Fabric P- state engu-1 kuzonciphisa ukubambezeleka kwenkumbulo ngenani lomkhawulokudonsa wenkumbulo. Lokhu kulungiselelwa kungase kuzuzise izinhlelo zokusebenza ezaziwa ngokuzwela ukubambezeleka kwenkumbulo. Ithebula lesi-7 lifingqa izilungiselelo.
Ithebula 7. Ukulungiselelwa kwe-APBDIS
| Ukusetha | Izinketho |
| I-APBDIS | ● Okuzenzakalelayo (0): kusetha i-APBDIS ezenzakalelayo ye-SMU. Lena inketho ezenzakalelayo.
● 0: ishintsha ngokuguqukayo i-Infinity Fabric P-state ngokusekelwe ekusetshenzisweni kwesixhumanisi ● 1: inika amandla isilawuli se-Infinity Fabric P-state esilungisiwe |
I-SOC P-State elungisiwe SP5F 19h
Iphoqa i-P-state ukuthi izimele noma incike, njengoba kubikwe into ye-ACPI _PSD. Ishintsha i-SOC P-State uma i-APBDIS inikwe amandla. lapho, u-F ebhekisela emndenini wokucubungula.
| Ukusetha | Izinketho |
| I-SOC P-State elungisiwe SP5F 19h | ● I-P0: I-Infinity Fabric P-state esebenza kakhulu
● I-P1: i-Infinity Fabric P-state ehamba phambili ephezulu kakhulu ● I-P2: i-Infinity Fabric P-state ehamba phambili elandelayo ephezulu kakhulu ngemva kwe-P1 |
Izilungiselelo ze-xGMI: ukuxhumana phakathi kwamasokhethi
Ohlelweni olunamasokhethi amabili, amaphrosesa axhunywe ngezixhumanisi ze-xGMI zesokhethi-to-socket, ingxenye ye-Infinity Fabric exhumanisa zonke izingxenye ze-SoC ndawonye.
Imithwalo yemisebenzi engaziwa ye-NUMA ingase idinge umkhawulokudonsa omkhulu we-xGMI ngenxa yokuxhumana okubanzi kwama-cross-socket. I-NUMA-aware workloads ingase ifune ukunciphisa amandla e-xGMI ngoba ayinayo i-cross-socket traffic futhi ikhetha ukusebenzisa i-CPU ekhuphukayo. Ububanzi bomzila we-xGMI bungancishiswa ukusuka ku-x16 kuye ku-x8 noma ku-x2, noma isixhumanisi se-xGMI singakhutshazwa uma ukusetshenziswa kwamandla kuphezulu kakhulu.
Ukumiswa kwesixhumanisi se-xGMI kanye nesivinini esikhulu se-4-link xGMI (Cisco xGMI max Speed)
Ungasetha inombolo yezixhumanisi ze-xGMI nesivinini esiphezulu sesixhumanisi se-xGMI. Ukusetha leli nani kusivinini esiphansi kungalondoloza amandla angenasisekelo angasetshenziswa ukukhulisa imvamisa eyinhloko noma ukunciphisa amandla ewonke. Iphinde yehlise i-cross-socket bandwidth futhi ikhulise i-cross-socket latency. I-Cisco UCS C245 M8 Rack Server isekela izixhumanisi ezine ze-xGMI ezinesivinini esikhulu esingu-32 Gbps.
Izilungiselelo ze-Cisco xGMI max Speed zivumela ukumisa i-xGMI Link kanye ne-4-Link/3-Link xGMI Max Speed. Ukunika amandla isivinini esikhulu se-Cisco xGMI kuzosetha ukucushwa kwesixhumanisi se-xGMI kube ngu-4, kanye ne-4-Link xGMI Max Speed ngu-32 Gbps. Ukukhubaza izilungiselelo ze-Cisco xGMI Max Speed kuzosebenzisa amanani azenzakalelayo.
Ithebula lesi-8 lifingqa izilungiselelo.
Ithebula 8. Izilungiselelo zesixhumanisi se-xGMI
| Ukusetha | Izinketho |
| I-Cisco xGMI Max Speed | ● Kukhutshaziwe (okuzenzakalelayo)
● Inikwe amandla |
| I-xGMI Link Configuration | ● Okuzenzakalelayo
● 1 ● 2 ● 3 ● 4 |
| 4-Link xGMI Max Speed | ● Okuzenzakalelayo (25 Gbps)
● 20 Gbps ● 25 Gbps ● 32 Gbps |
| 3-Link xGMI Max Speed | ● Okuzenzakalelayo (25 Gbps)
● 20 Gbps ● 25 Gbps ● 32 Gbps |
Qaphela: Lesi sici se-BIOS sisebenza kuphela kuma-Cisco UCS X215c M8 Compute Nodes kanye ne-Cisco UCS C245 M8 Rack Server enokulungiselelwa okungu-2-socket.
Ukusebenza kwe-CPU okuthuthukisiwe
Le nketho ye-BIOS isiza abasebenzisi ukuguqula izilungiselelo ezithuthukisiwe zokusebenza kwe-CPU. Uma inikwe amandla, le nketho ilungisa izilungiselelo zephrosesa futhi inike amandla iphrosesa ukuthi isebenze ngokunamandla, okungathuthukisa ukusebenza kwe-CPU kukonke kodwa okungase kuphumele ekusetshenzisweni kwamandla okuphezulu. Amanani ale nketho ye-BIOS angaba Okuzenzakalelayo noma Akhutshaziwe. Ngokuzenzakalela, inketho yokusebenza kwe-CPU ethuthukisiwe ivaliwe.
Qaphela: Lesi sici se-BIOS sisebenza kuphela kuma-Cisco UCS X215c M8 Compute Nodes kanye ne-Cisco UCS C245 M8 Rack Servers. Uma le nketho inikwe amandla, sincoma kakhulu ukusetha inqubomgomo yabalandeli ngamandla amakhulu.
Ngokuzenzakalelayo, lesi silungiselelo se-BIOS Sivaliwe.
Izilungiselelo zememori
Ungamisa izilungiselelo zememori ezichazwe kulesi sigaba.
I-NUMA Nodes Per Socket (NPS)
This setting lets you specify the number of desired NUMA Nodes Per Socket (NPS) and enables a tradeoff between reducing local memory latency for NUMA-aware or highly parallelizable workloads and increasing per-core memory bandwidth for non-NUMA-friendly workloads. Socket interleave (NPS0) will attempt to interleave the two sockets together into one NUMA node. 4th Gen AMD EPYC processors support a varying number of NUMA NPS values depending on the internal NUMA topology of the processor. NPS2 and NPS4 may not be options on certain processors or with certain memory populations.
Kumaseva anesokhethi eyodwa, inani lamanodi e-NUMA isokhethi ngalinye lingaba ngu-1, 2, noma 4, nakuba kungewona wonke amanani asekelwa yiwo wonke amaphrosesa. Ukusebenza kwezinhlelo zokusebenza ezithuthukiswe kakhulu nge-NUMA kungathuthukiswa ngokusetha inani lamanodi e-NUMA isokhethi ngalinye libe inani elisekelwayo elikhulu kuno-1.
Ukucushwa okuzenzakalelayo (Isizinda esisodwa se-NUMA ngesokhethi) siyanconywa emisebenzini eminingi. I-NPS4 inconyelwe i-High-Performance Computing (HPC) neminye imithwalo yomsebenzi ehambisana kakhulu. Uma usebenzisa ama-adaptha enethiwekhi angu-200-Gbps, i-NPS2 ingancanyelwa ukunikeza ukuvumelana phakathi kokubambezeleka kwememori nomkhawulokudonsa wememori we-Network Interface Card (NIC).
Lesi silungiselelo sizimele kunqolobane Yokucushwa Okuthuthukile kanye Ne-Power Interface (ACPI) Static Resource Affinity Table (SRAT) Layer- 3 (L3) njengesilungiselelo sesizinda se-NUMA. Uma i-ACPI SRAT L3 Inqolobane njengesizinda se-NUMMA inikwe amandla, lesi silungiselelo sibe sesinquma ubumbudumbudu obuhlukanisa inkumbulo. Nge-NPS1, zonke iziteshi zememori eziyisishiyagalombili ziphambanisiwe. Nge-NPS2, zonke iziteshi ezine ziphambana. Nge-NPS4, zonke iziteshi ezipheya ziphambanisiwe. Ithebula 9 lifingqa izilungiselelo.
Ithebula 9. Izilungiselelo ze-NUMA NPS
| Ukusetha | Izinketho |
| NUM Amanodi ngesokhethi ngayinye | ● Okuzenzakalelayo (NPS1)
● I-NPS0: inkumbulo yekhefu ifinyelela kuzo zonke iziteshi kuwo womabili amasokhethi (akunconyiwe). ● I-NPS1: inkumbulo ye-interleave ifinyelela kuzo zonke iziteshi eziyisishiyagalombili kusokhethi ngasinye; ibika inodi eyodwa ye-NUMA isokhethi ngalinye (ngaphandle uma Inqolobane ye-L3 njengoba i-NUMA ivuliwe). ● I-NPS2: inkumbulo ye-interleave ifinyelela kuwo wonke amaqembu eziteshi ezine (ABCD ne-EFGH) kusokhethi ngayinye; ibika amanodi amabili e-NUMA isokhethi ngalinye (ngaphandle kwalapho Inqolobane ye-L3 njengoba i-NUMA ivuliwe). ● I-NPS4: inkumbulo yekhefu ifinyelela kuwo wonke amapheya eziteshi (AB, CD, EF, ne-GH) kusokhethi ngayinye; ibika ama-node amane e-NUMA isokhethi ngalinye (ngaphandle uma Inqolobane ye-L3 njengoba i-NUMA ivuliwe). |
I-I/O Memory Management Unit (IOMMU)
I-I/O Memory Management Unit (IOMMU) ihlinzeka ngezinzuzo ezimbalwa futhi iyadingeka lapho kusetshenziswa isilawuli esiphazamisayo esihlelekayo se-x2 (x2APIC). Ukunika amandla i-IOMMU kuvumela amadivayisi (afana nesilawuli se-SATA esihlanganisiwe se-EPYC) ukwethula izicelo zokuphazamiseka ezihlukene (ama-IRQ) kudivayisi ngayinye enamathiselwe esikhundleni se-IRQ eyodwa yesistimu engaphansi. I-IOMMU iphinde ivumele amasistimu okusebenza ukuthi anikeze ukuvikeleka okwengeziwe kwamadivayisi anamandla e-I/O e-Direct Memory Access (DMA)–. I-IOMMU iphinda isize ngokuhlunga nokuphinda kumephu iziphazamiso ezivela kumadivayisi axhumene ne-peripheral. Ithebula le-10 lifingqa izilungiselelo.
Ithebula 10. Izilungiselelo ze-IOMMU
| Ukusetha | Izinketho |
| IOMMU | ● Okuzenzakalelayo (kunikwe amandla)
● Kukhutshaziwe: khubaza usekelo lwe-IOMMU ● Inikwe amandla: nika amandla usekelo lwe-IOMMU |
Inkumbulo iyahlangana
I-Memory interleaving kuyindlela esetshenziswa ama-CPU ukukhulisa umkhawulokudonsa wememori otholakala kuhlelo lokusebenza. Ngaphandle kokushiyana phakathi, amabhlogo enkumbulo alandelanayo, ngokuvamile imigqa yenqolobane, afundwa ebhange lememori elifanayo. Isofthiwe efunda inkumbulo ilandelana ngakho-ke izodinga ukulinda umsebenzi wokudlulisa imemori ukuze uqedele ngaphambi kokuqala ukufinyelela kwememori okulandelayo. Ngokuvunyelwa kokushiya inkumbulo, amabhulokhi enkumbulo alandelanayo asemabhange ahlukene, ngakho-ke wonke angaba negalelo kumkhawulokudonsa wenkumbulo uwonke ongafinyelelwa wuhlelo.
I-AMD incoma ukuthi zonke iziteshi eziyisishiyagalombili zememori isokhethi le-CPU ngalinye zigcwaliswe ngawo wonke amashaneli anomthamo olinganayo. Le ndlela yenza i-subsystem yememori isebenze kumodi ye-interleaving yezindlela eziyisishiyagalombili, okufanele inikeze ukusebenza okungcono kakhulu ezimweni eziningi. Ithebula 11 lifingqa izilungiselelo.
Ithebula 11. Izilungiselelo zokuhlukanisa inkumbulo
| Ukusetha | Izinketho |
| Inkumbulo iyahlangana | ● Inikwe amandla: i-interleaving inikwe amandla ngememori esekelwe ukulungiselelwa kwe-DIMM.
● Khubaza: akukho ukuxhumana okwenziwayo. |
Izilungiselelo zamandla
Ungamisa izilungiselelo zesimo samandla ezichazwe kulesi sigaba.
Ukuthuthukisa ukusebenza okubalulekile
Isici esibalulekile sokuthuthukisa ukusebenza sivumela iphrosesa ukuthi ishintshele kumvamisa ephezulu kunemvamisa yesisekelo se-CPU, ngokusekelwe ekutholakaleni kwamandla, i-thermal headroom, kanye nenani lama-cores asebenzayo ohlelweni. Ukuthuthukiswa kokusebenza okubalulekile kungabangela i-jitter ngenxa yokuguqulwa kwemvamisa yama-processor cores.
Eminye imithwalo yemisebenzi ayidingi ukuthi ikwazi ukusebenza ngesivinini esiphezulu ukuze kuzuzwe amazinga amukelekayo okusebenza. Ukuze uthole ukusebenza kahle kwamandla okungcono, ungasetha imvamisa ye-boost ephezulu kakhulu. Lesi silungiselelo asikuvumeli ukuthi usethe imvamisa engashintshi; ikhawulela kuphela imvamisa ye-boost ephezulu. Ukusebenza kwangempela kokuthuthukisa kuncike ezintweni eziningi nezinye izilungiselelo ezishiwo kulo mbhalo. Ithebula 12 lifingqa izilungiselelo.
Ithebula 12. Izilungiselelo zokuthuthukisa ukusebenza kahle
| Ukusetha | Izinketho |
| Ukuthuthukisa ukusebenza okubalulekile | ● Okuzenzakalelayo (kunikwe amandla): kuvumela iphrosesa ukuthi ishintshele kumafrikhwensi aphezulu (i-turbo frequency) kune
imvamisa yesisekelo se-CPU ● Kukhutshaziwe: kukhubaza imvamisa ye-CPU core boost |
Ukulawulwa kwe-C-state yomhlaba wonke
I-C-states yizimo zamandla ezingasebenzi eziyinhloko ze-CPU. I-C0 yisimo sokusebenza lapho iziyalezo zicutshungulwa khona, futhi izifunda ze-C ezinenombolo ephezulu (C1, C2, njll.) ziyizifunda zamandla aphansi lapho umnyombo ungenzi lutho. Ukulungiselelwa kwe-Global C-state kungasetshenziswa ukunika amandla nokukhubaza i-C-states kuseva. Ngokuzenzakalelayo, isilawuli se-C-state yomhlaba wonke sisethelwe kokuthi Okuzenzakalelayo, okwenza ama-cores akwazi ukungena ezimeni zamandla aphansi; lokhu kungabangela i-jitter ngenxa yokuguqulwa kwemvamisa yama-processor cores. Uma lesi silungiselelo sikhutshaziwe, ama-CPU cores azosebenza ezifundeni ze-C0 ne-C1. Ithebula 13 lifingqa izilungiselelo.
I-C-states ivezwa ngezinto ze-ACPI futhi ingacelwa ngamandla yisoftware. I-Software ingacela ushintsho lwe-C-state ngokwenza umyalo we-HALT noma ngokufunda ekhelini elithile le-I/O. Izenzo ezithathwe iphrosesa lapho ingena ku-C-state yamandla aphansi nazo zingalungiselelwa isofthiwe. Umongo wephrosesa we-4th Gen AMD EPYC yakhelwe ukusekela izifunda ezintathu ezishiwo yi-AMD:
I/O-based C0, C1, kanye ne-C2.
Ithebula 13. Izilungiselelo zomhlaba wonke ze-C-state
| Ukusetha | Izinketho |
| Ukulawulwa kwe-C-state yomhlaba wonke | ● Okuzenzakalelayo (kunikwe amandla): kunika amandla i-I/O-based C-states
● Kukhutshaziwe: kukhubaza i-I/O-based C-states |
Izendlalelo-1 kanye ne-Layer-2 yokusakaza yezingxenyekazi zekhompuyutha ezandulela kuqala
Umsebenzi omningi uyazuza ekusetshenzisweni kwe-Layer-1 kanye ne-Layer-2 stream prefetchers yehadiwe (i-L1 Stream HW Prefetcher kanye ne-L2 Stream HW Prefetcher) ukuze kuqoqwe idatha futhi kugcinwe ipayipi eliwumgogodla limatasa. Kodwa-ke, eminye imithwalo yomsebenzi ayihleliwe ngokwemvelo futhi izozuza ukusebenza okungcono kakhulu uma kukhutshaziwe oyedwa noma bobabili abalandeli bokuqala. Ngokuzenzakalelayo, kokubili ama-prefetcher anikwe amandla. Ithebula 14 lifingqa izilungiselelo.
Ithebula 14. Izilungiselelo ze-Layer-1 kanye ne-Layer-2 zokusakaza kwehadiwe yokusakaza kuqala
| Ukusetha | Izinketho |
| I-L1 Sakaza i-HW Prefetcher | ● Okuzenzakalelayo (Kunikwe amandla)
● Khubaza: ikhubaza i-prefetcher ● Nika amandla: yenza i-prefetcher |
| I-L2 Sakaza i-HW Prefetcher | ● Okuzenzakalelayo (Kunikwe amandla)
● Khubaza: ikhubaza i-prefetcher ● Nika amandla: yenza i-prefetcher |
Isilayida se-Determinism
Isilayida se-Determinism sivumela ukukhetha phakathi kokusebenza okufanayo kuwo wonke amasistimu amiswe ngendlela efanayo esikhungweni sedatha, ngokusetha iseva kusilungiselelo Sokusebenza, noma ukusebenza okuphezulu kwanoma iyiphi isistimu ngayinye kodwa ngokusebenza okuhlukahlukene kuyo yonke isikhungo sedatha, ngokusetha iseva kuzilungiselelo zamandla. Uma isilayidi se-Determinism sisethelwe kokuthi Ukusebenza, qiniseka ukuthi Amandla Okuklama Okushisayo (cTDP) okulungisekayo kanye Nomkhawulo Wamandla Wephakheji (PPL) asethelwe kunani elifanayo. Ukusetha okuzenzakalelayo (okuzenzakalelayo) kwamaphrosesa amaningi kuyimodi ye-Performance determinism, evumela iphrosesa ukuthi isebenze ezingeni eliphansi lamandla ngokusebenza okungaguquki. Ukuze uthole ukusebenza okuphezulu, setha isilayidi se-Determinism sibe Amandla. Ithebula 15 lifingqa izilungiselelo.
Ithebula 15. Izilungiselelo zesilayidi se-Determinism
| Ukusetha | Izinketho |
| Isilayida se-Determinism | ● Okuzenzakalelayo: lesi silungiselelo silingana nenketho Yokusebenza.
● Amandla: aqinisekisa amazinga aphezulu okusebenza e-CPU ngayinye kunani elikhulu lama-CPU amiswe ngendlela efanayo ngokucindezela ama-CPU kuphela lapho efinyelela ku-cTDP efanayo. ● Ukusebenza: kuqinisekisa amazinga okusebenza angaguquguquki kwinani elikhulu labantu bama-CPU amiswe ngendlela efanayo ngokucindezela amanye ama-CPU ukuze asebenze ngezinga eliphansi lamandla. |
I-CPPC: Ukulawula Ukusebenza Kwephrosesa Ehlanganyelwe
I-Collaborative Processor Performance Control (CPPC) yethulwe nge-ACPI 5.0 njengendlela yokuxhumana nokusebenza phakathi kwesistimu yokusebenza ne-hardware. Le modi ingasetshenziselwa ukuvumela i-OS ukuthi ilawule ukuthi i-turbo boost ingasetshenziswa nini futhi ingakanani emzamweni wokulondoloza ukusebenza kahle kwamandla. Akuwona wonke amasistimu wokusebenza asekela i-CPPC, kodwa iMicrosoft yaqala ukusekelwa nge-Microsoft Windows 2016 nakamuva.
Ithebula lesi-16 lifingqa izilungiselelo.
Ithebula 16. Izilungiselelo ze-CPPC
| Ukusetha | Izinketho |
| I-CPPC | ● Okuzenzakalelayo
● Kukhutshaziwe: kukhutshaziwe ● Inikwe amandla: ivumela i-OS ukuthi yenze izicelo zokusebenza namandla okuthuthukisa isebenzisa i-ACPI CPPC |
amandla profile ukukhetha F19h
Ukukhethwa kwe-DF P-state ku-profile Inqubomgomo ibhalwe ngaphezulu ububanzi be-P-state, inketho ye-BIOS, noma inketho ye-APB_DIS BIOS, lapho u-F ebhekisela emndenini womprosesa futhi uM esho imodeli.
| Izilungiselelo | Izinketho |
| Amandla profile ukukhetha F19h | ● Imodi yokusebenza kahle
● Imodi yokusebenza okuphezulu ● Imodi yokusebenza ye-I/O ephezulu ● Imodi yokusebenza kwenkumbulo elinganiselayo ● Imodi yokusebenza ewumongo ebhalansile ● Imodi yokusebenza kwenkumbulo ewumongo olinganiselayo ● Okuzenzakalelayo |
Inqubomgomo yokulawula abalandeli
Inqubomgomo yabalandeli ikuvumela ukuthi ulawule isivinini sabalandeli ukuze unciphise ukusetshenziswa kwamandla eseva namazinga omsindo. Ngaphambi kokusetshenziswa kwenqubomgomo yabalandeli, isivinini sabalandeli sikhuphuka ngokuzenzakalelayo lapho izinga lokushisa lanoma iyiphi ingxenye yeseva lidlula umkhawulo omisiwe. Ukusiza ukuqinisekisa ukuthi izivinini zabalandeli ziphansi, amazinga okushisa e-threshold wezingxenye ngokuvamile ayesethwa kumanani aphezulu. Nakuba lokhu kuziphatha kufaneleka kakhulu ukulungiselelwa kweseva, akuzange kubhekane nezimo ezilandelayo:
- Ukusebenza okuphezulu kwe-CPU: Ngokusebenza okuphezulu, ama-CPU athile kufanele apholiswe kakhulu ngaphansi kwezinga lokushisa elibekiwe. Lokhu kupholisa kudinga isivinini esiphezulu kakhulu sabalandeli, okuholela ekwenyukeni kokusetshenziswa kwamandla namazinga omsindo.
- Ukusetshenziswa kwamandla okuphansi: Ukusiza ukuqinisekisa ukusetshenziswa kwamandla okuphansi kakhulu, abalandeli kufanele bagijime kancane kakhulu futhi, kwezinye izimo, bame ngokuphelele kumaseva avumela lokhu kuziphatha. Kodwa isivinini sabalandeli esihamba kancane singabangela amaseva ukuthi ashise kakhulu. Ukuze ugweme lesi simo, udinga ukugijima abalandeli ngesivinini esishesha ngokumaphakathi kunejubane eliphansi kakhulu.
Ungakhetha izinqubomgomo zabalandeli ezilandelayo:
- Kulinganisiwe: Lena inqubomgomo ezenzakalelayo. Lesi silungiselelo singapholisa cishe noma yikuphi ukucushwa kweseva, kodwa singase singafaneli amaseva anamakhadi e-PCIe, ngoba lawa makhadi ashisa kakhulu.
- Amandla Aphansi: Lesi silungiselelo sifaneleka kahle kumaseva wokulungiselela okuncane angaqukethe amakhadi e-PCIe.
- Amandla Aphezulu: Lesi silungiselelo singasetshenziselwa ukulungiselelwa kweseva okudinga isivinini sabalandeli ukusuka kumaphesenti angu-60 ukuya kwangu-85. Le nqubomgomo ifaneleka kahle kumaseva aqukethe amakhadi e-PCIe ashisisa kalula futhi anamazinga okushisa aphezulu. Ubuncane besivinini sabalandeli esisethwe ngale nqubomgomo siyahlukahluka kunkundla yeseva ngayinye, kodwa cishe sisebangeni lamaphesenti angu-60 ukuya kwangu-85.
- Amandla Aphezulu: Lesi silungiselelo singasetshenziselwa ukulungiselelwa kweseva okudinga isivinini sabalandeli esiphezulu kakhulu esisukela phakathi kwamaphesenti angu-70 no-100. Le nqubomgomo ifaneleka kahle kumaseva aqukethe amakhadi e-PCIe ashisisa kalula futhi anamazinga okushisa aphezulu kakhulu. Isivinini esincane sabalandeli esisethwe ngale nqubomgomo siyahlukahluka kunkundla yeseva ngayinye, kodwa sicishe sisebangeni elingamaphesenti angama-70 ukuya kwangu-100.
- I-Acoustic: Isivinini sabalandeli sincishisiwe ukuze kuncishiswe amazinga omsindo ezindaweni ezizwela i-acoustic. Kunokuba ulawule ukusetshenziswa kwamandla kanye nokuvimbela ukunyakaziswa kwengxenye njengakwezinye izindlela, inketho ye-Acoustic ingase ibangele ukushayeka kwesikhashana ukuze kuzuzwe ileveli yomsindo eyehlisiwe. Ukusebenzisa le nqubomgomo yokulawula abalandeli kungase kubangele imithelela yokusebenza kwesikhashana kwesikhashana.
Qaphela: Le nqubomgomo ilungiseka kumaseva ezimele we-Cisco UCS C-Series M8 kusetshenziswa ikhonsoli ye-Cisco Integrated Management Controller (IMC) kanye nomphathi we-Cisco IMC. Kusuka ku-Cisco IMC web ikhonsoli, khetha Ukubala > Izinqubomgomo Zamandla > Inqubomgomo Yabalandeli Elungiselelwe > Inqubomgomo Yabalandeli.
Kumaseva e-Cisco Intersight®–aphethwe i-C-Series M8, le nqubomgomo iyalungiseka kusetshenziswa izinqubomgomo zabalandeli.
Izilungiselelo ze-BIOS ze-Cisco UCS X215c M8 Compute Nodes, Cisco UCS C245 M8 Rack Servers, kanye ne-Cisco UCS C225 M8 Rack Servers
Ithebula 17 libala amagama amathokheni we-BIOS, okuzenzakalelayo, namanani asekelwe amaseva e-Cisco UCS M8 anemindeni ye-AMD EPYC 4th gen kanye ne-5th Gen processor.
Ithebula 17. Amagama wethokheni ye-BIOS namanani
| Igama lethokheni ye-BIOS | Inani elizenzakalelayo | Amanani asekelwe |
| Iphrosesa | ||
| Imodi ye-CPU SMT | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| Imodi ye-SVM | Inikwe amandla | Inikwe amandla, Ikhutshaziwe |
| I-DF C-states | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| Inqolobane ye-ACPI SRAT L3 njenge-NUMA
Isizinda |
Okuzenzakalelayo (kukhutshaziwe) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| I-APBDIS | Okuzenzakalelayo (0) | Okuzenzakalelayo, 0, 1 |
| I-SOC P-State elungisiwe SP5F 19h | P0 | P0, P1, P2 |
| 4-link xGMI isivinini esikhulu* | Okuzenzakalelayo (32Gbps) | Okuzenzakalelayo, 20Gbps, 25Gbps, 32Gbps |
| Ukusebenza kwe-CPU okuthuthukisiwe* | Ikhutshaziwe | Okuzenzakalelayo, Kukhubazekile |
| Inkumbulo | ||
| NUM amanodi isokhethi ngalinye | Okuzenzakalelayo (NPS1) | Okuzenzakalelayo, NPS0, NPS1, NPS2, NPS4 |
| IOMMU | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| Inkumbulo iyahlangana | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| Amandla/ukusebenza | ||
| Ukuthuthukisa ukusebenza okubalulekile | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kukhubazekile |
| Ukulawulwa kwe-C-state yomhlaba wonke | Ikhutshaziwe | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| I-L1 Sakaza i-HW Prefetcher | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| I-L2 Sakaza i-HW Prefetcher | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo, Kunikwe amandla, Kukhutshaziwe |
| Isilayida se-Determinism | Okuzenzakalelayo (amandla) | Okuzenzakalelayo, Amandla, Ukusebenza |
| I-CPPC | Okuzenzakalelayo (kukhutshaziwe) | Okuzenzakalelayo, Kukhutshaziwe, Kunikwe amandla |
| Igama lethokheni ye-BIOS | Inani elizenzakalelayo | Amanani asekelwe |
| Amandla profile ukukhetha F19h | Imodi yokusebenza okuphezulu | Imodi yokusebenza kwenkumbulo ebhalansile, imodi yokusebenza kahle, imodi yokusebenza okuphezulu, imodi yokusebenza ye-I/O ephezulu, imodi yokusebenza ewumgogodla ebhalansisiwe, imodi yokusebenza kwenkumbulo ewumgogodla ebhalansile |
Izincomo ze-BIOS zemisebenzi ehlukahlukene yenhloso evamile
Lesi sigaba sifingqa izilungiselelo ze-BIOS ezinconyiwe ukuze kukhuliswe umthwalo womsebenzi wezinjongo ezijwayelekile:
- I-Computing-intensive
- I/O-intensive
- Ukusebenza kahle kwamandla
- Ukubambezeleka okuphansi
Izigaba ezilandelayo zichaza umsebenzi ngamunye.
Umsebenzi onzima we-CPU
Ngomthwalo onzima we-CPU, umgomo uwukusabalalisa umsebenzi womsebenzi owodwa kuma-CPU amaningi ukuze kuncishiswe isikhathi sokucubungula ngangokunokwenzeka. Ukuze wenze lokhu, udinga ukusebenzisa izingxenye zomsebenzi ngokuhambisana. Inqubo ngayinye, noma uchungechunge, lusingatha ingxenye ethile yomsebenzi futhi lwenze izibalo kanyekanye. Ama-CPU ngokuvamile adinga ukushintshanisa ulwazi ngokushesha, okudinga ihadiwe yokuxhumana ekhethekile.
Umthwalo onzima we-CPU ngokuvamile uzuza kumaphrosesa noma inkumbulo efinyelela ubuningi befrikhwensi ye-turbo yanoma yimuphi umgogodla ngamunye nganoma yisiphi isikhathi. Izilungiselelo zokuphathwa kwamandla ephrosesa zingasetshenziswa ukuze kuqinisekiswe ukuthi noma yikuphi ukukhuphuka kwemvamisa yengxenye kungafinyelelwa kalula. Umthwalo onzima we-CPU uwumthwalo wokusebenza wenhloso evamile, ngakho ukulungiselelwa kwenziwa ngokujwayelekile ukuze kukhushulwe umgogodla wephrosesa nesivinini sememori, futhi kusetshenziswe ukuhlelwa kokusebenza okuvame ukuzuza esikhathini esisheshayo sekhompuyutha.
I/O-imisebenzi enzima
Ukulungiselelwa okujulile kwe-I/O kuwukucupha okuncike ekuphumeni okukhulu phakathi kwe-I/O nenkumbulo. Izici zokuphatha amandla ezisekelwe ekusetshenzisweni kweprosesa ezithinta ukusebenza kuzixhumanisi phakathi kwe-I/O nenkumbulo zivaliwe.
Imisebenzi eyonga amandla
Ukulungiselela okonga amandla yizilungiselelo ezivame kakhulu zokusebenza okunokulinganisela. Zihlomula imisebenzi eminingi yohlelo lokusebenza kuyilapho zinika amandla izilungiselelo zokuphatha amandla ezingenamthelela omncane ekusebenzeni kukonke. Izilungiselelo ezisetshenziselwa imisebenzi eyonga amandla zikhulisa ukusebenza kohlelo lokusebenza okuvamile kunokonga amandla. Izilungiselelo zokuphatha amandla eprosesa zingathinta ukusebenza uma kusetshenziswa amasistimu wokusebenza we-virtualization. Ngakho-ke, lezi zilungiselelo zinconyelwe amakhasimende angashuni i-BIOS ngomthwalo wawo wokusebenza.
Imithwalo yemisebenzi yokubambezeleka ephansi
Imithwalo yomsebenzi edinga ukubambezeleka okuphansi, njengokuhweba ngezimali nokucubungula ngesikhathi sangempela, idinga amaseva ukuthi anikeze impendulo yesistimu engaguquki. Imithwalo yemisebenzi yokubambezeleka ephansi ingeyamakhasimende afuna inani elincane lokubambezeleka kokubambezeleka kwemithwalo yawo yomsebenzi. Isivinini esikhulu kanye nokuphumayo kuvame ukunikelwa ukuze kwehliswe ukubambezeleka okuphelele kwekhompyutha. Ukuphathwa kwamandla ephrosesa nezinye izici zokuphatha ezingase zethule ukubambezeleka kwekhompyutha kukhutshaziwe.
Ukuze uthole ukubambezeleka okuphansi, udinga ukuqonda ukucushwa kwehadiwe yesistimu ngaphansi kokuhlolwa. Izinto ezibalulekile ezithinta izikhathi zokuphendula zihlanganisa inani lama-cores, izintambo zokucubungula ngomongo ngamunye, inani lamanodi e-NUMA, i-CPU nokuhlelwa kwememori ku-topology ye-NUMA, kanye ne-cache topology endaweni ye-NUMA. Izinketho ze-BIOS zivame ukuzimela ku-OS, futhi uhlelo lokusebenza oluvunywe kahle oluphansi lokulinda luyadingeka ukuze kuzuzwe ukusebenza okunqunyiwe.
Isifinyezo sezilungiselelo ze-BIOS ezithuthukiselwe umthwalo wenjongo evamile
Ithebula 18 lifingqa izilungiselelo ze-BIOS ezilungiselelwe imithwalo yemisebenzi yenhloso evamile.
Ithebula 18. Izincomo ze-BIOS ze-CPU-intensive, I/O-intensive, amandla okusebenza kahle, kanye ne-low-latency workloads
| Izinketho ze-BIOS | Amanani e-BIOS (okuzenzakalelayo kwenkundla) | I-CPU enamandla | I/O inamandla | Amandla ukusebenza kahle | Ukubambezeleka okuphansi |
| Iphrosesa | |||||
| Imodi ye-CPU SMT | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe |
| Imodi ye-SVM | Inikwe amandla | Inikwe amandla | Inikwe amandla | Inikwe amandla | Ikhutshaziwe |
| I-DF C-states | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo | Ikhutshaziwe | Okuzenzakalelayo | Ikhutshaziwe |
| I-ACPI SRAT L3
Inqolobane njengesizinda se-NUMA |
Okuzenzakalelayo (kukhutshaziwe) | Inikwe amandla | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| I-APBDIS | Okuzenzakalelayo (0) | 1 | 1 | Okuzenzakalelayo | Okuzenzakalelayo |
| I-SOC P-State elungisiwe SP5F 19h | P0 | P0 | P0 | P2 | P0 |
| 4-link xGMI isivinini esikhulu | Okuzenzakalelayo (32Gbps) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Ukusebenza kwe-CPU okuthuthukisiwe | Ikhutshaziwe | Okuzenzakalelayo | Ikhutshaziwe | Ikhutshaziwe | Ikhutshaziwe |
| Inkumbulo | |||||
| Amanodi angu-NUMA isokhethi ngalinye | Okuzenzakalelayo (NPS1) | I-NPS4 | I-NPS4 | Okuzenzakalelayo | Okuzenzakalelayo |
| IOMMU | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo * | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe* |
| Inkumbulo ezingenelayo | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo * | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe* |
| Izinketho ze-BIOS | Amanani e-BIOS (okuzenzakalelayo kwenkundla) | I-CPU enamandla | I/O inamandla | Amandla ukusebenza kahle | Ukubambezeleka okuphansi |
| Amandla/ukusebenza | |||||
| Ukusebenza okubalulekile thuthukisa | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe |
| Global C-State control | Ikhutshaziwe | Ikhutshaziwe | Inikwe amandla | Inikwe amandla | Ikhutshaziwe |
| Ukusakaza kwe-L1 HW I-Prefetcher | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe | Okuzenzakalelayo |
| Ukusakaza kwe-L2 HW I-Prefetcher | Okuzenzakalelayo (kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Ikhutshaziwe | Okuzenzakalelayo |
| Isilayida se-Determinism | Okuzenzakalelayo (amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Ukusebenza |
| I-CPPC | Okuzenzakalelayo (kukhutshaziwe) | Okuzenzakalelayo | Okuzenzakalelayo | Inikwe amandla | Okuzenzakalelayo |
| amandla profile ukukhetha F19h | Imodi yokusebenza okuphezulu | Imodi yokusebenza okuphezulu | Imodi yokusebenza ephezulu ye-I/O | Imodi yokusebenza kahle | Imodi yokusebenza okuphezulu |
Qaphela: Amathokheni e-BIOS ane-* agqanyisiwe asebenza kuphela kuma-Cisco UCS X215c M8 Compute Nodes kanye ne-Cisco UCS C245 M8 Rack Servers.
Uma isimo sohlelo lwakho lokusebenza singadingi ukwenziwa kwezinto ezibonakalayo, vala i-AMD ubuchwepheshe bokwenza izinto. Uma i-virtualization ikhutshaziwe, vala futhi inketho ye-AMD IOMMU. Kungabangela umehluko ekubambezelekeni kokufinyelela kwimemori. Bona i- Umhlahlandlela wokushuna ukusebenza kwe-AMD ukuze uthole ulwazi olwengeziwe.
Izincomo ezengeziwe ze-BIOS zomthwalo womsebenzi webhizinisi
Lesi sigaba sifingqa izilungiselelo ezilungile ze-BIOS zomthwalo webhizinisi:
- I-Virtualization
- Izitsha
- Isizindalwazi Sobudlelwano (RDBMS)
- Isizindalwazi Sokuhlaziya (Bigdata)
- Imisebenzi ye-HPC
Izigaba ezilandelayo zichaza umsebenzi webhizinisi ngalinye.
Imisebenzi ye-Virtualization
I-AMD Virtualization Technology ihlinzeka ngokulawuleka, ukuphepha, nokuvumelana nezimo ezindaweni ze-IT ezisebenzisa izixazululo ze-virtualization ezisekelwe kusofthiwe. Ngalobu buchwepheshe, iseva eyodwa ingahlukaniswa futhi ingavezwa njengamaseva amaningana azimele, okuvumela iseva ukuthi isebenzise izinhlelo zokusebenza ezihlukene kusistimu yokusebenza ngesikhathi esisodwa. Kubalulekile ukunika amandla i-AMD Virtualization Technology ku-BIOS ukuze isekele imithwalo yemisebenzi ye-virtualization.
Ama-CPU asekela ukwenziwa kwezinto ezibonakalayo anika amandla iphrosesa ukuthi isebenzise amasistimu wokusebenza amaningi emishinini ebonakalayo. Lesi sici sibandakanya okungaphezulu ngenxa yokuthi ukusebenza kwesistimu yokusebenza ihamba kancane ngokuqhathaniswa naleyo ye-OS yomdabu.
Ukuze uthole ulwazi olwengeziwe, bheka ama-AMD I-VMware vSphere Tuning Guide.
Umsebenzi wesitsha
Ukuhlanganisa inkundla yohlelo lokusebenza kanye nokuncika kwayo okuhlobene kukhipha ingqalasizinda eyisisekelo kanye nomehluko we-OS ukuze usebenze kahle. Isiqukathi ngasinye sigoqwa sibe yiphakheji elilodwa eliqukethe yonke indawo yesikhathi sokusebenza, okuhlanganisa uhlelo lokusebenza olunakho konke ukuncika kwalo, amalabhulali nokunye okumbambambili, kanye nokucushwa. files ezidingekayo ukuze usebenzise lolo hlelo lokusebenza. Iziqukathi ezisebenzisa izinhlelo zokusebenza endaweni yokukhiqiza zidinga ukuphathwa ukuze kuqinisekiswe isikhathi sokusebenza esingaguquki. Uma isiqukathi sehla, esinye isitsha kumele siqale ngokuzenzakalelayo.
Imithwalo yomsebenzi ekala futhi isebenza kahle ensimbi engenalutho kufanele ibone ijika lokukala elifanayo endaweni yesiqukathi enomsebenzi omncane ongaphezulu. Eminye imithwalo yemisebenzi efakwe esitsheni ingakwazi ngisho nokubona umehluko wokusebenza osondele ku-0% uma kuqhathaniswa nensimbi engenalutho. I-overhead enkulu ngokuvamile isho ukuthi izilungiselelo zohlelo lokusebenza kanye/noma ukulungiselelwa kwesiqukathi akusethiwe kahle. Lezi zihloko zingaphezu kobubanzi balo mhlahlandlela wokushuna. Kodwa-ke, ukuziphatha kokulinganisa kokulayisha kwe-CPU kwe-Kubernetes noma abanye abahleli benkundla ye-orchestration yesiqukathi bangase banikeze noma balayishe ibhalansi izinhlelo zokusebenza ezifakwe esiqukathi ngendlela ehlukile kunendawo yensimbi engenalutho.
Ukuze uthole ulwazi olwengeziwe, bheka ama-AMD I-Kubernetes Container Tuning Guide.
Imithwalo yemisebenzi Yesizindalwazi Esihlobene
Ukuhlanganisa i-RDBMS njenge-Oracle, i-MySQL, i-PostgreSQL, noma i-Microsoft SQL Server enamaphrosesa e-AMD EPYC kungaholela ekusebenzeni okuthuthukisiwe kwedathabhesi, ikakhulukazi ezindaweni ezidinga ukuvumelana okuphezulu, ukucutshungulwa kwemibuzo ngokushesha, kanye nokusetshenziswa kahle kwensiza. Ukwakheka kwamaphrosesa e-AMD EPYC kuvumela imininingwane yolwazi ukuthi isebenzise ama-cores amaningi nemicu ngempumelelo, okuzuzisa kakhulu umthwalo wokuthengiselana, izibalo, kanye nokucubungula idatha ngezinga elikhulu.
Kafushane, ukusebenzisa ama-AMD EPYC processors ezindaweni ze-RDBMS kungaholela ekuthuthukisweni okuphawulekayo ekusebenzeni, ukulinganisa, kanye nokusebenza kahle kwezindleko, kukwenze kube ukukhetha okuqinile kwezixazululo zesizindalwazi sebhizinisi.
Amaphrosesa we-4th Gen AMD EPYC aletha Ukusebenza okuphezulu Kokufaka/Okukhiphayo Ngomzuzwana (i-IOPS) kanye nokuphumayo kuzo zonke izingosi zolwazi. Ukukhetha i-CPU efanele kubalulekile ukuze kugcinwe kungobo yomlando ukusebenza kwesicelo sesizindalwazi esifanele.
Ukuze uthole ulwazi olwengeziwe, bheka ama-AMD I-RDBMS Tuning Guide.
Izibalo zokusebenza ezinkulu zedatha
I-Big Data Analytics ibandakanya ukuhlolwa kwenani elikhulu ledatha ukuze kwembulwe amaphethini afihliwe, ukuhlobana, neminye imininingwane engasetshenziswa ukwenza izinqumo ezingcono. Lokhu kudinga amandla amakhulu okuhlanganisa, umthamo wenkumbulo, kanye nomkhawulokudonsa we-I/O—izindawo lapho amaphrosesa e-AMD EPYC ephumelela khona.
Amaphrosesa e-AMD EPYC ahlinzeka ngenkundla eqinile ye-Big Data Analytics, enikeza amandla okuhlanganisa, umthamo wenkumbulo, kanye nomkhawulokudonsa we-I/O odingekayo ukuze kusingathwe izidingo zokucubungula idatha ngezinga elikhulu. Ukukhula kwazo, ukusebenza kahle kwezindleko, nokusebenza kahle kwamandla kuzenza zibe yisinqumo esiphoqayo sezinhlangano ezifuna ukwakha noma ukuthuthukisa ingqalasizinda yazo ye-Big Data Analytics.
I-HPC (High-performance computing) imithwalo yomsebenzi
I-HPC isho ukusebenzisa ikhompuyutha okusekelwe kuqoqo esebenzisa amanodi amaningi angawodwana axhunyiwe futhi asebenza ngokufana ukuze kuncishiswe inani lesikhathi esidingekayo ukuze kucutshungulwe amasethi amakhulu edatha obekungathatha isikhathi eside kakhulu ukusebenza kunoma iyiphi isistimu eyodwa. Imithwalo yemisebenzi ye-HPC i-computing-intensive futhi ngokuvamile i-network-I/O ejulile. Imithwalo yemisebenzi ye-HPC idinga ikhwalithi ephezulu
Izingxenye ze-CPU nezindwangu zenethiwekhi ezinesivinini esikhulu, ezingabambeki kancane zoxhumo lwazo lwe-Message Passing Interface (MPI).
Amaqoqo e-Computing ahlanganisa i-head node ehlinzeka ngephoyinti elilodwa lokuphatha, ukuthumela, ukuqapha, nokuphatha iqoqo. Amaqoqo aphinde abe nengxenye yokuphatha umthwalo wangaphakathi, eyaziwa ngokuthi ishejuli, elawula zonke izinto ezingenayo zomsebenzi (okubhekiselwa kuzo njengemisebenzi). Ngokuvamile, imithwalo yomsebenzi ye-HPC idinga izinombolo ezinkulu zama-node anamanethiwekhi angavimbeli e-MPI ukuze akwazi ukukala. Ukukala kwamanodi kuyisici esisodwa esibaluleke kakhulu ekunqumeni ukusebenza okuzuziwe okusebenzisekayo kweqoqo.
I-HPC idinga inethiwekhi ye-I/O enomkhawulokudonsa ophezulu. Uma unika amandla ukusekela kwe-Direct Cache Access (DCA), amaphakethe enethiwekhi angena ngqo kusilondolozi se-Layer 3 processor esikhundleni sememori eyinhloko. Le ndlela inciphisa inani lemijikelezo ye-HPC I/O ekhiqizwe umthwalo wokusebenza we-HPC lapho kusetshenziswa ama-adaptha athile e-Ethernet, okubuye kwandise ukusebenza kwesistimu.
Ukuze uthole ulwazi olwengeziwe, bheka ama-AMD I-High-Performance Computing (HPC) Tuning Guide.
Isifinyezo sezilungiselelo ze-BIOS ezinconyelwe umthwalo wokusebenza webhizinisi
Ithebula 19 lifingqa amathokheni we-BIOS nezilungiselelo ezinconyelwe imithwalo yemisebenzi ehlukahlukene yebhizinisi.
Ithebula 19.
Izincomo ze-BIOS zokwenza izinto ezibonakalayo, iziqukathi, i-RDBMS, ukuhlaziya idatha enkulu, kanye nemithwalo yomsebenzi yebhizinisi le-HPC
| Izinketho ze-BIOS | Amanani e-BIOS (okuzenzakalelayo kwenkundla) | I-Virtualization / isitsha | I-RDBMS | Idatha enkulu izibalo | I-HPC |
| Iphrosesa | |||||
| Imodi ye-CPU SMT | Inikwe amandla | Inikwe amandla | Inikwe amandla | Ikhutshaziwe | Ikhutshaziwe |
| Imodi ye-SVM | Inikwe amandla | Inikwe amandla | Inikwe amandla | Inikwe amandla | Inikwe amandla |
| I-DF C-states | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Ikhutshaziwe | Okuzenzakalelayo | Okuzenzakalelayo |
| Inqolobane ye-ACPI SRAT L3
njengesizinda se-NUMA |
Okuzenzakalelayo (Kukhutshaziwe) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| I-APBDIS | Okuzenzakalelayo (0) | Okuzenzakalelayo | 1 | 1 | 1 |
| I-SOC P-State elungisiwe SP5F 19h | P0 | P0 | P0 | P0 | P0 |
| 4-link xGMI ubuningi isivinini* | Okuzenzakalelayo (32Gbps) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Ukusebenza kwe-CPU okuthuthukisiwe* | Ikhutshaziwe | Ikhutshaziwe | Ikhutshaziwe | Ikhutshaziwe | Okuzenzakalelayo |
| Izinketho ze-BIOS | Amanani e-BIOS (okuzenzakalelayo kwenkundla) | I-Virtualization / isitsha | I-RDBMS | Idatha enkulu izibalo | I-HPC |
| Inkumbulo | |||||
| Amanodi angu-NUMA isokhethi ngalinye | Okuzenzakalelayo (NPS1) | Okuzenzakalelayo | I-NPS4 | Okuzenzakalelayo | I-NPS4 |
| IOMMU | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Inkumbulo iyahlangana | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Amandla/ukusebenza | |||||
| Ukusebenza okubalulekile thuthukisa | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Global C-State control | Ikhutshaziwe | Inikwe amandla | Inikwe amandla | Inikwe amandla | Inikwe amandla |
| Ukusakaza kwe-L1 HW I-Prefetcher | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Ukusakaza kwe-L2 HW I-Prefetcher | Okuzenzakalelayo (Kunikwe amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| Isilayida se-Determinism | Okuzenzakalelayo (Amandla) | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo | Okuzenzakalelayo |
| I-CPPC | Okuzenzakalelayo (Kukhutshaziwe) | Inikwe amandla | Okuzenzakalelayo | Inikwe amandla | Okuzenzakalelayo |
| amandla profile ukukhetha F19h | Imodi yokusebenza okuphezulu | Imodi yokusebenza okuphezulu | Imodi yokusebenza ephezulu ye-I/O | Imodi yokusebenza okuphezulu | Imodi yokusebenza okuphezulu |
Qaphela: Amathokheni e-BIOS ane-*agqanyisiwe awasebenzi endaweni eyodwa kuphela elungiselelwe isokhethi njenge-Cisco UCS C225 M8 1U Rack Server.
- Uma umthwalo wakho wokusebenza unama-vCPU ambalwa ngomshini ngamunye (okungukuthi, ngaphansi kwekota yenani lama-cores isokhethi ngalinye), izilungiselelo ezilandelayo zivame ukunikeza ukusebenza okungcono kakhulu:
- I-NUMA NPS (amanodi isokhethi ngalinye) = 4
- I-LLC Njengoba i-NUMA ivuliwe
- Uma imishini yakho ebonakalayo yomthwalo womsebenzi inenani elikhulu lama-vCPU (okungukuthi, ngaphezu kwesigamu senani lama-cores isokhethi ngalinye), izilungiselelo ezilandelayo zivame ukunikeza ukusebenza okungcono kakhulu:
- I-NUMA NPS (amanodi isokhethi ngalinye) = 1
- I-LLC Njengoba i-NUMA ivaliwe
Ukuze uthole ukwaziswa okwengeziwe, bheka i- I-VMware vSphere Tuning Guide.
Umhlahlandlela wokushuna wesistimu yokusebenza ukuze usebenze okuphezulu
I-Microsoft Windows, i-VMware ESXi, i-Red Hat Enterprise Linux, kanye nezinhlelo zokusebenza ze-SUSE Linux ziza nezici eziningi zokuphatha amandla ezinikwe amandla ngokuzenzakalelayo. Ngakho-ke, kufanele ushune isistimu yokusebenza ukuze uzuze ukusebenza okuhle kakhulu.
Ukuthola eminye imibhalo yokusebenza, bheka i Imihlahlandlela yokusebenza ye-AMD EPYC.
I-Linux (Isigqoko Esibomvu ne-SUSE)
Umbusi we-CPUfreq uchaza izici zamandla zesistimu ye-CPU, okubuye kube nomthelela ekusebenzeni kwe-CPU. Umbusi ngamunye unokuziphatha kwakhe okuhlukile, inhloso, nokufaneleka ngokomthwalo womsebenzi.
Umbusi wokusebenza uphoqa i-CPU ukuthi isebenzise imvamisa yewashi ephezulu kakhulu. Le mvamisa isethwe ngokwezibalo futhi ayishintshi. Ngakho-ke, lo mbusi othile akanikezi inzuzo yokonga amandla. Ifanele amahora omsebenzi onzima kuphela, futhi nalapho, kuphela ngezikhathi lapho i-CPU ingavamile (noma ingalokothi) ingenzi lutho. Ukusetha okuzenzakalelayo “kungokufunwayo,” okuvumela i-CPU ukuthi ifinyelele ubuningi bemvamisa yewashi lapho umthwalo wesistimu uphezulu, kanye nefrikhwensi encane yewashi lapho isistimu ingenzi lutho. Nakuba lesi silungiselelo sivumela isistimu ukuthi ilungise ukusetshenziswa kwamandla ngokuvumelana nomthwalo wesistimu, ikwenza lokho ngezindleko zokubambezeleka kusuka ekushintsheni imvamisa.
Umbusi wokusebenza angasethwa kusetshenziswa umyalo we-cpupower: cpupower frequency-set -g ukusebenza
Ukuze uthole ulwazi olwengeziwe, bheka izixhumanisi ezilandelayo:
- I-Red Hat Enterprise Linux: Setha ukusebenza kwe-CPUfreq governor.
- Iseva ye-SUSE Enterprise Linux: Setha ukusebenza kwe-CPUfreq governor.
I-Microsoft Windows Server 2019 kanye ne-2022
Ku-Microsoft Windows Server 2019, ngokuzenzakalelayo, kusetshenziswa uhlelo lwamandla olubhalansile (olunconyiwe). Lesi silungiselelo sinika amandla ukongiwa kwamandla, kodwa singabangela ukubambezeleka okwengeziwe (isikhathi sokuphendula esinensayo kweminye imisebenzi), futhi singabangela izinkinga zokusebenza kwezinhlelo zokusebenza ezidinga i-CPU. Ukuze uthole ukusebenza okuphezulu, setha uhlelo lwamandla lube Ukusebenza Okuphezulu.
Ukuze uthole ulwazi olwengeziwe, bheka isixhumanisi esilandelayo:
I-Microsoft Windows ne-Hyper-V: Setha inqubomgomo yamandla ibe Ukusebenza Okuphezulu.
I-VMware ESXi
Ku-VMware ESXi, ukuphathwa kwamandla kosokhaya kuklanyelwe ukunciphisa ukusetshenziswa kwamandla kwabasingathi be-ESXi ngenkathi bevuliwe. Setha inqubomgomo yamandla kokuthi Ukusebenza Okuphezulu ukuze ufinyelele ukusebenza okuphezulu.
Ukuze uthole ulwazi olwengeziwe, bheka izixhumanisi ezilandelayo:
I-VMware ESXi: Setha inqubomgomo yamandla ibe Ukusebenza Okuphezulu.
Isiphetho
Lapho ulungisa izilungiselelo zesistimu ye-BIOS ukuze usebenze, udinga ukucabangela inani leprosesa nezinketho zememori. Uma ukusebenza okuhle kakhulu kuwumgomo wakho, qiniseka ukuthi ukhetha izinketho ezithuthukisa ukusebenza ngokuncamela ukonga amandla. Futhi zama nezinye izinketho, ezifana ne-memory interleaving kanye ne-CPU hyperthreading. Okubaluleke kakhulu, hlola umthelela wanoma yiziphi izilungiselelo ekusebenzeni okudingwa izinhlelo zakho zokusebenza.
Ukuze uthole olunye ulwazi
Ukuze uthole ulwazi olwengeziwe mayelana ne-Cisco UCS M8 Server ene-AMD 4th gen & 5th gen processors, bona izinsiza ezilandelayo:
- Umhlahlandlela wethokheni ye-IMM BIOS:
/b_IMM_Server_BIOS_Tokens_Guide.pdf
- I-Cisco UCS X215c M8 Compute Node:
- Iseva yeRakhi ye-Cisco UCS C245 M8:
- Iseva yeRakhi ye-Cisco UCS C225 M8:
- Imihlahlandlela yokushuna ye-AMD EPYC:
- https://developer.amd.com/resources/epyc-resources/epyc-tuning-guides/
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/58015- I-epyc-9004-tg-architecture-overview.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/white- amaphepha/58649_amd-epyc-tg-low-latency.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/57996- epyc-9004-tg-rdbms.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning- guides/58002_amd-epyc-9004-tg-hpc.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/58013- epyc-9004-tg-hadoop.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/58007- epyc-9004-tg-mssql-server.pdf
- https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning- guides/58001_amd-epyc-9004-tg-vdi.pdf
Indlunkulu yaseMelika
Inkampani Cisco Systems, Inc.
San Jose, CA
Indlunkulu yase-Asia Pacific
ICisco Systems (USA) Pte. Ltd.
Singapore
Europe Indlunkulu
I-Cisco Systems International BV Amsterdam,
Abase Netherlands
ICisco inamahhovisi angaphezu kuka-200 emhlabeni jikelele. Amakheli, izinombolo zocingo, nezinombolo zefeksi zibhalwe kuCisco Webindawo ku https://www.cisco.com/go/offices. I-Cisco kanye nelogo ye-Cisco yizimpawu zokuthengisa noma izimpawu zokuthengisa ezibhalisiwe ze-Cisco kanye/noma izinhlaka zayo e-US nakwamanye amazwe, view uhlu lwezimpawu zokuthengisa ze-Cisco, hamba kulokhu URL: https://www.cisco.com/go/trademarks. Izimpawu zokuthengisa zezinkampani zangaphandle ezishiwo ziyimpahla yabanikazi bazo. Ukusetshenziswa kwegama elithi uzakwethu akusho ubudlelwano bokusebenzisana phakathi kweCISCO nanoma iyiphi enye inkampani. (1 1 1 NOMA)
Iphrintiwe e-USA
Shayela-4692101-03
07/25
© 2025 Cisco kanye/noma amanxusa ayo. Wonke Amalungelo Agodliwe.
Amadokhumenti / Izinsiza
 |
Cisco Performance Tuning for Cisco UCS M8 Platforms [pdf] Incwadi Yeziqondiso I-C245 M8, Ukucushwa Kokusebenza Kwezinkundla ze-Cisco UCS M8, Ukushuna Izinkundla ze-Cisco UCS M8, Cisco UCS M8 Platforms, UCS M8 Platforms, M8 Platforms, Platforms |