I-GPU Passthrough (vDGA) ku-VMware: Umhlahlandlela Ophelele

Ukuxhuma i-GPU ebonakalayo ngqo emshinini obonakalayo ku-VMware Kungenye yalezo zinguquko ezenza umehluko omkhulu lapho usebenza ngemithwalo yezithombe esindayo. IA noma ukwenziwa kwe-3D. Ukushintsha kusuka ekhadini lehluzo elilingisiwe uye ekufinyeleleni okuqondile nge-passthrough (vDGA / VMDirectPath I/O) kungasondeza ukusebenza kwe-VM komshini ongokoqobo, kodwa ngokubuyisela kunezela izidingo eziningi kanye nemikhawulo okufanele icace kakhulu ngaphambi kokuqala.

Ngaphezu kwalokho, ohlelweni lwamanje lwe-ecosystem, kunezindlela eziningana zokusebenzisa i-GPU ezindaweni ezisebenza nge-virtualized: I-passthrough enikezelwe, i-vGPU eyabiwe, kanye nobuchwepheshe obufana ne-BitFusion noma i-GPU partitioningUkuqonda ukuthi ngayinye yenzani, ukuthi ifanelana kanjani nokuthi ihlelwe kanjani ku-vSphere/ESXi (nokuthi ihlobene kanjani nobuchwepheshe obufanayo njenge-Hyper-V DDA) kubalulekile ekungangeni emgomweni ophelele ngehadiwe noma inguqulo ye-hypervisor ekhethiwe.

Kuyini i-GPU passthrough (vDGA / VMDirectPath I/O) ku-VMware

I-GPU passthrough ku-VMware, eyaziwa nangokuthi i-vDGA noma i-VMDirectPath I/OLena indlela yokusebenza lapho ikhadi lehluzo elingokoqobo elifakwe ku-host ye-ESXi linikezwa ngqo emshinini obonakalayo. Esikhundleni sokusebenzisa i-adaptha yehluzo elingiswe yi-hypervisor, uhlelo lokusebenza lwezivakashi lubona i-GPU sengathi ixhunywe kubhodi lomama elingokoqobo.

Lesi sinqamuleli sivumela i-VM ukuthi isize wonke amandla e-chip yehluzo, inkumbulo yayo yevidiyo kanye nezici ezithuthukisiwe njenge-CUDA, i-OpenCL, i-Direct3D, kanye ne-OpenGL, ngaphandle kwezindleko ezengeziwe ezivela ku-hypervisor. Ekuhlolweni kwelebhu ye-VMware, ukulahlekelwa kokusebenza okucishe kube ngu-4-5% kuvame ukubikwa uma kuqhathaniswa nokusebenzisa i-GPU efanayo ku-bare metal.

Empeleni, ukusebenzisa i-GPU passthrough kusho ukuthi Lelo khadi linikezelwe ngokuphelele kumshini owodwa obonakalayo.Akukho ukwabiwa kwezinsizakusebenza okuhle phakathi kwama-VM amaningi, futhi akukho ungqimba lwesofthiwe lomuntu wesithathu olufakwe ku-ESXi ukuze kwabelwane ngekhadi lehluzo, ngokungafani nalokho okwenzekayo ngezixazululo ze-vGPU ezifana I-NVIDIA I-GRID.

Kubalulekile ukuhlukanisa le ndlela kwezinye izindlela zokusebenzisa ama-GPU ku-virtualization, njenge I-NVIDIA vGPU (i-vGPU eyabiwe), i-RemoteFX/i-partitioning kuzixazululo ze-Hyper-V noma ze-BitFusion, ezifuna ukusabalalisa i-GPU noma iqoqo lama-GPU phakathi kwemishini eminingi enezindlela ezahlukene ze-virtualization noma zokuqondisa kabusha okude.

Uma sikhuluma nge-vDGA ezweni le-VMware, empeleni sichaza lokhu ukunikezwa okuqondile kwedivayisi ye-PCIe ye-GPU ku-VM kusetshenziswa i-VMDirectPath I/O, nazo zonke izinto ezinhle (ukusebenza) nezimbi (ukuhamba kanye nemikhawulo ephezulu yokutholakala) okuhambisana nalokhu.

Izinzuzo zokusebenzisa i-GPU passthrough ku-vSphere

Isizathu esiyinhloko sokushintshela ku-vDGA ukuthi Ihluzo kanye nokusebenza kwekhompyutha kufana kakhulu nokwekhompyutha ebonakalayo.Ngokushiya ingxenye enkulu yesendlalelo se-virtualization saleyo divayisi ye-PCIe, izithiyo ezivamile ze-GPU elingisiwe ziyanyamalala futhi i-VM ingasebenza nemidlalo, izinhlelo zokusebenza ze-3D noma izinjini ze-AI ngokushelela kakhulu.

Lokhu kubonakala kakhulu ezimweni lapho ama-GPU ahlanganisiwe noma ikhadi lehluzo elibonakalayo elilingisiwe elizenzakalelayo lihluleka khona: Umklamo wezithombe othuthukisiwe, i-CAD, ukumodela nokudweba kwe-3D, ukuhlela ividiyo, ukugqwayiza, kanye nokuthuthukiswa kwemidlaloKubalulekile futhi ekuqeqeshweni kwamamodeli okufunda komshini kanye nemithwalo yemisebenzi ye-AI encike kakhulu ku-CUDA noma okulinganayo.

Enye inzuzo ecacile ukusetshenziswa kwehadiwe okuguquguqukayo ezingeni lesikhungo sedatha. Esikhundleni sokuba ne- indawo yokusebenza ebonakalayo ngomsebenzisi ngamunye noma ngephrojekthi ngayinyeKungenzeka ukunikezela i-host ye-ESXi enkulu kuma-VM amaningana, ngayinye ine-GPU yayo ekudluleni, bese udlala ngamashejuli noma isidingo esiphezulu.

Ezindaweni ezithile, ikakhulukazi uma amaseva anezikhala ze-PCIe zamahhala esevele etholakala, izindleko ngomsebenzisi ngamunye noma ngephrojekthi ngayinye Kungase kube ngaphansi kokugcina inqwaba yezindawo zokusebenza ezinamandla, ikakhulukazi uma ikhadi lehluzo lingadingeki amahora angama-24 ngosuku futhi lingalungiswa kabusha ngokwezikhathi zomsebenzi onzima.

Okokugcina, kukhona futhi inzuzo engaqondile maqondana nokuphepha nokusebenza: ngokugcina umthwalo wemisebenzi yezithombe ngaphakathi kwemishini ebonakalayo ehlukanisiweUma kukhona okungahambi kahle (ukuxhashazwa, umshayeli onenkinga, ukucushwa okubi) kulula ukuvimbela umthelela, ukubuyela esithombeni sangaphambilini noma ukubuyisela kusuka kubhekhaphu, inqobo nje uma imikhawulo ye-passthrough, esizoyibona kamuva, ihlonishwa.

Ukudlula kwe-GPU uma kuqhathaniswa ne-vGPU nezinye izindlela

Ngaphakathi kwe-VMware ecosystem, kunezindlela eziningana zokusebenzisa ikhadi lehluzo, futhi akuzona zonke ezihilela ukulinikezela ngokuphelele kwi-VM eyodwa. Ezaziwa kakhulu yilezi: i-vDGA / VMDirectPath I/O, i-vGPU (i-NVIDIA GRID noma ezinye) kanye nezixazululo zokufinyelela/ikhompyutha ezikude njenge-BitFusion.

Kumodi yokudlula ngqo (i-vDGA), i-GPU yabelwe kumshini obonakalayo kuphelaAma-compute cores kanye ne-VRAM awabiwa phakathi kwama-VM amaningi, futhi indima ye-hypervisor cishe ayikho ngaphandle kokuhambisa idivayisi ye-PCIe esivakashini. Yindlela elula ukuyiqonda futhi leyo efana kakhulu neseva ebonakalayo enekhadi lehluzo elizinikele.

Ngendlela ye-vGPU, isofthiwe ekhethekile (isibonelo I-NVIDIA GRID vGPU ku-VMware vSphereIphatha i-GPU virtualization ezingeni lomlawuli futhi iveza izimo ezibonakalayo ze-GPU ezingabelwa ama-VM amaningi ngasikhathi sinye. Umshini ngamunye wezivakashi ubona "ucezu" lwe-GPU olunezinsiza eziqinisekisiwe noma ezabiwe.

Ama-vGPU avumela amadeskithophu amaningi noma amaseva ukuthi abelane ngekhadi elilodwa lehluzo, okuwusizo kakhulu ku I-VDI, izindawo zehhovisi ezisheshayo ezilula, izindawo zokusebenza ezibonisa imidwebo ezitolo noma ezinakekela izivakashinoma izimo lapho ukusetshenziswa kwehluzo eziphakeme kungalingani kubasebenzisi bonke. Ngenxa yalokho, kukhona okunye okungaphezulu, futhi ukusebenza okuphezulu okufanayo akufezwa njengase-GPU ephelele enikezelwe ku-VM eyodwa.

Kukhona nezixazululo ezifana nalezi: I-BitFusion Flexdirect kanye nobuchwepheshe obufanayookuvumela ama-GPU ukuthi asetshenziswe kunethiwekhi evela kuma-VM ahlukene, okulungele imithwalo yemisebenzi ye-AI kanye ne-HPC lapho i-GPU isebenza kakhulu njengomthombo wekhompyutha okude kunekhadi levidiyo lesikhombikubona sesithombe somsebenzisi.

Khetha phakathi kwe-vDGA, i-vGPU, noma imodeli ye-GPU ekude Kuya ngokuthi udinga ukusebenzisa i-GPU ngokugcwele yini emshinini owodwa (i-passthrough), ukuthi ufuna ukusabalalisa ikhadi elibizayo phakathi kwabasebenzisi abaningi abanemithwalo yemisebenzi ephakathi nendawo (i-vGPU), noma ukuthi isihluthulelo siwukuhlela iqembu lama-GPU ekhompyutha esabalalisiwe (i-BitFusion nezinye ezifanayo).

Izidingo zehadiwe zokusebenzisa i-vDGA ku-ESXi

Ngaphambi kokuhlela ukuthunyelwa kwe-GPU passthrough ku-VMware, udinga ukuqinisekisa ukuthi Ipulatifomu yehadiwe ihlangabezana nochungechunge lwezimo lokho kudlula "ukuba nekhadi lehluzo elixhunywe kuseva".

Okokuqala, iprosesa kanye ne-chipset yebhodi lomama le-ESXi kumele zisekele ukwenza i-virtualization nge-IOMMU. e Intel Lokhu kufezwa nge-Intel VT-x kanye ne-VT-dy, kanye nange-AMD nge-AMD-V ene-IOMMU. I-BIOS/UEFI yeseva ivame ukuba nezinketho ezithile zalokhu. sebenzisa izandiso ze-virtualization we-I/O.

Okwesibili, udinga ukuhlola ukuthi ipuleti lokusekela Ukumepha imemori ye-MMIO ngaphezu kwe-4 GB (ngezinye izikhathi kubhalwe ngokuthi “Ngaphezulu kwe-4G decoding”, “imemori efakwe imephu ye-I/O ngaphezu kwe-4G” noma okufanayo). Lokhu kubaluleke kakhulu ngama-GPU aphezulu njenge Tesla, P100, V100 kanye nezinto ezifanayo, ezimemezela izindawo zememori ezinkulu kakhulu ku-BAR yazo (Amarejista Ekheli Eliyisisekelo).

Amanye ala makhadi aphezulu angawabala isikhala se-MMIO esingaphezu kuka-16 GBNgakho-ke, ngaphezu kokudlala i- I-BIOSNgemuva kwalokho, amapharamitha athile kuzilungiselelo ze-VM ezithuthukisiwe ku-vSphere kuzodingeka alungiswe ukuze akwazi ukuqala ngaleyo GPU ngaphandle kwamaphutha ezinsiza anganele.

Vele, i-GPU ngokwayo kumele ibe iyahambisana neplatifomu yeseva futhi isekelwe umenzi we-host (Dell, HPE, Lenovonjll.) uma isetshenziswa kwimodi yokudlula. Empeleni, ama-PCIe GPU amaningi esimanje ayasebenza, kodwa kuyalulekwa ukuhlola uhlu lokuhambisana, ikakhulukazi kumakhadi e-GRID noma amamodeli amasha kakhulu.

Izidingo zesofthiwe kanye nokuhambisana kwenguqulo

Ezingeni lesofthiwe, kubalulekile ukucaca ukuthi I-VMware iyayisekela i-vDGA ku-vSphere 6.x kanye nezinguqulo zakamuvaKodwa-ke, abanye abasebenzisi babike izinkinga ezithile ngezinhlanganisela ezithile zehadiwe (isibonelo, ama-GPU e-NVIDIA GRID kumaseva e-Dell R720 ane-ESXi 6.x).

Kulezi zimo, kuvamile ukubona amaphutha anjengokuthi “idivayisi isivele isetshenziswa” noma izimpawu ezibonisa ukuthi I-Passthrough iyeke ukusebenza ngemuva kokuthuthukisa kusuka ku-ESXi 5.5 kuya ku-6.xkanti empeleni kumayelana namaphutha athile, izinguquko ekuphathweni kwedivayisi ye-PCI noma abashayeli, kunokuba kuhoxiswe ngokusemthethweni ukwesekwa.

Uhlelo lokusebenza lwezivakashi oluzosebenzisa i-GPU ekudluleni kumele lube nalo abashayeli abasemthethweni bomkhiqizi bafakiwe ngaphakathi kwe-VM (NVIDIA, AMD, Intel), njengoba i-ESXi ingalayishi noma yimuphi umshayeli othize walelo khadi uma usebenzisa i-VMDirectPath I/O; i-hypervisor imane iveze idivayisi esivakashini.

Ngaphezu kwalokho, i-VM kumele ilungiselelwe ukuqala kabusha Imodi ye-EFI noma ye-UEFI uma usebenzisa ama-GPU amemezela izifunda zememori ezinkulu ze-MMIO. Le mininingwane ibalulekile: i-firmware ye-VM engalungile ingaholela ekuhlulekeni ibhuthini noma ukuthi i-GPU ayiqalisi kahle kusukela ohlelweni lokusebenza lwezivakashi.

Ngasohlangothini lweklayenti, uma ukufinyelela ku-VM kwenzeka nge Ideskithophu ekude (i-RDP noma ezinye izinqubo)Izinqubomgomo ezifanele kuzodingeka zivulwe ukuze uhlelo lwezivakashi lusebenzise i-adaptha yehluzo zehadiwe kumaseshini akude futhi lungabambeki kumshayeli ojwayelekile ngaphandle kokusheshisa.

Ukulungiselela i-host ye-ESXi ukuze isebenzise i-GPU kumodi yokudlula

Isinyathelo sokuqala esiwusizo ukulungiselela iseva ye-vSphere/ESXi ukuze iveze i-GPU njengedivayisi ye-DirectPath I/OLokhu kuhilela ukufinyelela i-BIOS, ukuhlola i-PCI inventory ku-host, nokumaka ikhadi ukuze likwazi ukwabelwa ama-VM.

Uma i-GPU idinga izindawo ezinkulu zememori ye-MMIO (16 GB noma ngaphezulu), kufanele ubheke ku-BIOS/UEFI yeseva ukuze uthole izinketho ezifana nalezi “Ukubhala ikhodi okungaphezu kwe-4G” noma “Ukuphathwa kwezinsiza ze-PCI 64bit ngaphezu kwe-4G” bese uzisebenzisa. Igama elithile liyahlukahluka kuye ngomkhiqizi, kodwa livame ukutholakala esigabeni sokucushwa kwe-PCI noma izinsiza ezithuthukisiwe.

Uma i-ESXi isiqalisiwe ngalezi zilungiselelo, kuklayenti le-vSphere ungaya kumphathi ohambisanayo bese ufinyelela “Lungiselela → Ihadiwe → Amadivayisi e-PCI → Hlela” Ukuze ubuke uhlu lwamadivayisi e-PCI atholakele, uzobona amakhadi e-NVIDIA, AMD, noma afanayo kanye nezinye izingxenyekazi ze-PCI zeseva.

Uma i-GPU ingakasebenzi ku-DirectPath I/O, mane uthikhe ibhokisi. ibhokisi lokungena emnyango wakho ngaphakathi kwalolo hlu. Uma ulondoloza izinguquko, i-vSphere izokucela ukuthi uqale kabusha i-host ukuze usebenzise ukucushwa, njengoba i-hypervisor idinga ukubhukha nokulungiselela idivayisi ukuthi inikezwe kabusha kuma-VM.

Ngemva kokuqala kabusha, lapho ubuyela esigabeni “Lungiselela → Ihadiwe → Amadivayisi e-PCI” Kuzovela iwindi elinesihloko esithi “DirectPath I/O PCI Devices Available to VMs”, elibala wonke amadivayisi asetholakalela ukusetshenziswa emishinini ebonakalayo, kufaka phakathi ama-GPU, futhi ezimweni eziningi, ama-adapter enethiwekhi athuthukile njengeMellanox.

Ukulungiselela nokulungiselela umshini obonakalayo

Njengoba i-host isilungile, isinyathelo esilandelayo ukudala noma ukulungisa umshini obonakalayo ozosebenzisa i-GPU. Into yokuqala ukuqinisekisa ukuthi i-VM Yakhiwe nge-firmware efanelekile ye-EFI/UEFI., ikakhulukazi ezimweni ezinama-GPU aphezulu kanye ne-MMIO ephezulu.

Kuklayenti le-vSphere, khetha nje i-VM, iya ku- “Hlela Izilungiselelo → Izinketho ze-VM → Izinketho Zokuqalisa” futhi uqinisekise ukuthi i-"EFI" noma i-"UEFI" ikhethiwe ensimini ethi "Firmware". Uma kungenjalo, kuzodingeka ishintshwe (futhi kwezinye izimo, i-VM noma uhlelo lokusebenza kuzodingeka luphinde lwenziwe uma lungasekeli lokhu kushintshana okushisa).

Uma usebenzisa i-passthrough ngamakhadi abonisa isikhala se-MMIO esingaphezu kuka-16 GB, kungcono ukulungisa amanye amapharamitha athuthukile ekucushweni kwe-VM, atholakala kusuka ku- “Hlela Izilungiselelo → Izinketho ze-VM → Okuthuthukisiwe → Amapharamitha Okucushwa → Hlela Ukucushwa”Lapho ungangeza okhiye abahlobene ne-pciPassthru ukulawula ukuthi isikhala sekheli sigcinwa kanjani.

Ngokuqondile, ukusetshenziswa kwe-64-bit MMIO kuvame ukuvunyelwa futhi kuchazwa usayizi waleso sifunda, kubalwa kusukela ku- Mangaki ama-GPU aphezulu azokwabelwa i-VM?Umthetho oyinhloko uvame ukuphindaphinda u-16 ngenani lama-GPU bese uzungeza umphumela uye emandleni alandelayo amabili (isibonelo, ama-GPU amabili anjalo angagcina eno-64 GB we-64-bit MMIO).

Ngemva kokulungisa la mapharamitha, ukufakwa kuyenziwa noma kuqinisekiswa ukuthi Uhlelo lokusebenza lwezivakashi lusekela i-EFI/UEFI futhi luyakwazi ukuphatha usayizi wememori kanye ne-GPU okukhulunywa ngayo.Kuleli qophelo, ikhadi lehluzo alikaxhunywa ku-VM; indawo ilungiswa nje ukuze, uma isixhunyiwe, konke kuqale ngaphandle kwamaphutha ngenxa yokuntuleka kwezinsiza noma i-firmware engahambisani.

Nika i-GPU ku-VM usebenzisa i-VMDirectPath I/O

Uma i-host isimake i-GPU njengetholakalayo kwi-DirectPath I/O futhi i-VM isilungiselelwe kahle, sekuyisikhathi sokuthi hlanganisa ikhadi ngqo nalowo mshini obonakalayoLesi sinyathelo kumele senziwe i-VM icishiwe ngokuphelele.

Kusuka kuklayenti le-vSphere, khetha i-VM bese ufaka “Hlela Izilungiselelo” ukuze ubukeze ihadiwe ebonakalayoOhlwini lwamadivayisi, ungachofoza okuthi "Engeza Idivayisi Entsha" bese ukhetha okuthi "Idivayisi ye-PCI" uma i-GPU ingakafakwa ohlwini. Bese ukhetha idivayisi ye-PCI ehambisana nekhadi lehluzo (isibonelo, ikhadi le-NVIDIA noma le-AMD elitholwe kumphathi).

Uma ukucushwa kulondolozwa, i-VM izobonisa into efana nale kwihadiwe yayo “Idivayisi ye-PCI 0” ehlotshaniswa ne-GPU ethileKusukela kuleli phuzu kuqhubeke, lapho uhlelo lokusebenza lwezivakashi luqala, luzobona i-adaptha eyengeziwe ye-PCIe ehambisana nekhadi lehluzo elingokoqobo.

Kubalulekile ukuthi umshini obonakalayo ube igcine yonke imemori eyabelwe yonaKu-vSphere, lokhu kulungiselelwe ku-“Hlela Izilungiselelo → Ihadiwe Ebonakalayo → Imemori”, kusetha inani “Lokubhukha” enanini le-RAM elilungiselelwe i-VM. Ngaphandle kwalokhu kubhukha okugcwele, i-PCI passthrough ingase yehluleke noma ibhekane nezinkinga ngezikhathi ezithile.

Ngemva kokuvula i-VM, kusistimu Linux Ukuba khona kwe-GPU kungaqinisekiswa nge imiyalo unomusa lspci | grep nvidia, ngenkathi ku-Windows kuzovela ngaphansi kokuthi “Ama-adaptha okubonisa” ku- Umlawuli wedivayisiKujwayelekile ukubona i-adaptha yehluzo elingisiwe ye-VMware kanye ne-GPU ebonakalayo ezinikele.

Isinyathelo sokugcina ukufaka okulandelayo kudivayisi yesimenywa: abashayeli Izikhulu zomkhiqizi we-GPU, okulandiwe kumawebhusayithi e-NVIDIA, AMD noma e-Intel, kugwema ukuthembela kubashayeli abajwayelekile noma labo abanikeziwe I-Windows Updateokungenzeka ukuthi ayilungiselelwe izimo zokudlula.

Imikhawulo ye-vSphere nezici ezingasebenzi ne-vDGA

Uhlangothi B lwe-GPU passthrough ku-VMware ukuthi Izici eziningana ezithuthukisiwe zepulatifomu ziyalahleka ngokunikela idivayisi ebonakalayo ngqo ku-VM. Leyo yintengo okufanele uyikhokhele ukusebenza okuseduze nendawo.

Umhlatshelo wokuqala omkhulu i-vMotion kanye ne-DRSUmshini obonakalayo one-GPU kumodi yokudlula awukwazi ukuthuthelwa komunye umsingathi ngoba ikhadi likhiyiwe ngokomzimba kuseva yokuqala. Izinqubomgomo zokulinganisela umthwalo ezizenzakalelayo ezihilela ukuhambisa i-VM phakathi kwama-host ku-cluster nazo azikwazi ukusetshenziswa.

Izici ezifana nokuthi Izithombe zendabuko noma izindlela ezithile zokutholakala okuphezulu ngaleyo VM ethile. Ngenxa yokuthi incike kuhadiwe ethile ebonakalayo, ikhono lokuqandisa nokubuyisela izimo eziyinkimbinkimbi liyaphazamiseka.

Esinye isici okufanele sicatshangelwe ukuthi, kule modi, I-GPU ayabelwana ngayo phakathi kwama-VM amaningiUma kudingeka amadeskithophu amaningi noma amaseva ane-graphics ascent emshinini ofanayo, kuzodingeka ikhadi elilodwa nge-VM ngayinye, noma ngenye indlela, imodeli ye-vGPU ingasetshenziswa lapho ikhadi lenziwa nge-virtual ezimweni eziningi.

Ngasohlangothini lokusekela, kungase kube nezimo ezithile lapho Ukuhlanganiswa okuthile kwehadiwe kanye nabashayeli kungabangela izinkingaNjengoba abanye abasebenzisi bebonile lapho bethuthukela ku-ESXi 6.x ngamakhadi e-NVIDIA GRID kumaseva athile (isb., i-Dell R720), kuyalulekwa ukuthi ubuyekeze amadokhumenti omkhiqizi we-VMware kanye ne-GPU kulezi zimo, futhi uvule amacala okusekela uma kudingeka.

Okokugcina, kufanele kuqashelwe ukuthi ubuchwepheshe noma izinsizakalo ezithile ezisebenzisana nezithombe, njenge amadeskithophu akude, izinhlelo ezingaphansi kwe-Linux ku-Windows, noma izici zesistimu yokusebenza ezithuthukisiweBangaphazamisa noma babangele amaphutha "e-Code 43" kubashayeli be-NVIDIA uma bethola ukuthi usebenza ngaphakathi kwe-VM ene-GPU passthrough.

Ukudlula kwe-GPU kwamanye ama-hypervisor: kuhambisana ne-Hyper-V

Nakuba ukugxila lapha ku-VMware, kufanelekile ukuqonda ukuthi kanjani amanye ama-hypervisor (isibonelo ukwenza i-virtualization nge-KVM kanye ne-virt-manager) ukubhekana nesidingo esifanayo sokwabela i-GPU ebonakalayo ku-VMngoba amagama namathuluzi ayashintsha, kodwa umqondo oyisisekelo uyafana.

Ku-Hyper-V, okulingana ne-VMware VMDirectPath I/O yi- ukwabiwa kwedivayisi okuqondile kusetshenziswa i-DDA (Discrete Device Assignment)Le ndlela ivumela ukumaka idivayisi ethile ye-PCIe, njenge-GPU noma i-NVMe, ngqo ngaphakathi komshini we-Windows virtual, enezinga lokulawula nokusebenza okufana nokuphumelela ku-ESXi.

Izinguqulo ezindala ze-Windows Server zisebenzise lobu buchwepheshe I-RemoteFX ukunikeza i-GPU virtualization nokwabelana ngekhadi lehluzo phakathi kwama-VM amaningi. el tiempoNgenxa yezinkinga zokuphepha kanye nemikhawulo yokusebenza (njenge-1GB VRAM nge-VM cap kanye ne-30 FPS), iMicrosoft yathatha umhlalaphansi i-RemoteFX futhi yashiya i-DDA njengendlela eyinhloko yezimo ze-GPU ezizinikele.

Ku-Windows 10 futhi Windows 11Ikakhulukazi ngamaqoqo athile, ukwesekwa kuye kwavela Ukuhlukanisa i-GPU kanye nezindlela ezisetshenziswe kabusha ezivela ku-WSL2 kanye ne-Windows SandboxKodwa-ke, ukucushwa kwayo kuvame ukuhilela izikripthi eziyinkimbinkimbi nokukopisha abashayeli kusuka kumphathi kuya kusivakashi, okungelula njengokwabela idivayisi ku-vSphere.

Ukwazi lezi zindlela kusenza sibone ukuthi Ifilosofi yokunikeza ukufinyelela okuseduze nendawo kwi-GPU ngesiteshi se-PCIe esiqondile Kuvamile kuma-hypervisor amaningana, yize ngalinye linama-nuances alo, imiyalo, kanye nemikhawulo yokuhambisana.

Lonke lolu hlelo lokusebenza lwe-passthrough, i-vGPU, kanye ne-DDA lubonisa ukuthi, uma luhlelwe kahle futhi lunehadiwe efanele, Kungcono kakhulu ukusebenzisa ama-GPU anamandla ngaphakathi imishini ebonakalayo yokukhiqiza Ngemithwalo yemisebenzi esukela kuma-desktop ehluzo adingayo kuya ku-AI kanye ne-HPC, njalo ucabanga ukuthi kuzodingeka ulahle ezinye izinto ezilula ze-virtualization yendabuko futhi unake kakhulu abashayeli, izinguqulo ze-hypervisor kanye nokusekelwa komkhiqizi we-GPU.

I-athikili ehlobene:

Ungavumela kanjani i-virtualization kusuka ku-BIOS noma i-UEFI isinyathelo ngesinyathelo

Isaka

Umbhali oshisekayo ngomhlaba wamabhayithi nobuchwepheshe ngokujwayelekile. Ngiyathanda ukwabelana ngolwazi lwami ngokubhala, futhi yilokho engizokwenza kule bhulogi, ngikubonise zonke izinto ezithakazelisayo kakhulu ngamagajethi, isofthiwe, ihadiwe, izitayela zobuchwepheshe, nokuningi. Inhloso yami ukukusiza ukuthi uzulazule emhlabeni wedijithali ngendlela elula nejabulisayo.