Eminyakeni yamuva nje kube nokukhulunywa okuningi mayelana nezixazululo ze-4K, ukulandelela imisebe, kanye ne-DLSS, kodwa enye yemikhawulo emikhulu, evame ukunganakwa yahlala iyimemori yehluzo. ukuthungwa okusindayo kanye nemidlalo edinga kakhuluNgisho namakhadi amaningi ehluzo aphakathi nendawo eYurophu aphelelwa yi-VRAM ngokushesha, okwaphoqa abasebenzisi ukuthi banciphise izilungiselelo zekhwalithi noma baphile ngokuthuthumela kanye nokwehla kokusebenza.
Kulowo mongo kuyavela Ukucindezelwa Kwe-Neural Texture ye-NVIDIA (NTC), ubuchwepheshe bokucindezela ukuthungwa obusekelwe kumanethiwekhi e-neural okwethulwe ngokuningiliziwe ngesikhathi se-GTC kanye ne-GDC ngo-2026. Isiphakamiso sayo silula: ukunciphisa ngamandla ukusetshenziswa kwememori kokuthungwa, kodwa ngaphandle kokubi kakhulu isithombe esibonwa esikrinini, nokusithuthukisa kwezinye izimo uma kuqhathaniswa nezindlela zakudala.
Indlela Ukusebenza Kokucindezela Kwezinzwa Nokwakheka Kwezinzwa kanye nokuthi yini eyenza kube okuhlukile
Isisekelo se-NTC ukusetshenziswa amanethiwekhi amancane e-neural aqeqeshwe ngokukhethekile ukwakhekaEsikhundleni sokuthembela kuphela kumafomethi okucindezela amabhulokhi e-BCn (BC5, BC6, BC7, njll.) imboni ebilokhu iwasebenzisa iminyaka eminingi. Ngokwesiko, lawa ma-texture agcinwa kakade ecindezelwe ku-VRAM futhi i-GPU iyawachaza ngokushesha, kodwa asaphethe inani elikhulu lememori.
Ngokucindezelwa Kwesimo Sezinzwa, Ulwazi lokuthungwa lugcinwa ngendlela ecacile kakhuluLolu uhlobo lokufaka ikhodi okufihliwe okuqoshwa yinethiwekhi ye-neural ngesikhathi sangempela lapho kwenziwa uhlaka ngalunye. Esikhundleni sokuphatha ama-gigabytes amamephu okusabalalisa, okujwayelekile, ukurhabaxa, njalo njalo, i-GPU isebenza nesethi yedatha encane kakhulu.
Ngokusho kwezincazelo ze-NVIDIA, lawa mamodeli e-neural ane- baqeqeshwe ukuqonda ukuthi i-texel kufanele ibukeke kanjani ("iphikseli" yokwakheka) kwezinto ezithile: itshe, ukhuni, ukuthungwa kwensimbiizinto zobumba, izindwangu, njll. Kusukela kulokhu kufunda, inethiwekhi ingakha kabusha ukubukeka kokugcina kusuka kudatha ecindezelwe, ilingise umphumela obonakalayo esingaba nawo ngezindwangu eziningi kakhulu.
Umphumela osebenzayo uwukuthi ukuthungwa kuyayeka ukuba "umthwalo" oqinile enkumbulweni futhi kuncike ku- ikhodi encane ecindezelwe kanye namakhono okucabanga e-AILokhu kuhambisana nomkhuba we-NVIDIA wokushintsha inkumbulo kanye nomthwalo we-bandwidth uye kwikhompyutha ehlakaniphile kuma-GPU ayo.
Idemo ethi "Tuscan Wheels / Tuscan Villa": kusukela ku-6,5 GB kuya ngaphansi kwe-1 GB ye-VRAM
Ukuze kuboniswe amandla obuchwepheshe, i-NVIDIA ibonise imiboniso eminingana yobuchwepheshe, okuhlanganisa nendawo eyaziwayo manje. “Amasondo aseTuscan” noma “I-Tuscan Villa”, indawo yokuhlala yendlu yaseMedithera enezakhiwo zangaphakathi ezinemininingwane esebenza njengendawo yokuhlola esezingeni eliphezulu yemininingwane.
Ekucushweni kwendabuko, kusetshenziswa amafomethi Isigcawu esijwayelekile se-BCn sidinga cishe i-6,5 GB ye-VRAM Lokhu kusebenza kuphela ekuthungeni. Ngokuvumela ukucindezelwa kwe-Neural Texture, indawo efanayo isebenza cishe 970 MB yememori yehluzoOkusho ukuthi, ukwehla cishe ngo-85% uma kuqhathaniswa nokusetshenziswa kokuqala. Izibalo ezifanayo zibonwe kwezinye imiboniso, kanye nokwehla cishe ngo-80% (kwehle kufike cishe ku-670 MB kwezinye izinhlobo zezigcawu).
Okubalulekile akukhona nje ubukhulu bokwehla kwe-VRAM, kodwa lokho Ukuqhathanisa okubonakalayo phakathi kwalezi zinguqulo ezimbili cishe akubonakali. kumsebenzisi ojwayelekile. Ngokusho kwe-NVIDIA, uma "isabelomali" sememori esifanayo sigcinwa, i-NTC iyakwazi ngisho nokugcina imininingwane emincane kakhulu kunezimo ze-BCn ezincishisiwe noma eziphinde zalinganiswa.
Isibonelo esibonakalayo singabonakala ngaphakathi kwendlu, enetafula elimbozwe ngezitsha zokudlala, amabhodlela, nezinto zokuhlobisa. Kulokho kuqhathaniswa, Ingxenye ecutshungulwe nge-NTC, enenani elifanayo lememori, ikhombisa ubukhali obukhulu kanye nemininingwane emincane. ukuthi isigaba esihunyushiwe esinezimo ze-BCn ezincishisiwe silingane nesabelomali se-VRAM esifanayo.
Lezi zinhlobo zemiphumela ziphakamisa ukusetshenziswa okubili kobuchwepheshe: izifundo zingakhetha kunciphisa kakhulu ukusetshenziswa kwememori ngaphandle kokulahlekelwa ikhwalithi noma gcina ukusetshenziswa futhi uthuthukise ukwethembeka kokubona, into ekhangayo kakhulu kumaphrojekthi afisa ukubukeka okufana nokungokoqobo kwesithombe.
Izinzuzo ezisebenzayo zemidlalo nezinjini zehluzo
Ngokombono wentuthuko, umphumela oyinhloko wakho konke lokhu ukuthi Ukuthungwa akusanciphisi kangako lapho kuklanywa izigcawu eziyinkimbinkimbi.I-VRAM encane ehlala kumamephu obuso isho indawo enkulu yezinye izinhlelo noma yokwandisa inani lokuqukethwe okuboniswa ngasikhathi sinye.
Ngemidlalo ye-PC ehloselwe imakethe yaseYurophu, lapho ingxenye enkulu yabasebenzisi isasebenza khona... ama-GPU aphakathi nendawo noma lawo anememori engu-8 GBUkucindezela okunjalo okunamandla kuvula umnyango wokuvumela ukuthungwa okunesinqumo esiphezulu ezinhlelweni, okwamanje, eziphoqelekile ukusika ikhwalithi ukuze zigweme ukudlula umkhawulo we-VRAM.
Phakathi kwezinzuzo i-NVIDIA ezigqamisile yilezi:
- Ukwehla okukhulu kokusetshenziswa kwe-VRAM, kuze kube izikhathi eziyisikhombisa ezimweni ezithile.
- Amandla okuphatha ukuthungwa okunesisombululo esiphezulu ngaphandle kokudala izidingo zememori.
- Ukukhululeka kwe-Bottleneck okuhlobene nomkhawulokudonsa wememori kanye nokusakaza kwempahla.
- Kungenzeka ukufakwa okuncane kanye nosayizi we-patch, ngokufaka ukwakheka okuqinile okwengeziwe kudiski.
- Ukusetshenziswa okungcono amadivayisi aphathekayo kanye nama-consoles esikhathi esizayolapho inkumbulo iyinsiza elinganiselwe kakhulu.
Konke lokhu kuhambisana nemakethe lapho, ngisho naseYurophu, imidlalo idlula kalula usayizi wokulanda ongu-100 GB futhi lapho i-bandwidth etholakalayo ingahlali yanele, ikakhulukazi ezindaweni zasemakhaya noma ezinokuxhumana okulinganiselwe. Nciphisa usayizi wokuthungwa ngaphandle kokulahlekelwa ikhwalithi Kungenza umehluko ezikhathini zokulanda kanye nezibuyekezo.
Ngaphezu kwalokho, ngokuthembela kakhulu ekubalweni okuhlakaniphile, okunye kokucindezeleka kwimemori ebonakalayo kuyancishiswa futhi amakhono okusebenzisa ama-GPU anamuhla asetshenziswa kahle kakhulu, into i-NVIDIA ebilokhu iyikhuthaza isikhathi esithile ngezinye izixazululo ze-AI.
Izinto Zezinzwa: iziteshi ezimbalwa, isivinini esengeziwe
Kanye ne-NTC, i-NVIDIA iphinde yethula umqondo wokuthi Izinto Zezinzwa, ukunwetshwa kwemvelo komqondo wokucindezela kwemizwa okusetshenziswa hhayi kuphela ekubunjweni, kodwa nasemodeli ebonakalayo yezinto ezisetshenziswa ekudwebeni.
Kuhlelo lokusebenza lwendabuko, ukuchaza ukuthi ubuso buziphatha kanjani uma buphendula ekukhanyeni, okulandelayo kuyahlanganiswa iziteshi eziningi namamephuUmbala wesisekelo, okujwayelekile, ukurhabaxa, insimbi, ukuvaleka, kanye nolunye ulwazi oluthile oluxhunywe ku-equation ye-BRDF esetshenziswa yinjini yehluzo. Lokhu kuhumushela kudatha eningi, ukufinyelela okuningi kwememori, kanye nemisebenzi eminingi yezibalo ngephikseli ngayinye.
Ngezinto Zokusebenza Kwezinzwa, Leli sethi leziteshi lincishisiwe laba ukumelwa okucashile okuncane kakhulu inethiwekhi encane ye-neural inesibopho sokubhala ikhodi ngesikhathi sangempela, yakha kabusha izakhiwo ezibonakalayo zezinto ngesikhathi sokudweba.
Ekuhlolweni okwabiwe yi-NVIDIA, ukucushwa kwe Iziteshi zezinto ezingu-19 zancishiswa zaba iziteshi ezingu-8 kuphela, okuhunyushwe kuzo, ezigcawini zokuhlola ku-resolution engu-1080p ukusheshisa okuphakathi kwezikhathi ezimbili neziyishumi ngesikhathi sokukhishwa, kuye ngesimo esithile.
Le ndlela ayigcini nje ngokusindisa inkumbulo, kodwa futhi Kwenza kube lula inani lokufinyelela idatha kanye nokusebenza ngephikseli ngayinye.Lokhu kubaluleke kakhulu ekulandeleni imisebe kanye nokuhlelwa kokulandelela indlela, lapho izindleko zokukhanya ngakunye ziphindaphindwa khona.
I-NTC ngaphakathi kwepayipi lehluzo elisha le-NVIDIA elisebenzisa i-AI
Ukucindezelwa Kwesimo Sezinzwa akuzi kodwa. Kuyingxenye yesu elibanzi lapho I-NVIDIA iklama kabusha izingxenye ezibalulekile zombhobho wehluzo ngokusebenzisa amanethiwekhi e-neural.Ubuchwepheshe obufana ne-DLSS, ukukhiqizwa kohlaka, noma i-DLSS 5 ezayo busekelwe emcabangweni ofanayo oyisisekelo: ukushintsha omunye womsebenzi "onzima" wendabuko wokwenza amamodeli e-AI alungiselelwe.
Ezingxoxweni eziningana zobuchwepheshe ze-GTC kuchazwe ukuthi, ngale kokucubungula isithombe okulula, Ukucindezelwa kanye nokuvezwa kwemizwa kokuthungwa nezinto zokwakha kuyizinto ezibalulekile ukuze leyo ndlela yokusebenza isebenze kahle. Ukunciphisa inkumbulo, ukunciphisa izithiyo, nokukhulula izinsiza kuvumela ezinye izigaba, njengokwakhiwa kabusha kwesithombe se-DLSS, ukuthi zibe nesikhundla esikhulu.
Iphuzu elilodwa eligcizelelwa yi-NVIDIA ukuthi, endabeni ye-NTC kanye nezinto ze-Neural, I-AI ekhiqizayo ayisetshenziswa "ukusungula" okuqukethwe kobucikoKunalokho, ziyimodeli zokucabanga ezenzelwe ukuphinda ngokwethembeka ukubukeka kwezindwangu nezinto ezichazwe ngaphambilini yiqembu lobuciko. Lokhu kuhlose ukubhekana nokugxekwa okuye kwavela emphakathini mayelana nomthelela ongaba khona we-AI enhlosweni yokuqala ebonakalayo yemidlalo.
Empeleni, umgomo oshiwoyo ukuthi amathuluzi e-neural asebenze njenge izisheshisi zobuchwepheshefuthi hhayi njengezinto ezithatha indawo yomsebenzi wokudala wabaculi nama-studio, into ezoqhubeka nokudala impikiswano njengoba lezi zixazululo zifinyelela ezihlokweni zezentengiselwano.
Ngokusho kwe-NVIDIA, Amanethiwekhi asekela i-NTC asevele eqeqeshwe ngezinto eziningi ezitholakala emidlalweni yevidiyo.Lokhu kuzokwenza kube lula ukuhlanganiswa kwayo ezinjinini zezentengiselwano uma ubuchwepheshe sebuvulelwe onjiniyela.
Umthelela ongaba khona emakethe yaseYurophu kanye nokuthunyelwa kwesikhathi esizayo
Kuze kube manje, i-NVIDIA ayikakabeki usuku oluthile lokuthi i-Neural Texture Compression isetshenziswe kabanzi emidlalweni yezentengiselwano, kodwa Imiboniso yobuchwepheshe ikhomba isimo lapho ukusetshenziswa kwe-VRAM kungaba ncane khona.ikakhulukazi kwi-PC.
EYurophu, lapho isimo sehadiwe sihluke kakhulu, lolu hlobo lwesisombululo lungaba nomthelela ocacile. Abasebenzisi abaningi badlala kusukela Ama-laptop noma ama-desktop emidlalo anama-GPU ane-6 noma i-8 GB ye-VRAMLeli nani seliqala ukuncipha kwezinye izinguqulo ze-AAA ezinezimo "eziphezulu". Uma i-NTC igcwalisa isithembiso sayo, ingavumela lezi zilungiselelo eziphezulu ukuthi zigcinwe ngaphandle kokuthi umdlalo ubhekane nokushintshana noma ukwehla kokusebenza ngokuzumayo.
Ngokombono wezifundo, kukhona nezikhuthazo ezisebenzayo: Idatha encane yokuthungwa isho ukwakheka okululaUkulanda okuncane kwesilingo kanye nezikhathi zokubuyekeza ezinengqondo. Kubadlali baseYurophu, lapho kungebona bonke abantu abanokufinyelela kokuxhumeka kwe-fiber optic okusheshayo, lokhu kungahumushela ekubeni yinto engakhungathekisi kangako lapho kufakwa noma kubuyekezwa imidlalo emikhulu.
Kodwa-ke, kuzoba nezici okufanele ziqashwe. Ukwamukelwa kwangempela kwe-Neural Texture Compression kuzoncika ku- ukulula kokuhlanganiswa ezinjinini ezifana ne-Unreal Engine, i-Unity, noma ezinye izinjini zangaphakathi, kuye ngokusekelwa okunikezwa yizizukulwane ezahlukene zama-GPU kanye nebhalansi phakathi kwekhwalithi, ukusebenza kanye nezindleko zokusebenzisa zesitudiyo ngasinye.
Kunoma yikuphi, okubonakala kucacile yilokho Imemori yezithombe isibe yinto ebaluleke kakhulu yokwenza ngcononokuthi iziphakamiso ezifana ne-NTC ziyahambisana nomkhuba obanzi obheke "ekuguquleni kwe-neural", lapho ukubala okuhlakaniphile kuthatha indawo yezinye zezixazululo zendabuko ezisekelwe kuphela emandleni amakhulu.
Uma sibheka isithombe esikhulu, i-Neural Texture Compression, i-Neural Materials, kanye namanye amasu amenyezelwe azungeze i-DLSS akhomba isizukulwane sezinjini zehluzo lapho I-AI ayidali amaphikseli kuphela, kodwa futhi inquma ukuthi iwagcina kanjani, iwacindezele, futhi iwakhe kabusha kanjani.Uma izithembiso zokonga i-VRAM, imininingwane ethuthukisiwe, kanye nezikhathi zokukhiqiza ezincishisiwe zibonakala emidlalweni yezentengiselwano, singase sibheke olunye lwezinguquko ezibaluleke kakhulu eminyakeni yamuva endleleni izithombe zesikhathi sangempela ezikhiqizwa futhi zifezwe ngayo.
