Indlela yokuthuthukisa izibalo zememori ye-Linux ngama-NVIDIA GPU

Informatec Digital » Izinsiza » Indlela yokuthuthukisa izibalo zememori ye-Linux ngama-NVIDIA GPU

Uma usebenza nama-NVIDIA GPU ku-Linux, kuzodingeka uqonde ukuthi kwenzekani ngenkumbulo yesistimu yakho, ukusebenza kwayo, kanye nezibalo zangaphakathi . Kungakhathaliseki ukuthi imidlalo, ukufunda komshini, ukulingisa kwesayensi, noma isikhungo sedatha, ukuqapha le datha kwenza umehluko phakathi komshini "osebenza" kanye neplatifomu elungiselelwe ngempela.

Ngesikhathi esifanayo, abasebenzisi abaningi beza ku-Linux belindele ukusebenza okungcono kunaku-Windows nge-NVIDIA GPU yabo , kodwa-ke bahlangabezana nezilinganiso ezingabonisi njalo imiphumela elindelekile. Isihluthulelo sisekwazini ukuthi ungaqapha kanjani, uhumushe, futhi ulungise kanjani kokubili uhlelo lokusebenza kanye nokucushwa kwe-GPU, usebenzisa amathuluzi athile afana ne -DCGM, i-dcgm-exporter, i-gpu-monitoring-tools , kanye nezixazululo zokubuka njenge-Elastic.

Ukusebenza kwe-NVIDIA GPU ku-Linux: okulindelwe kanye neqiniso

Uma umuntu onekhadi lehluzo lesimanje, njenge- RTX 4070 noma enye i-GPU ephezulu , ecabanga ukushintshela ku-Linux, kuvamile kakhulu ukuthi acwaninge kumaforamu, i-Reddit, amavidiyo e-YouTube, kanye nazo zonke izinhlobo zeziqondiso. Sekuyiminyaka eminingi, umqondo wokuthi i-Linux isebenza kangcono kune-Windows uye waphindwa, kodwa iqiniso, uma ubheka izilinganiso ezingathi sína kanye nokuhlolwa kokuqhathanisa , liyinkimbinkimbi kakhulu.

Emkhakheni wemidlalo, i-Linux ithuthuke kakhulu ngenxa ye -Steam, i-Proton, ne-DXVK , evumela ukuthi izihloko eziningi ze-Windows zisebenze kule nkundla. Kodwa-ke, lapho kubuyekezwa imiphumela emidlalweni edingayo, amazinga afanayo e-FPS njengakwi-Windows awatholakali njalo, ikakhulukazi lapho umdlalo ungenayo inguqulo yomdabu futhi uthembele ezingqimbeni zokuhambisana.

Ngaphandle kwemidlalo, ama-GPU e-NVIDIA aseyisisekelo sengqalasizinda esebenza kahle kakhulu : ukuqeqeshwa kwenethiwekhi ye-neural, ukulingisa okuyinkimbinkimbi kwefiziksi, ukuhumusha, kanye nemithwalo yemisebenzi emikhulu yesikhungo sedatha. Kulesi simo, ukuqapha ukusetshenziswa kwememori ye-Linux nokukhulisa ukusebenza kwe-NVIDIA GPU kubalulekile, futhi yilapho uhlelo oluhle lokuqapha lwenza khona umehluko omkhulu.

Kubalulekile futhi ukukhumbula ukuthi, ku-Linux, ukusebenza akuxhomekile kuphela ku-GPU uqobo, kodwa ekuhlanganisweni kwabashayeli be-NVIDIA, i-kernel, ukusatshalaliswa okukhethiwe , kanye nezilungiselelo zamandla newashi lekhadi. Uma noma yiziphi zalezi zinto zingalingani kahle, ama-benchmark angabonisa imiphumela edumazayo ngisho nangehadiwe enamandla kakhulu.

Kungani ukuqapha i-GPU kubaluleke kangaka ku-Linux?

Ama-GPU e-NVIDIA awasewona nje "amakhadi okudlala"; aseyisici esiyinhloko ezikhungweni zedatha, emafwini omphakathi, nasezindaweni ezisebenzisa kakhulu amakhompyutha . Ezindaweni eziningi zokufakwa, iningi lamandla ekhompyutha livela ku-GPU kune-CPU, ngakho-ke ukunganaki izilinganiso zokusebenza kwayo kumane kuyinto yokunethezeka amabhizinisi noma abasebenzisi bamandla abangenakukwazi ukuyikhokhela.

Kulezi zimo, akwanele ukwazi ukuthi unememori engakanani iyonke noma ukuthi iyiphi imodeli ye-GPU efakiwe. Kubalulekile ukukwazi ukubona ukusetshenziswa kwememori ngesikhathi sangempela, umthwalo we-GPU, izinga lokushisa, ukusetshenziswa kwamandla , isivinini sewashi, kanye nezinye izinto zokubala zangaphakathi ezinikeza izinkomba mayelana nezinkinga zokuzinza noma izinkinga zokuqina.

Ngaphezu kwalokho, i-Linux iyisistimu yokusebenza evelele kuma -computing clusters, i-HPC, kanye namapulatifomu e-cloud AI , ngakho-ke yonke ingqalasizinda yokuqapha ivame ukuyijikeleza. I-NVIDIA inikeza i-stack yayo yokuveza ama-metrics, kanye nezixazululo ezifana ne-Elastic Observability zivumela ukufakwa kwale datha phakathi, ukubonakala kwayo, kanye nokukhiqizwa kwezixwayiso ezihlakaniphile ngendlela eguquguqukayo kakhulu.

Uma usebenzisa ama-GPU e-NVIDIA nabahlinzeki abanjengo- Google Cloud, i-AWS, noma i-Genesis Cloud , ukuqapha akugcini nje ngokukusiza ukuthi uthole amaphutha; futhi kukusiza ukuthi uqinisekise ukuthi lokho okukhokhelayo efwini kusetshenziswa kahle, kugweme izindleko ezingadingekile ezivela ezimweni ezingasetshenziswa kahle noma ezingalungiselelwe kahle.

Elinye iphuzu elibalulekile ukuthi, ngokungafani nama-metric e-CPU, ama-metric amaningi e-GPU awahlanganiswa kahle namathuluzi ajwayelekile e-Linux. Yingakho izingxenye ezifana ne -NVIDIA Datacenter GPU Manager (DCGM) kanye nabathumeli be-metric abazinikele abahlanganiswa nezinhlelo zesimanje zokubuka ziwusizo kakhulu.

Ukuthembela okuyisisekelo: abashayeli be-NVIDIA, i-DCGM, namathuluzi okuqapha

Ukuze uthole izibalo zememori ye-Linux ezilungiselelwe kahle ku-NVIDIA, isinyathelo sokuqala ukuba nendawo elungiselelwe kahle: abashayeli abafanele, amathuluzi e-GPU, kanye nesisombululo sokubona idatha. Akusizi ngalutho ukuba namagrafu amahle uma uhlelo lungakwazi ukubonisa njalo izibalo.

Ezindaweni zesikhungo sedatha, i-NVIDIA inikeza i-Datacenter GPU Manager (DCGM) , esebenza njengesisekelo sokuqoqa ububanzi bezilinganiso, kusukela ekusetshenzisweni kwememori kanye nokushisa kuya kolwazi oluningiliziwe lokusebenza kwangaphakathi. Ukufakwa kuvame ukuncika kumaphakheji okusabalalisa okudumile njenge-Ubuntu 18.04 nezinye izinguqulo ezihambisanayo.

Ngesikhathi senqubo yokusetha echazwe yi-NVIDIA, kubalulekile ukunaka imininingwane efana nepharamitha ezindaweni zokugcina izimpahla ze-CUDAKumasistimu ajwayelekile angama-64-bit, umphumela womyalo uname -a kukhombisa ukuthi ukwakheka kwezakhiwo x86_64Ngakho-ke, imigqa engeziwe ohlelweni lwephakheji kumele isebenzise lelo nani ngendlela efanele.

Isibonelo esijwayelekile sokwengeza indawo yokugcina ye-CUDA ku-distro esekelwe ku-Debian/Ubuntu kungaba into efana nale:
echo "deb http://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64 /" | sudo tee /etc/apt/sources.list.d/cuda.list

Kunconywa futhi ukuhlola ukuthi imibhalo inayo yini iphutha lokubhala emiyalweni esemthethweni, njengokusebenzisa uphawu olwengeziwe > lapho kuchaza i-variable $distribution kulayini lapho okhiye be-GPG bangeniswa khona, okumele kulungiswe ngaphambi kokusebenzisa:

sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64/7fa2af80.pub

Uma abashayeli kanye ne-DCGM sezifakiwe, singahlola isimo esiyisisekelo se-GPU ngethuluzi elijwayelekile le-nvidia-smi , elibonisa imodeli, inkumbulo ephelele nesetshenzisiwe, izinga lokushisa, kanye nezinqubo ezisebenzisa ikhadi ngaleso sikhathi.

Ukufaka amathuluzi okuqapha i-gpu kanye ne-dcgm-exporter

Ukuze uhlanganise imemori ye-Linux kanye nezibalo ze-NVIDIA GPU nezinhlelo zangaphandle ezifana ne-Prometheus noma i-Elastic, i-NVIDIA inikeza i-gpu-monitoring-tools , ehlanganisa ingxenye ye-dcgm-exporter. Lezi zinsiza zisatshalaliswa njengekhodi yomthombo we-Go, ngakho-ke udinga ukuthi i-Go ifakwe futhi ilungiselelwe kahle ohlelweni lwakho.

Esimweni esivamile, i-Go iyalandwa futhi ifakiwe ku- /usr/localKhipha ifayela le-tar.gz bese wengeza i-Go binary ku-PATH yakho. Bese, bhala kabusha indawo yokugcina esemthethweni:
cd /tmp
git clone https://github.com/NVIDIA/gpu-monitoring-tools.git
cd gpu-monitoring-tools/
sudo env "PATH=$PATH:/usr/local/go/bin" make install

Ingxenye ebalulekile kule phakheji yile umthumeli-ngaphandle we-dcgmLe ngxenye iveza amamethrikhi e-DCGM ngefomethi uPrometheus angayiqonda. Uma iqalwa, iqala iseva ye-HTTP, ngokuzenzakalelayo, elalela ekhelini lendawo njenge localhost:9090 futhi inikeza zonke izilinganiso ezitholakalayo.

Ukucushwa kwezibalo okufanele ziboniswe kuchazwe kufayela elithi /etc/dcgm-exporter/default-counters.csv , eliqukethe izibalo eziningi ezichazwe ngaphambilini. Lokhu kufaka phakathi izibalo ezihlobene nememori ye-GPU, izinga lokushisa, ukusetshenziswa kwamandla, kanye nezinkomba zokusebenza zangaphakathi ezahlukahlukene . Ukuze uthole ukulawula okwengeziwe, imibhalo ye-DCGM Library API ichaza uhlu oluphelele lwezibalo ezingavulwa.

Uma uqala i-dcgm-exporter ngomyalo onjengalona:
dcgm-exporter --address localhost:9090
Uzobona imiyalezo yokuqalisa ekhombisa ukuthi i-DCGM iqalile ngempumelelo. Kuvamile ukuthola izexwayiso mayelana namamojula athile angalayishi, njengezilinganiso ze-DCP, okungenzeka kulesi simo ukunganakwa ngaphandle kwenkinga enkulu uma zingasetshenziswa.

Ukuhlanganiswa ne-Elastic Observability kusetshenziswa i-Metricbeat

Uma ama-metric e-NVIDIA GPU esevezwe nge-dcgm-exporter, isinyathelo esilandelayo esinengqondo ukuthumela leyo datha epulatifomu yokubona . Inketho enamandla yi-Elastic Observability, esebenzisa izingxenye ezifana ne-Metricbeat ukuqoqa ama-metric bese uwafaka phakathi ekusetshenzisweni kwe-Elastic Cloud noma isitaki esiziphethe.

Ukuze wenze lokhu, kuqala ifakiwe I-Metricbeat ohlelweni lwe-Linux lapho i-dcgm-exporter isebenza khona. Landa iphakheji ye-.deb ehambisana nenguqulo yakamuva, bese uyifaka nge dpkg -i futhi ifayela eliyinhloko lilungiselelwe /etc/metricbeat/metricbeat.yml ukuze kubhekisele ekusetshenzisweni kwe-Elastic Cloud kusetshenziswa amapharamitha cloud.id y cloud.auth.

Inani le- cloud.id Ngokuvamile inefomu elinekhodi elikhomba isifunda kanye nokufakwa kwe-Elastic, kuyilapho cloud.auth Hlanganisa igama lomsebenzisi nephasiwedi, isibonelo:
cloud.id: "staging:dXMtY2VudHJhbDEuZ2NwLmNsb3VkLmVzLmlvJDM4ODZkYmUwMWNjODQ2NDM4YjRlNzg5OWEyZDAwNGM5JDBiMTc0YzYyMTVlYTQwYWQ5M2NmMGY4MjVhNzJmOGRk"
cloud.auth: "elastic:J7KYiDku2wP7DFr62zV4zL4y"
(kusobala ukuthi, endaweni yangempela, kuzosetshenziswa iziqinisekiso eziphephile zomsebenzisi).

Okufakwayo kweMetricbeat kuyi-modular, futhi kulokhu sidinga ukuvumela Imodyuli ye-Prometheusngoba i-dcgm-exporter ishicilela amamethrikhi ayo ngaleyo fomethi. Ukusebenza kwenziwa nge:
sudo metricbeat modules enable prometheus

Ngaphambi kokusebenzisa konke, kuyalulekwa ukuthi usebenzise izivivinyo zokucushwa ukuqinisekisa ukuthi iMetricbeat ihlelwe kahle futhi ingaxhuma ku-Elastic Cloud kanye ne-Prometheus endpoint. Lokhu kwenziwa kusetshenziswa imiyalo efana nalokhu:
sudo metricbeat test config
sudo metricbeat test output
sudo metricbeat modules list

Uma lezi zivivinyo zingaphumeleli, kungcono ukubuyekeza imibhalo ye- Ukuxazulula izinkinga zeMetricbeatnjengoba amaphutha ngokuvamile ahlobene neziqinisekiso ezingalungile, ama-endpoint angafinyeleleki, noma amamojula angavulwanga kahle. Uma konke sekuqinisekisiwe, ungaqalisa:
sudo metricbeat setup
ukulayisha amadeshibhodi azenzakalelayo nokuchaza amamephu enkomba adingekayo.

Ekugcineni, iMetricbeat iqalwa kumodi yokukhipha yekhonsoli, noma njengesevisi, nge:
sudo metricbeat -e
Lokhu kuzokwenza ukuthi amamethrikhi e-GPU avezwe yi-dcgm-exporter aqale ukuthunyelwa njalo ku-Elastic, lapho azogcinwa khona futhi alungele ukuhlaziywa.

Ukubuka ngeso lengqondo inkumbulo kanye nezilinganiso ze-GPU eKibana

Uma idatha isingene ku-Elastic Observability, ingxenye ethakazelisayo iqala: ukuhlola, ukuhlunga, kanye nokubhekisela ku-GPU kanye ne-CPU ukuze uqonde ngempela ukuthi kwenzekani engqalasizinda yakho. Lokhu kwenziwa ngokuyinhloko kusetshenziswa i-Kibana, i-web-based analytics kanye ne-visual interface ye-Elastic.

Isinyathelo sokuqala ukuqinisekisa iphethini yenkomba metricbeat-* ilungiselelwe kahle esigabeni Ukuphathwa Kwezitaki > Kibana > Amaphethini EzinkombaUkusuka lapho, khetha iphethini bese uchofoza ku-"Refresh field list" ukuze uKibana akwazi ukubona amasimu amasha ahlobene nezilinganiso ze-GPU.

Amamethrikhi avela ku-dcgm-exporter, aqoqwe yi-module ye-Prometheus, ngokuvamile avela nesiqalo esifana ne -prometheus.metrics.DCGM_ . Lezi zinkambu zifaka izibalo zokusetshenziswa kwememori, imemori yamahhala, iphesenti lokusetshenziswa kwe-GPU, amazinga okushisa, ukusetshenziswa kwamandla, nezinye izinkomba ezifanele.

Njengoba lezi zinkambu sezitholakala manje, umbono we-Discover kaKibana ungasetshenziswa ukwenza usesho lwe-ad hoc , ukuhlunga nge-host, igama le-GPU, noma ibanga lesikhathi. Kungenzeka futhi ukwakha imiboniso esigabeni samadeshibhodi, ukukhombisa, isibonelo, ukuvela kokusetshenziswa kwememori ye-GPU usuku lonke noma ukuqhathanisa izimo ezahlukene kugrafu efanayo.

Enye yemibono ewusizo kakhulu yi-Metrics Explorer, ekuvumela ukuthi uqhathanise ukusebenza kwe-CPU ne-GPU esikrinini esifanayo . Lokhu kusiza ukuthola amaphethini lapho, isibonelo, i-GPU incishisiwe khona kuyilapho i-CPU ingasetshenziswa kahle, noma okuphambene nalokho, okubonisa izinkinga ezingaba khona zokulinganisela umthwalo womsebenzi.

Ngaphezu kwalokho, ukubuka kwe-Inventory kukuvumela ukuthi ukhombe amaphuzu abalulekile okusetshenziswa kwe-GPU ezindaweni ezinkulu zokusetshenziswa , ukhombe ukuthi yimaphi ama-node asemkhawulweni wawo, angasebenzi, nokuthi ama-bottlenecks agxilile kuphi. Ukusuka lapho, ungachaza izexwayiso eziqala lapho, isibonelo, inkumbulo ye-GPU idlula umkhawulo othile isikhathi esithile.

Amapharamitha okuqapha abalulekile ngokusho kwe-NVIDIA

Akuzona zonke izindlela zokulinganisa ezifaneleka ngokulinganayo uma kuthuthukiswa indawo ye-NVIDIA GPU ku-Linux. Ngokwezincomo zomenzi, kunezinye izindlela ezibalulekile okufanele ziqashwe njalo ukuze kuvinjelwe ukwehluleka, kutholakale ukwehla kokusebenza, futhi kulungiswe ukusetshenziswa kwamandla.

Izinga lokushisa le-GPU lingenye yezinkomba ezisobala kakhulu. Ukwanda okuqhubekayo ngaphezu kwebanga eliphephile kungabonisa izinkinga zokupholisa, amafeni angcolile, ukuhamba komoya okungekuhle ku-chassis, noma ngisho nomklamo wokushisa onganele umthwalo uhlelo olungaphansi kwawo.

Esinye isimiso esibalulekile ukusetshenziswa kwamandla e-GPU . Uma ikhadi liqala ukusebenzisa amandla amaningi kunokujwayelekile ngomthwalo womsebenzi ofanayo, kungabonisa izinkinga zehadiwe ezivelayo noma ukusethwa kwamandla okunamandla ngokweqile. Ezindaweni ezinkulu, le datha ibalulekile futhi ekulawuleni izindleko zikagesi nasekuhleleni umthamo wengqalasizinda.

Ijubane lamanje lewashi le-GPU liphinde lembule okuningi ngempilo yesistimu. Uma lawa mawashi ehlala engaphansi kwamanani alindelekile ngaphansi komthwalo osindayo, kungaba ngenxa yokulinganiselwa kwamandla, ukucindezelwa kokushisa, noma izilungiselelo zamandla ezilondolozayo ezibamba ukusebenza kwangempela okutholakalayo.

Ukuze kuhlolwe noma kuqinisekiswe ukucindezeleka, i-NVIDIA inikeza amathuluzi anjenge i-dcgmproftester10Le miyalo ikuvumela ukuthi ulingise imithwalo ye-GPU enamandla futhi uqinisekise ukuthi ama-metric aphendula kanjani. Umyalo ojwayelekile ungaba:
dcgmproftester10 --no-dcgm-validation -t 1004 -d 30
eqhuba ukuhlolwa kokusebenza okuthile imizuzwana engama-30 ngaphandle kokuqinisekiswa okuthile kwe-DCGM, okulungele ukuhlola ukuthi uhlelo luhlala luzinzile yini ngaphansi kwengcindezi.

Ngokuhlanganisa lezi zilinganiso namakhono okuxwayisa e-Elastic, kungenzeka ukwenza ngokuzenzakalelayo izincomo eziphakanyiswe yi-NVIDIA , kubangele izaziso lapho imingcele yokushisa idlula, lapho ukusetshenziswa kwememori kusondela ngo-100% ngendlela eyingozi, noma lapho ukusetshenziswa kwamandla kuphuma kakhulu ekulawuleni uma kuqhathaniswa nezilinganiso ezijwayelekile.

Ukuthuthukisa izilungiselelo ze-NVIDIA GPU ku-Linux

Ngaphandle kokuqapha okungenamsebenzi, kunezinguquko eziningana ezingasetshenziswa ku-Linux ukuze kufezwe ukusebenza okuzinzile, futhi ezimweni eziningi, okuphakeme ngama-NVIDIA GPU. Esinye sezici ezifanele kakhulu ukuphatha izici zokukhulisa okuzenzakalelayo kanye nokukala iwashi okuguquguqukayo okufakiwe kumshayeli uqobo.

Kwezinye izimo ze-GPU, umshayeli we-NVIDIA usebenzisa isici sokukhulisa okuzenzakalelayo esilungisa amaza e-GPU ngokusekelwe ezimweni zomthwalo kanye nokushisa. Nakuba lokhu kuwusizo ezindaweni ezijwayelekile, ezimweni ezidinga ukusebenza okuphezulu kanye nemiphumela ephindaphindwayo, kungcono ukukhubaza lezi zici bese usetha isivinini sewashi ngesandla kumanani aso aphezulu aphephile.

Ngokukhubaza ukukhulisa okuzenzakalelayo nokusetha imvamisa ephezulu, ukusebenza kuba okuvumelana kakhulu futhi okubikezelwayo phakathi kokugijima . Lokhu kubaluleke kakhulu ekulinganisweni kwezilinganiso, ukuqinisekiswa kwehadiwe, izikhungo zedatha, kanye namapulatifomu e-AI lapho umehluko omncane phakathi kokugijima kubalulekile khona.

Elinye iphuzu okufanele uliqaphele yimodi yamandla ye-GPU. Ku-Linux, amathuluzi e-NVIDIA akuvumela ukuthi ulungise amaphrofayili wamandla kusukela kumamodi asebenzisa amandla amaningi kuya kumamodi anikeza ukusebenza okuphezulu okuqhubekayo. Ukukhetha iphrofayili egcina kakhulu kunganciphisa kakhulu i-FPS emidlalweni noma ukusebenza emisebenzini yekhompyutha ebanzi , noma ngabe kunciphisa ukusetshenziswa kwamandla.

Kuma-PC abathengi, ikakhulukazi lawo aklanyelwe imidlalo ye-Linux enamakhadi ehluzo esimanje, kungumqondo omuhle ukuhlola ukuthi abashayeli basesikhathini nokuthi imodi yokuphikelela, umkhawulo wamandla, kanye nezilungiselelo zephrofayili yewashi kuhambisana nemigomo yokusebenza efiselekayo. Lena yindawo lapho imiphumela eminingi emibi yokulinganisa ibangelwa nje ukucushwa kwefektri okungalungiswanga kahle.

I-Linux nemidlalo enama-GPU e-NVIDIA: isimo samanje

Nakuba iLinux isahlushwa idumela lokungabi "yiplatifomu efanele" yokudlala imidlalo, iqiniso liwukuthi, nge -NVIDIA GPU enhle kanye ne-Steam ecosystem , kungenzeka ngokuphelele ukujabulela izihloko eziningi ngolwazi olufana kakhulu nolwe-Windows, ikakhulukazi uma usebenzisa ama-API ehluzo anamuhla.

I-Steam Play, isebenzisa i-Proton ne-DXVK (ifoloko ethile yewayini) , ihumusha izingcingo ze-DirectX zibe yi-Vulkan, okuvumela inani elikhulu lemidlalo eyaklanywa ekuqaleni ukuze i-Windows isebenze. Nakuba ukusebenza okuqondile kungatholakali ngaso sonke isikhathi, ukulahlekelwa kwe-FPS kuvame ukuba kuncane futhi, ezimweni eziningi, kuyamukeleka kubasebenzisi abaningi.

Uma kuqhathaniswa namakhadi esizukulwane esidlule, kusukela ku -NVIDIA GeForce GTX 980 Ti kuya ku-TITAN RTX , kanye namamodeli e-AMD afana ne-Radeon RX Vega 56 noma i-Radeon VII, kubonakale ukuthi ama-GPU aphezulu e-NVIDIA anama-Turing cores avelele ikakhulukazi ku- 4K resolution ku-Linux uma kukhulunywa ngemidlalo edinga amandla.

Isibonelo esibonakalayo singatholakala emidlalweni efana ne- Total War: Three Kingdoms , lapho amakhadi amaningi ehluzo angu-18 avivinyiwe angafinyelela i-60 FPS efiselekayo kusisombululo se-4K uma kugcinwa izinga lemininingwane ephansi. Uma imininingwane ikhuliswa ibe maphakathi, amamodeli anamandla kakhulu kuphela njenge- RTX 2080 Ti noma i-TITAN RTX akwazi ukudlula lowo mkhawulo ku-Linux njalo.

Ezinye izihloko ezifana ne -Counter-Strike: Global Offensive noma i-DOTA 2 zibonisa ukuziphatha okuthethelelayo kakhulu: ngokusebenzisa i-Vulkan API uma itholakala, cishe wonke amakhadi avivinyiwe angawasebenzisa kahle ku-4K, ngamazinga aphezulu e-FPS ngisho nakuhadiwe engathuthuki kakhulu ngaphakathi kobubanzi obuhlaziyiwe.

Emidlalweni edinga kakhulu imidwebo efana ne -Deus Ex: Mankind Divided noma i-Dawn of War III , isidingo samakhadi ehluzo aphezulu ukuze adlale ku-4K enezilungiselelo eziphezulu siphinde sicace. Ezihlokweni ezifana ne -Rise of the Tomb Raider noma i-A Total War Saga: Thrones of Britannia , ama-GPU amaningi angaphatha isixazululo se-4K uma izinga lemininingwane lisethwe phansi noma eliphakathi, okuqinisa umqondo wokuthi i -NVIDIA inenzuzo kune-AMD kubashayeli kanye nokwenza ngcono i-Linux kulezi zimo.

Isihloko esisodwa esiphindaphindayo kulezi zivivinyo ukuthi i-AMD isenendawo yokuthuthuka kubashayeli bayo bemidlalo ye-Linux , kuyilapho i-NVIDIA inikeza ukwesekwa okuvuthiwe kanye nezibuyekezo ezihambisanayo. Kulabo abafuna ukusebenza okungcono kakhulu emidlalweni enesinqumo esiphezulu ku-Linux, lokhu kusalokhu kuyisici esibalulekile lapho ukhetha i-GPU.

Uma sibheka ndawonye, ​​ukuqonda ukuthi izibalo zememori ye-Linux kanye nezilinganiso ze-NVIDIA GPU zisebenza kanjani, kuhlanganiswe namathuluzi okuhle okulungiselela umshayeli kanye nokuqapha, kukuvumela ukuthi uthole okuningi ekhadini, kungaba ku-PC yokudlala, iseva ye-AI, noma isikhungo sedatha esinama-node amaningi asebenza ngokugcwele.

Kusobala ukuthi ukulawula okuningiliziwe kwememori, ukusebenza, kanye nesimo sama-GPU e-NVIDIA ku-Linux akusadingeki, kodwa kuyingxenye ebalulekile kokubili ezindaweni zokukhiqiza kanye nabasebenzisi abafuna kakhulu: i-driver stack enhle kanye ne-DCGM, ungqimba lokuthumela ngaphandle lwe-metrics olune-dcgm-exporter, iqoqo eline-Metricbeat, kanye nokuboniswa ku-Kibana, kuhlanganiswe nokulungiswa kahle kwamandla e-GPU kanye nesivinini sewashi, yikona okwenza umehluko phakathi kohlelo olusebenza kalula kanye nolucindezela yonke i-watt kanye nayo yonke i-gigabyte yememori etholakalayo ngokugcwele.