Ukukhetha ikhadi lehluzo kulezi zinsuku akuseyona nje into yokuthi lithola ama-FPS amangaki kumdlalo wakho owuthandayo. Kubaluleke kakhulu. Kudingeka ugesi ongakanani ukuze isebenze kahleLokhu kubalulekile ngenxa yebhili kagesi kanye nokushisa okukhiqizwayo, kanye nomthelela wayo emvelweni. Futhi yilapho kungena khona ukuqhathanisa okuhle kokusetshenziswa kwamandla, kusukela kuma-GPU amancane e-Ryzen 5 7600 kuya ezikhungweni zamandla zokubala zobuhlakani bokwenziwa.
Emigqeni elandelayo uzothola umhlahlandlela ophelele lapho imihlaba emibili ihlangana khona: ngakolunye uhlangothi, ama-GPU anamandla aphansi, asebenza kahle kakhulu Ngakolunye uhlangothi, kukhona ama-GPU emidlalo nama-PC asekhaya, kanti ngakolunye uhlangothi, izilo zamaseva nezikhungo zedatha eziqeqesha amamodeli e-AI futhi zidle ama-watt njengokungathi akukho kusasa. Umqondo uwukuthi uqonde ukuthi yini eyenza i-GPU isebenze kahle, ukuthi ukusebenza kahle kwamandla kushintshe kanjani ku-AMD naku-NVIDIA, nokuthi yimaphi amamodeli athile okwamanje anikeza isilinganiso esihle kakhulu sokusebenza nge-watt ngayinye.
Kungani ukusebenza kahle kwamandla kwe-GPU kubaluleke kangaka?
Uma sikhuluma ngokusetshenziswa kwamandla kumakhadi wehluzo, sivame ukubhekisela ku-TGP noma ku-TDP, okungukuthi, amandla aphezulu kakhulu angasetshenziswa yi-GPU Ngaphansi komthwalo. Leli nani lithinta izinto eziningana: usayizi kanye nekhwalithi yokunikezwa kwamandla, inani lokushisa okukhiqizwayo, umsindo wabalandeli, kanye, nebhili kagesi uma usebenzisa imishini kakhulu.
Empeleni, ngisho nama-GPU amabili anegama lomkhiqizo elifanayo angaziphatha ngendlela ehluke kakhulu, ngoba Abakhiqizi babeka imikhawulo yamandla ehlukene kuye ngemodeli kanye nomklamo we-heatsink. Akuvamile ukubona, isibonelo, i-RTX 4050 yeselula isebenza ku-35W kanti enye ku-115W; igama liyafana, kodwa Ukusebenza kushintsha kakhulu ngalowo mkhawulo wokusetshenziswaLokhu kwenza ukuqhathanisa kube nzima kakhulu uma idatha ye-wattage ingacatshangelwa.
Ngaphezu kwalokho, yonke i-GPU inikwa amandla nge- Ibhasi le-PCI Express kanye nezixhumi zamandla ezengeziwe ngezikhonkwane ezingu-6, 8 noma 16, okuphoqa umklamo wokwakha olinganiselayo onenhloso ecacile: ukufeza ukusebenza okwengeziwe ngenkathi kugcinwa noma kunciphisa ukusetshenziswa kwamandla ngo lungiselela amaphrofayela okusebenzaEminyakeni yamuva nje, kube nentuthuko enkulu kule ndawo, kokubili emkhakheni wezemidlalo kanye nasemkhakheni wamaseva anama-GPU e-AI kanye ne-HPC (ikhompyutha esebenza kahle kakhulu).
Ukuthuthukiswa kokusebenza kahle kwamandla kuma-GPU emidlalo ye-AMD kanye ne-NVIDIA
Uma sibheka emuva eminyakeni engaba yishumi, sithola izizukulwane ezifana I-ATI Radeon HD 5000 (Terascale 2) kanye ne-NVIDIA GTX 400Ngaleso sikhathi, ukusebenza nge-watt ngayinye kwakungadingeki kakhulu uma kuqhathaniswa nalokho esinakho namuhla. Kusukela ngaleso sikhathi, ukwakheka okusha ngakunye kuye kwaqinisa izikulufo ekusebenziseni kahle amandla.
Enye yezinyathelo ezihlaba umxhwele kakhulu ifike ne- I-NVIDIA GTX 900Kwakuyisizukulwane sokuqala sama-NVIDIA GPU ukwamukela ngokucacile amasu athuthukisiwe okuhluza asekelwe kuthayela, okwasiza ekusebenziseni kangcono amandla asetshenziswayo. I-AMD yathembisa into efanayo ngobuchwepheshe bayo be-DSBR kuma-Vega GPU, kodwa ukusetshenziswa okuqinile ngempela akuzange kufike kwaze kwaba ngawo-2000. I-RDNA yokuqala (RX 5000) kanye nokuvela kwayo i-RDNA 2 (RX 6000)lapho ibhalansi yokusebenza/yokusetshenziswa ithuthuke kakhulu.
Ngokuphathelene nokusebenza nge-watt ngayinye, imizamo ye-AMD ne-RDNA ibe nkulu kangangokuthi, kwezinye izinhlobo zemikhiqizo, Bakwazile ukuzibeka ngaphezu kwe-NVIDIA RTX 30Kulokhu kumele sengeze intengo: ngokwesiko, amakhadi e-AMD abeshibhile kancane kune-NVIDIA elingana nawo, okwenza akhange kubasebenzisi abaningi, yize leyo ntengo ephansi nayo ibonise lokho. I-AMD ibisilele kancane ekusebenzeni okumsulwa izizukulwane eziningana..
Esinye isici esibalulekile kube yi-node yokukhiqiza. Ngesikhathi i-AMD ibheja ku- I-TSMC's 7nm ezizukulwaneni zayo zakamuvaUbuchwepheshe be-NVIDIA busukela enqubweni ye-TSMC engu-12nm (uchungechunge lwe-RTX 20) kuya enqubweni ye-Samsung engu-8nm (uchungechunge lwe-RTX 30). Lokhu kunikeze i-AMD inzuzo ekuqineni kwe-transistor kanye nokusebenza kahle okungabonakali ngaso sonke isikhathi uma ubheka kuphela imininingwane yobuchwepheshe. Ukusebenzisa i-node ethuthukisiwe kuvumela... ukufaka ama-transistors amaningi endaweni encane futhi kudle amandla amancane ngokusebenza okulinganayo, ngakho-ke ukuqhathanisa phakathi kwezakhiwo akuzange kube okulungile ngokuphelele.
Ngokufika kwezizukulwane ezintsha, i-AMD kanye ne-NVIDIA kulindeleke ukuthi zamukele ama-node athuthukile kakhulu (njengenqubo ye-TSMC ye-5nm kuzakhiwo zakamuva), okuyinto Kuzolinganisa inkundla yokudlala kakhulu. futhi kuzosenza sikwazi ukuqhathanisa ukusebenza kahle cishe ngokulinganayo, okwenza ikhwalithi yomklamo wezakhiwo omsulwa ibe sobala.
Ama-GPU amaseva kanye ne-AI: amandla amakhulu kanye nokusetshenziswa okukhulu kakhulu
Enye ingxenye enkulu lapho kulwiwa khona impi yokusebenza kahle yileya Ama-GPU amaseva nezikhungo zedathaUmgomo lapha akukhona ukufeza i-FPS ephezulu emidlalweni, kodwa kunalokho ukusheshisa amamodeli obuhlakani bokwenziwa, idatha enkulu, kanye nokulingisa kwesayensi okuyinkimbinkimbi kakhulu. Futhi yize ukusebenza kahle kuyinto eza kuqala, ukusetshenziswa kwamandla okuphelele kuyanda kakhulu ezibalweni ezimangalisayo.
I-NVIDIA H100: Isilo se-AI esidla ama-watts
La I-NVIDIA H100 Kuze kube namuhla, ingenye yama-GPU anamandla kakhulu kukhathalogi ye-NVIDIA ye-AI ne-HPC. Isekelwe kusakhiwo se-Hopper futhi yenzelwe izinhlangano ezidinga ukusebenza okuhle kakhulu qeqesha futhi usebenzise amamodeli e-AI esizukulwane esilandelayo, amamodeli amakhulu olimi kanye nokulingisa okuyinkimbinkimbi kwesayensi.
Phakathi kwezici zayo zobuchwepheshe, okulandelayo kuyagqama: Ubuchwepheshe bokukhiqiza obuthuthukisiwe be-4nm (TSMC)Ihambisana nememori ye-HBM3 efika ku-80 GB kanye namandla amakhulu ekhompyutha: amashumi e-TFLOPS ku-FP64 kanye nama-TFLOPS angaphezu kwe-1000 asebenzisa ama-Tensor cores ayo ngendlela encane. Iphinde ifake ukwesekwa kwe- izinhlobo zedatha njenge-FP8okuvumela ukuthuthukiswa okukhulu ekusebenzeni kahle kokuqeqesha amamodeli amakhulu ngaphandle kokulahlekelwa ukunemba okukhulu.
Ngenxa yesizukulwane sesine se-NVLink, i-H100 can xhumana namanye ama-GPU ukwakha amaqoqo angempela asebenza kahle kakhulu, okwandisa amandla okusebenzisa ikhompyutha. Intengo yakho konke lokhu ukusetshenziswa kukagesi okukhulu: ekucushweni kwe-AI sikhuluma ngakho cishe ama-watts angu-700 nge-GPU ngayinyeUkuze sikubeke esimweni salokho, lokho kungaphezu kwalokho okusetshenziswa ama-PC amaningi aphelele edeskithophu, kanye namakhadi ehluzo emidlalo asezingeni eliphezulu ahlanganisiwe.
Uma uphinda lokho kusetshenziswa nge izigidi zamayunithi okulindeleke ukuthi asetshenziswe eminyakeni ezayoIsibalo sokusetshenziswa kwamandla sonyaka sifinyelela amazinga afana nokusetshenziswa kukagesi kwamanye amazwe. Kulinganiselwa ukuthi, ngama-H100 kuphela, ugesi ongu-13.000 GWh ngonyaka ungadlulwa kalula, okungaphezu kokusetshenziswa okuphelele kwamazwe afana neGuatemala noma iLithuania. Lokhu kuphakamisa. imibuzo engathi sÃna yezemvelo kanye namandla ekukhuleni kwe-AI.
I-NVIDIA A30: ukulinganisela ukusebenza kanye nokusetshenziswa kwamandla ukuze kutholakale isiqondiso
Ngaphambi kwe-H100, i- I-NVIDIA A30 Kuyi-GPU yeseva elinganiselayo, esekelwe kusakhiwo se-Ampere. Iqondiswe kumabhizinisi adinga Ukusebenza kahle, kodwa futhi kuqaphela ukusetshenziswa kanye nesabelomali.Kuyindlela esetshenziswa kabanzi yokuqagela i-AI (kusetshenziswa amamodeli aqeqeshwe kusengaphambili), i-HPC elinganiselayo, kanye nokwenza izinto zibe yi-virtualization.
Ikhiqizwe nge-7 nm yi-TSMC, i-A30 inikeza kufika ku- Ama-TFLOPS ayi-10 ku-FP64 kanye nama-TFLOPS angaba ngu-165 kuma-Tensor coresinememori ye-HBM2 engu-24 GB kanye nekhono lokuxhumanisa amakhadi amabili nge-NVLink. Inzuzo yayo eyinhloko ukuthi Ithola amandla amakhulu okusebenzisa amandla ngokuqhathaniswa nokusetshenziswa kwamandla okuphansi.Lokhu kwenza kube kuhle kakhulu ezikhungweni zedatha ezifuna ukusebenza kahle kanye nezindleko zikagesi ezifanele.
I-Intel Gaudi 2: enye indlela eqinile esikhundleni se-NVIDIA ye-AI
I-Intel, ngokusebenzisa i-Habana Labs, ithuthukise i- Izisheshisi zeGaudi njengendlela ehlukile kuma-GPU e-NVIDIA yemithwalo yemisebenzi ye-AI. Imodeli Gawudi 2 Yenzelwe ngqo ukuqeqeshwa kwenethiwekhi ye-neural, enama-cores angu-24 e-Tensor kanye nokugxila okucacile ekwandiseni ukusebenza kahle kanye nokusebenza kahle kulolu hlobo lomsebenzi.
I-Gaudi 2 ikhiqizwa ku-7 nm, futhi ihlanganisa 96 GB yememori ye-HBM2e kanye nesistimu yesofthiwe evulekile. Amandla ayo amakhulu ukuhlanganiswa kobuchwepheshe obufana ne-RDMA ne-RoCE, okuvumela ukufinyelela okuqondile kwememori phakathi kwama-node kanye nokukhuliswa okuthuthukisiwe lapho kusetshenziswa ama-accelerator amaningi ngesikhathi esisodwa. Nakuba ukusetshenziswa kwamandla ayo ngqo kungaxoxwa ngakho kakhulu njengasendabeni ye-H100, ukwakheka kwayo kuhloselwe ukunikeza isilinganiso sokusebenza/sokusetshenziswa esincintisana kakhulu.
I-Intel Gaudi 3: Ukusebenza okwengeziwe kanye nokusebenza kahle okungcono ngezinga
El Intel Gaudi 3 Kungukuvela kwemvelo kwe-Gaudi 2 futhi kumelela intuthuko enkulu ekusebenzeni nasekukhuleni kwe-AI yokukhiqiza, amamodeli olimi olunwetshiwe (i-LLM), kanye nokucutshungulwa kwezithombe. Nge-node engu-5nm kanye nokwakheka okwenziwe ngcono kwe-FP8, ingafinyelela kufika ku-1,835 PFLOPS ngaleyo fomethi, futhi uphathe imemori ye-HBM2e efika ku-120 GB.
Ingqalasizinda yayo yokuxhumana ethuthukisiwe ivumela yakha amaqoqo amakhulu ngendlela ephumelela kakhuluLokhu kubalulekile lapho kuqeqeshwa amamodeli amakhulu. Ezivivinyweni zokuqeqeshwa ze-LLM eziqhathaniswayo, i-Gaudi 3 ingaba ngokushesha okuphindwe izikhathi ezingu-1,7 kunezinye izindlela ezisungulwe kahle, okuhunyushwa ngokuthi isikhathi esincane sokubala kanye nokusebenza kahle kwamandla okungcono okudingekayo ukuze kufinyelelwe izinga elithile lokusebenza.
Umthelela kwezemvelo kanye nesidingo samandla ahlanzekile
Ukukhula kokusetshenziswa kwama-GPU e-AI kanye nezikhungo zedatha kuholele ekutheni ukusetshenziswa kwamandla kwalezi zingqalasizinda kukhuphuka kakhuluAkukhona nje kuphela ngamandla adingwa yizisheshisi ezifana ne-H100 noma i-Gaudi 3, kodwa futhi nezinye izingxenye zehadiwe ezisekelayo, futhi ngaphezu kwakho konke, izinhlelo zokupholisa ezidingekayo ukuze zigcinwe ngaphakathi kwemingcele yazo yokushisa.
Izinkampani ezinkulu zobuchwepheshe ezifana ne-Meta, i-OpenAI, nezinye ezifanayo sezivumile kakade ukuthi zizokondla ingqalasizinda yazo ye-AI. Kudingeka amandla amaningi kakhulu...kuze kube seqophelweni lokuxoxa ngokusobala ngezitshalo zamandla enuzi ezizinikele noma ukutshalwa kwezimali okukhulu ezintweni ezivuselelekayo. Lokhu kuphakamise ukukhathazeka okukhulu mayelana nomthelela wesimo sezulu wokuthuthukiswa kwe-AI, ikakhulukazi njengoba eziningi zalezi zinhlelo zokusebenza zifinyelela kubasebenzisi bokugcina njengezinsizakalo ezibonakala zikhululekile.
Uma sibheka lesi simo, kubalulekile ukuthi Izinkampani eziklama ama-GPU kufanele zithuthukise kokubili ukusebenza kahle kanye nomthombo wamandla awanikeza amandla.Akusizi ngalutho ukuthi i-GPU isebenze kahle ngo-20% uma inani eliphelele lama-accelerator asetshenziswayo liphindaphindwa kayishumi futhi uphethiloli wezinto ezisetshenziswayo uqhubeka nokusetshenziswa. Lapha, ukuqhathaniswa kokusetshenziswa kanye nokusebenza kahle akusekho nje indaba yezikweletu zikagesi futhi kuba umbuzo womthwalo wemfanelo wezemvelo.
Amakhadi ehluzo anamandla aphansi ama-PC asekhaya kanye nemidlalo
Emkhakheni wama-PC asekhaya kanye nemidlalo, isimo sihlukile. Kubasebenzisi abaningi, ukuba ne-GPU esebenza kahle kusho gcina amandla kagesi, nciphisa ukushisa ebhokisini futhi ugweme umsindo omkhulu...ngaphezu kokukhokha kancane ekupheleni kwenyanga. Ama-GPU anamandla aphansi ahlala eyisigaba esithakazelisayo kakhulu, ikakhulukazi ebangeni eliphakathi nendawo... Ukuthuthukiswa kwegeyimu nokuphathwa kwayo ku-Steam.
I-GPU ngokuvamile ibhekwa njengenamandla aphansi uma Akudluli cishe ku-150W ngaphansi komthwalo wangempelaFuthi esimweni esibi kakhulu, idinga isixhumi esisodwa samaphini ayi-8 noma ingadingi nhlobo, idonsa amandla kuphela kusuka ku-PCI Express slot. Lokhu kuyivumela ukuthi isebenze kahle ezinhlelweni ezineziphakeli zamandla ezingu-300 kuya ku-400W, ezilungele kakhulu yakha ama-PC angabizi kakhulu noma uthuthukise imishini emidala ngaphandle kokushintsha ingxenye yesikhulumi.
Yini elethwa yi-GPU enamandla aphansi kwi-PC yangempela?
Isibonelo esijwayelekile kungaba yi-PC ene- I-Ryzen 5 5600X, i-16 GB ye-DDR4 RAM, i-PCIe SSD, kanye ne-cooler ye-AIO engu-240 mmUma ufaka i-GPU esebenza kahle njenge-RTX 4060 noma i-RTX 5060 yesikhathi esizayo kulolo hlelo, ukusetshenziswa kwamandla esistimu okuphelele ngaphansi komthwalo kuvame ukuba phakathi kuka-280 no-320 W, okushiya ugesi ongu-400 W unesikhala esanele. Ngaphezu kwalokho, i-GPU efanele Ukuthuthukiswa kokusebenza kahle ku-Windows 11 Kuyasiza ukugcina lawo mazinga okusebenzisa elawulwa.
Ngaphezu kwalokho, la makhadi ehluzo, Azidingi ama-CPU asezingeni eliphezulu ukuze asebenze kahle.Ngakho-ke, zingahlanganiswa nama-processor ashibhile ngaphandle kokwesaba ukuthi kuzoba nenkinga enkulu. Ngokuvamile zincane futhi kulula ukuzipholisa, zinamafeni angadingi ukujikeleza ngesivinini esiphezulu, okuhunyushwa ngokuthi imishini epholile futhi ethulile.
Amamodeli athile ama-GPU asebenzayo kanye nokusetshenziswa kwawo kwamandla
I-GeForce GTX 1650 GDDR6: inketho engabizi kakhulu futhi engabizi kakhulu
La I-GeForce GTX 1650 enememori ye-GDDR6 Kungenye yezinketho ezinhle kakhulu uma ufuna imodeli eshibhile kakhulu esetshenzisiwe yokuvuselela ikhompyutha endala. Ine-TGP engu-75W kuphela futhi inikwe amandla yi kusuka kuphela echwebeni le-PCI Express ezinguqulweni eziningi futhi ingasebenza ngaphandle kwezinkinga ngezinsizakusebenza zikagesi ezindala ezingama-300W.
Ngokuphathelene nokusebenza, le GPU isebenza ngendlela efanayo ne- I-GeForce GTX 970 noma i-Radeon RX 570Kodwa-ke, inenzuzo yokusebenzisa ukwakheka kwesimanje (i-Turing) kanye nokusekelwa kwamanje komshayeli. Isengakwazi ukudlala imidlalo engadingi noma emidala ku-1080p ngekhwalithi eyamukelekayo, okwenza kube mnandi kakhulu ezinhlelweni ezidinga kuphela ukuthuthukiswa kwehluzo okulinganiselwe.
I-GeForce RTX 4060: Ukusebenza kahle kakhulu kwe-1080p
La I-NVIDIA GeForce RTX 4060 Liyikhadi elihlonishwa kakhulu, elibaluleke kakhulu ngenxa yokulinganisela kwalo phakathi kokusebenza kanye nokusetshenziswa kwamandla. Ngokusekelwe kusakhiwo se-Ada Lovelace (i-AD107 core), linikeza amandla angaphezu kwanele dlala ku-1080p ngekhwalithi ephezulu noma ephezulu kakhulu ezihlokweni zamanje, ngokusekelwa kobuchwepheshe obufana ne-DLSS.
I-TGP yakhe imayelana 115 W futhi isebenzisa isixhumi esisodwa samaphini ayi-8Ngakho-ke, isebenza kahle kakhulu ngezinsiza zikagesi ezingu-350-400W, uma nje lonke uhlelo lungadingi amandla amaningi. Iza nama-shaders angu-3.072, i-8GB ye-GDDR6 ku-17Gbps, ibhasi elingu-128-bit, kanye nenani elihle lama-Tensor nama-RT cores. Intengo yayo ivame ukuba ngaphansi kuka-€300, okuthi, uma kucatshangelwa ukusetshenziswa kwayo kwamandla kanye nesipiliyoni sokudlala esikunikezayo, kuyenze ibe inketho enhle kakhulu. elinye lamakhadi ehluzo asebenza kahle kakhulu i-NVIDIA ewakhiphe eminyakeni yamuva nje.
I-GeForce RTX 5060: Ukusebenza okusebenzayo kuyanda ngokusetshenziswa kwamandla okuphansi
La I-GeForce RTX 5060 Kuthatha isinyathelo esengeziwe ekusebenzeni, kufaka phakathi ukwakheka kweBlackwell (i-GB206 core) kanye nememori ye-GDDR7 eshesha kakhulu. Kwandisa inani lama-shaders libe ngu-3.840 futhi kuthuthukisa ukusebenza uma kuqhathaniswa ne-RTX 4060, okwenza kube ngokoqobo. ezingeni le-8GB RTX 4060 Ti emandleni aluhlaza.
I-TGP yakhe ikhuphukela ku 145 WIsalokhu ingaphakathi kwalokho okungabhekwa njengokusetshenziswa kwamandla okulingene. Idinga isixhumi esingu-8-pin, futhi ngokulungiswa okuvamile kwe-Ryzen 5 5600X, i-RAM engu-16 GB, kanye ne-SSD, uhlelo oluphelele akufanele ludlule i-320 W ngaphansi komthwalo. I-GDDR7 yayo engu-8 GB kanye nokusekelwa kwamasu okukhiqiza uhlaka lwesizukulwane esilandelayo kuyayinika amandla. Isikhathi eside sokuphila ku-1080p kanye nokufakwa ku-1440pngaphandle kokusetshenziswa okukhulu.
I-AMD Radeon RX 7600: enye indlela ephumelelayo esikhundleni se-RTX 4060
Kulabo abathanda i-AMD, i- Radeon RX 7600 Kuyinselele ethakazelisayo esezingeni eliphakathi. Ngokusekelwe ku-RDNA 3 (Navi 33), inama-shaders angu-2.048, i-8 GB ye-GDDR6 ku-18 Gbps kanye nebhasi engu-128-bit, kanye namayunithi angu-32 okulandelela imisebe azinikele. Ngokuphathelene nokusebenza, Isebenza eduze kakhulu ne-RTX 4060 ku-1080p, okwenza kube inketho enhle yaleso sinqumo.
I-TGP yakhe imayelana 165 W futhi idinga isixhumi esinamaphini angu-8Kuhlelo olujwayelekile olune-Ryzen 5 5600X kanye nokulungiselelwa okulinganiselayo, luhlala lusebenza kahle ngogesi wekhwalithi engu-400W. Lapho lukhanya khona kakhulu kuse- ibhalansi phakathi kokusetshenziswa nokusebenza manje njengoba intengo yayo yehlileUma uyithola ngama-euro angaba ngu-260 noma ngaphansi, ngokuvamile kuyintengo enengqondo.
I-GeForce RTX 5060 Ti 16 GB: Amandla aphezulu ngaphandle kokusetshenziswa okuphuthumayo
Uma sikhuphuka isinyathelo esisodwa sithola I-GeForce RTX 5060 Ti 16GBLeli khadi lisebenza kahle kakhulu ngenkathi ligcina ukusetshenziswa kwamandla okunengqondo. Linomklamo we-Blackwell kanye ne-core ye-GB206, linikeza ama-shaders angu-4.608, i-16 GB ye-GDDR7 ku-28 Gbps, ibhasi le-128-bit, kanye nenani elihle lama-Tensor nama-RT cores.
I-TGP yakhe iseduze 180 WNoma kunjalo isakwazi ukwenza kahle kakhulu kune-Radeon RX 7700 XT, ene-TGP engu-245W. Idinga kuphela isixhumi esisodwa esinamaphini angu-8, okwenza kube lula kakhulu ukufaka. Ku-PC ene- I-Ryzen 5 7600, i-32 GB ye-DDR5, i-NVMe SSD kanye ne-cooler engu-240 mmUkusetshenziswa kwamandla okuphelele kungama-360W ngaphansi komthwalo ngale GPU. Kungcono kakhulu, kulokhu, kufanele ukhethe izinsiza zikagesi zekhwalithi enhle ezingama-450-500W ukuba nendawo eningi nokugcina umthombo usendaweni yawo yokusebenza kahle kakhulu.
I-GeForce RTX 4070: umkhawulo ophezulu "wokusetshenziswa kwamandla okunengqondo"
Okokugcina, imodeli ehambisana nalokho abanye abangeke besakubheka njengokusetshenziswa kwamandla okuphansi, kodwa isencane kakhulu uma iqhathaniswa namakhadi aphezulu kakhulu: I-GeForce RTX 4070Ngokwakhiwa kwe-Ada Lovelace (AD104), ama-shaders angu-5.888, i-12 GB ye-GDDR6 ku-21 Gbps kanye nebhasi engu-192-bit, iyi-GPU eyenzelwe dlala ngo-1440p konke kuphelele futhi ubheke ngisho ne-4K.
I-TGP yakhe imayelana 200 Wfuthi ngokuvamile idinga izixhumi ezimbili ezinamaphini ayi-8 noma isixhumi esisodwa samaphini ayi-16, kuye ngohlobo. Ekucushweni okuvamile nge-Ryzen 5 7600, i-32 GB ye-DDR5 RAM, kanye nokupholisa koketshezi, uhlelo oluphelele ludonsa cishe ama-380-390 W ngaphansi komthwalo. Kule nhlanganisela, okulandelayo kuyanconywa: ugesi omuhle ongu-500-600Wokuvumela i-margin futhi okuqinisekisa ukusebenza kahle okuphezulu kwe-PSU uqobo.
Ukuqhathanisa okusekelwe ekusetshenzisweni kwangempela: inselele enkulu
Inkinga ephindaphindayo lapho kuqhathaniswa ama-GPU ukuthi amawebhusayithi amaningi agxila ku- imiphumela yokulinganisa yeqembu ngemodeliLokhu kwenziwa ngaphandle kokucabangela amandla angempela okukhipha inguqulo ngayinye, ikakhulukazi kuma-laptop. Lokhu kuholela ekuqhathaniseni okudukisayo lapho, isibonelo, i-RTX 3050 Ti yeselula eboshwe ku-35W ibonakala ixubene nezinye izinhlobo ku-60 noma ku-80W, okuhlanekezela umbono wokusebenza kwayo kwangempela.
Okungcono kakhulu, uma uqhathanisa ukusetshenziswa kwamandla, umuntu kufanele akwazi ukukhetha hhayi nje imodeli ye-GPU kodwa futhi nomkhawulo wayo wamandla othize (isibonelo, i-RTX 3050 Ti ku-35W uma iqhathaniswa ne-RTX 4050 ku-45W) bese ubona ukusebenza okuhlobene esimweni ngasinye. Nakuba kunezizindalwazi kanye nohlu oluhlela ama-GPU nge-wattage, kokubili ideskithophu ne-laptop, Azihlanganisi njalo imiphumela yokusebenza ehlotshaniswa nokucushwa kwamandla ngakunyeyikho okusivumela ngempela ukuthi sihlole ukusebenza kahle.
Ngaphezu kwalokho, kufanele kukhunjulwe lokho I-silicon efanayo ingaziphatha ngendlela ehluke kakhulu kuye ngomklamo womenzi. (i-VRM, ukupholisa, imikhawulo yamandla), ikakhulukazi kuma-laptop, lapho i-variant enamandla aphansi ihlose ukwandisa impilo yebhethri nokunciphisa ukushisa, kuyilapho i-variant enamandla aphezulu ibeka phambili ukusebenza okuluhlaza. Ngakho-ke, lapho kuhlaziywa ukusetshenziswa kwamandla, ukubheka nje imininingwane yobuchwepheshe akwanele: Kubalulekile ukucabangela ukuhlolwa komthwalo kanye nezimo eziqondile zemodeli ngayinye. bese wenza i- ukuhlanza uhlelo olujulile ngaphambi kokulinganisa.
Lesi sithombe sonke, kusukela kuma-GPU angabizi njenge-GTX 1650 GDDR6 kuya ezilo ezifana ne-NVIDIA H100 noma i-Intel Gaudi 3, sibonisa ukuthi ukusebenza kahle kwamandla sekuyinto ebalulekile ekuklanyweni nasekukhethweni kwamakhadi ehluzo.
Ukuqonda ukuthi ukusetshenziswa, ukwakheka, indawo yokukhiqiza, ukusebenza, kanye nokusetshenziswa okuthile kuhlangana kanjani kuvumela izinqumo ezinengqondo kakhulu, kungakhathaliseki ukuthi ukwakha i-PC yokudlala engabizi enesilinganiso esihle sokusebenza/i-watt noma ukulinganisa i-AI cluster engaguquli ibhili kagesi ibe yinto eshaqisayo engapheli. Yabelana ngalolu lwazi ukuze abantu abaningi bakwazi ukufunda ngalesi sihloko.