Indlela yokufaka i-CUDA ku-Windows ngaphandle kwamaphutha: umhlahlandlela wabathuthukisi nabadali

Ukufaka i-CUDA ku-Windows Akumele kube yikhanda elibuhlungu uma wazi ukuthi uzoqala kuphi nokuthi yini okumele uyihlole esinyathelweni ngasinye. Kulesi sihloko ngizokuqondisa ngendlela engokoqobo, nawo wonke ama-nuances wokuhambisana, ukufakwa, ukuqinisekiswa kanye nokuxazulula izinkinga okuvamile ukuqinisekisa ukuthi ikhithi yamathuluzi isebenza kahle kukhompyutha yakho okokuqala ngqa.

Ngaphezu kokumboza ukufakwa kwekhithi yamathuluzi yakudala ku-Windows, uzobona futhi indlela yokusebenzisa i-CUDA nge-WSL, ukuyifaka nge-Conda noma nge-pip, hlanganisa izibonelo nge-Visual Studio, futhi uqonde amamodeli ahlukene abashayeli be-NVIDIA ku-Windows. Ulwazi luhlangene futhi luhambisana nesikhathi. Ngokusekelwe kumihlahlandlela esemthethweni nezimo zempilo yangempela ezingase zenzeke kuwe, njengekhompyutha ephathekayo ene-hybrid AMD iGPU + NVIDIA dGPU GPU.

Iyini i-CUDA futhi inikeza ini kuWindows?

CUDA Kuyinkundla yezinhlelo ezifanayo ze-NVIDIA kanye nemodeli evumelayo sheshisa izinhlelo zokusebenza nge-GPUUkusuka ku-AI nesayensi yedatha kuye ekulingiseni nasekucubunguleni izithombe. Ezingeni elingokoqobo, ukufaka i-CUDA Toolkit ku-Windows kukunikeza i-nvcc compiler, isikhathi sokusebenza, imitapo yolwazi efana ne-cuBLAS, cuFFT, cuRAND, ne-cuSOLVER, amathuluzi okulungisa iphutha nawokwenza iphrofayela, kanye nezibonelo ezilungele ukuhlanganisa.

Idizayini ye-CUDA yenza kube lula ukuhlanganisa i-CPU ne-GPU kuhlelo lokusebenza olufanayo: izingxenye ama-serial ku-processor kanye nezigaba ezihambisanayo ku-GPU, ezihlinzeka ngamakhulu noma izinkulungwane zemicu egijima ngokuhambisana. Ngenxa yenkumbulo ye-chip eyabiwe kanye namalabhulali athuthukisiwe, ukweqa ukusebenza Ngokuvamile kubonakala ngaphansi kwemithwalo enzima.

Ukuhambisana kwesistimu nokuhlanganisa ku-Windows

Ngaphambi kokusebenzisa isifaki, kuyatuseka ukuthi uhlole ukuhambisana. IWindows ehambisanayo Izinguqulo zakamuva zekhithi yamathuluzi zihlanganisa: Windows 11 24H2, 23H2 kanye 22H2-SV2; Windows 10 22H2; kanye ne-Windows Server 2022 kanye ne-2025.

Kubahlanganisi, ukwesekwa okujwayelekile kuhlanganisa I-MSVC 193x ene-Visual Studio 2022 17.x kanye ne-MSVC 192x ene-Visual Studio 2019 16.x, enezilimi zesigodi ezingu-C+++11, C++14, C++17, no-C++20 (kuye ngokuthi inguqulo). I-Visual Studio 2015 yehlisiwe ku-CUDA 11.1; I-VS 2017 yehlisiwe ngo-12.5 futhi yasuswa ngo-13.0. Hlola i-matrix enembile yenguqulo yakho ukugwema ukwethuka.

Okubalulekile kumaphrojekthi wefa: Kusukela ku-CUDA 12.0, ukuhlanganiswa kwe-32-bit kuyasuswa, futhi ukusetshenziswa kwamabhinari angama-32-bit x86 kumasistimu we-x64 kukhawulelwe umshayeli, i-quart nezibalo kuma-GeForce GPUs kuze kufike ekwakhiweni kwe-Ada; I-Hopper ayisasekeli amabhithi angu-32.

Khetha bese ufaka Ikhithi yamathuluzi ku-Windows

Landa isifaki kuwebhusayithi esemthethweni ye-NVIDIA CUDA. Ungakhetha Isifaki Senethiwekhi (ukulanda okuncane okusebenzisa i-inthanethi kuzo zonke ezinye) noma Isifaki Esigcwele (konke kuphakheji eyodwa, kuyasebenziseka imishini ngaphandle kwenethiwekhi noma ukuthunyelwa kwebhizinisi). Ngemva kokulanda, qinisekisa ubuqotho nge-checksum (isb., MD5) ukuze ukhiphe inkohlakalo.

Qalisa isifaki sesithombe bese ulandela izinyathelo ezikusikrini. Funda Amanothi Okukhishwa enguqulo yakho ngoba inemininingwane yezinguquko, ukuhambisana ngqo, nezixwayiso ezibucayi. Ukuqala nge-CUDA 13, isifaki sekhithi yamathuluzi asisamfaki umshayeli. Umshayeli we-NVIDIA ufakwe ngokuhlukile. kusuka ekhasini elihambisanayo labashayeli.

Ukufaka buthule nokukhethwa kwengxenye

Uma udinga ukuphakela buthule, isifaki samukela imodi ye-interface-less enenketho -s futhi ivumela khetha amaphakheji amancane athile ngegama esikhundleni sokufaka yonke into. Ungaphinda uvimbele ukuqalisa kabusha okuzenzakalelayo ngo-n. Lobu bumbudumbudu buwusizo ekwenzeni ngokwezifiso izindawo zokwakha kanye nokunciphisa umkhondo wakho.

Phakathi kwama-subpackages ajwayelekile uzothola izinto ezinjengokuthi nvcc, cudart, cuBLAS, cuFFT, cuRAND, cuSOLVER, cuSPARSEI-Nsight Compute, i-Nsight Systems, ukuhlanganiswa kwe-Visual Studio, i-NVRTC, i-NVTX, i-NVJitLink, ama-demanglers, nezinsiza ezifana ne-cuobjdump noma i-nvdisasm. Uma uzohlanganisa kanye nephrofayili, khetha amathuluzi e-NsightUma uyisebenzisa nje, isikhathi sokusebenza singase senele.

Khipha isifaki futhi ubuyekeze okuqukethwe

Ukuze kuhlolwe noma ukupakishwa kwezinkampani, isifaki esiphelele singakhishwa kusetshenziswa amathuluzi asekela i-LZMA njenge-7-Zip noma i-WinZip. Uzothola isihlahla se-CUDAToolkit namamojula Amafayela okuhlanganiswa kwe-Visual Studio abekwa kumafolda ahlukene. Amafayela we-.dll kanye ne-.nvi kulawo mafolda awayona ingxenye yokuqukethwe okufakekayo ngokwako.

Faka i-CUDA ku-Windows nge-Conda

Uma ukhetha ukuphatha indawo nge-Conda, i-NVIDIA ishicilela amaphakheji kokuthi anaconda.org/nvidia. Ukufakwa okuyisisekelo kwekhithi yamathuluzi Kwenziwa ngomyalo owodwa, `i-conda install`, futhi ungakwazi ukulungisa izinguqulo zangaphambilini ngokwengeza ithegi ethi `khulula`, isibonelo, ukukhiya enguqulweni 11.3.1. khipha Iqondile nje.

Faka i-CUDA ngepayipi (amasondo)

I-NVIDIA inikeza amasondo ePython agxile ku-CUDA isikhathi sokusebenza seWindows. Zihloselwe ngokuyinhloko usebenzisa i-CUDA ngePython futhi awabandakanyi amathuluzi okuthuthukisa aphelele. Okokuqala, faka i-nvidia-pyindex ukuze i-pip yazi inkomba ye-NVIDIA NGC, futhi uqiniseke ukuthi une-pip ne-setuptools ebuyekeziwe ukuze ugweme amaphutha. Bese ufaka ama-metapackage okudingayo, njenge-nvidia-cuda-runtime-cu12 noma i-nvidia-cublas-cu12.

Lawa ma-metapackage aqondise kumaphakheji athile afana ne-nvidia-cublas-cu129, nvidia-cuda-nvrtc-cu129, nvidia-npp-cu129, namanye. Khumbula ukuthi imvelo iphethwe yi-pip.Uma ufuna ukusebenzisa i-CUDA ngaphandle kwe-virtualenv, uzodinga ukulungisa izindlela zesistimu nokuguquguqukayo ukuze uxhume kahle.

Qinisekisa ukufakwa ku-Windows

Vula ukwaziswa komyalo bese usebenzisa i-nvcc -V ukuze uqinisekise inguqulo efakiwe. Hlanganisa amasampula e-CUDA Landa izibonelo ku-GitHub futhi uzihlanganise ne-Visual Studio. Qalisa idivayisiQuery kanye ne-bandwidthTest: uma kukhona ukuxhumana okuyimpumelelo ne-GPU, uzobona idivayisi itholiwe futhi ukuphumelela izivivinyo Awekho amaphutha. Uma idivayisiQuery ingawatholi amadivaysi, hlola umshayeli nokuthi i-GPU iyabonakala ohlelweni.

I-WSL ene-CUDA ukusheshisa

Windows 11 kanye nezinguqulo zakamuva ze Windows 10 ukwesekwa esebenzisa izinhlaka ze-ML ezisheshiswe nge-CUDA namathuluzi ngaphakathi kwe-WSL, okuhlanganisa I-PyTorch, i-TensorFlow ne-Docker Usebenzisa i-NVIDIA Container Toolkit, qala ufake umshayeli onikwe amandla i-CUDA ku-WSL, bese unika amandla i-WSL futhi ufake ukusabalalisa kwe-glibc okufana no-Ubuntu noma i-Debian.

Qiniseka ukuthi une-WSL kernel ebuyekeziwe (ubuncane obungu-5.10.43.3). Kuhlole nge Sebenzisa `wsl cat /proc/version` kusuka ku-PowerShell. Bese ulandela inkomba yomsebenzisi ye-CUDA ku-WSL ukuze ufake imitapo yolwazi neziqukathi bese uqala ukusebenzisa ukugeleza komsebenzi wakho we-Linux ku-Windows ngaphandle kokushiya indawo yakho.

Khipha i-CUDA ku-Windows

Ngemva kokufaka i-CUDA ku-Windows, ingabe ufuna ukubuyela enguqulweni yangaphambilini? Wonke amaphakheji amancane angabuyiselwa emuva. Khipha ku-Control Panel Ukusebenzisa Izinhlelo Nezici. Uma uphatha ikhithi yamathuluzi nge-Conda noma i-pip, sebenzisa izindlela zokukhipha zomphathi ngamunye ukuze ugweme ukushiya noma yiziphi izinsalela zephakheji.

Amanothi ahambisana nenguqulo

I-CUDA 11.8 ibingukukhululwa okudume kakhulu ngenxa yokuzinza nokusekelwa kwe-ecosystem. Izidingo ezijwayelekile Ku-11.8: I-GPU Enekhono Lekhompyutha 3.0 noma ngaphezulu, i-64-bit, ubuncane obungu-8 GB we-RAM kanye okungenani nememori engu-4 GB ye-GPU. Ku-Linux, ihlanganisa kahle nokusabalalisa okufana no-Ubuntu 18.04/20.04, RHEL/CentOS 7/8, njll.

I-CUDA 12.x yethula ukuthuthukiswa kwesikhathi sokusebenza kanye nelabhulali futhi iphushe ukuncika kwe abashayeli bakamuvaI-CUDA 13 ihlukanisa unomphela umshayeli kusifaki se-Toolkit: khumbula ukufaka umshayeli ngokwakho. Ukucaciswa okubalulekileI-CUDA ubuchwepheshe be-NVIDIA futhi idinga ama-NVIDIA GPU; uma ubona noma yikuphi ukuthi iyahambisana nama-AMD GPU, lokho akulungile kusitaki se-CUDA.

Ukufaka i-CUDA ku-Windows: Ukuxazulula Izinkinga Ezivamile

• Isifaki siyehluleka noma asiwuqedi umsebenzi.Hlola amalogi wesifaki futhi uqinisekise isivikeli magciwane sakho, isikhala sediski, nezimvume zokuphatha. Zama futhi Ngesifaki Esigcwele uma inethiwekhi ingazinzile, noma ikumodi ethulile uma kunokungqubuzana kwe-UI.
• deviceQuery ayiboni i-GPUHlola ukuthi umshayeli ulungile, ukuthi i-GPU iyasebenza, nokuthi uhlelo lokusebenza lisebenzisa i-dGPU. Buyekeza umshayeli bese ufaka kabusha Ikhithi Yamathuluzi uma kunesidingo.
• Ukungqubuzana nezitolo zezincwadiUma unamathuluzi amaningi afakiwe, qinisekisa i-CUDA_PATH kanye ne-PATH. Ku-Python, hlola ukuthi izinguqulo ze-PyTorch noma ze-TensorFlow kanye nokucushwa kwazo kuyahambisana nenguqulo yakho ye-CUDA/cuDNN.
• I-Visual Studio ayihlanganisi i-.cuEngeza i-CUDA Yakha Ukwenza Ngokwezifiso kuphrojekthi yakho futhi umake amafayela .cu njenge-CUDA C/C++. Qinisekisa ukuthi i-MSVC iyasebenzisana nekhithi yakho yamathuluzi.

Amathuluzi, amasampula kanye nemibhalo

Ngokungeziwe ku-nvcc nemitapo yolwazi, i-Toolkit yokufaka i-CUDA ku-Windows ihlanganisa amaphrofayili nabahlaziyi njenge-Nsight Systems ne-Nsight Compute, kanye nemibhalo ye-HTML/PDF yolimi lwe-CUDA C++ kanye imikhuba engconoIzibonelo ezisemthethweni ziku-GitHub futhi ziyisisekelo esihle kakhulu sokuqinisekisa abashayeli, ukusebenza kwenkumbulo, nama-multiprocessors.

Isetshenziswa nini i-Conda noma ipayipi uma iqhathaniswa nesifaki sakudala

I-Conda ne-pip zilungile uma ugxile ekusebenziseni izinhlaka ze-ML esezivele zincike kumaphakheji ngokuhambisana nezinguqulo ezithile ze-CUDA. InzuzoUkuhlukaniswa kwemvelo kanye nokungqubuzana okuncane. Ububi: Ekuthuthukisweni komdabu kwe-C++ noma ukuhlanganiswa okugcwele ne-VS, isifaki sakudala samathuluzi esinikezwayo wonke amathuluzi kanye nolwazi oluphelele kakhulu.

I-FAQ Esheshayo

• Ngazi kanjani uma i-GPU yami ihambisana ne-CUDA? Vula Isiphathi Sedivayisi, hamba kokuthi Bonisa ama-adaptha, bese uhlola imodeli; iqhathanise nohlu olusemthethweni lwe-NVIDIA lwama-CUDA GPU. Ungaphinda usebenzise i-nvidia-smi futhi uqinisekise lokho I-GPU yakho iyavela.
• Ngingakwazi ukuqeqesha ngaphandle kwe-CUDA? Yebo, izosebenza ku-CPU, kodwa izohamba kancane. Ukuze usebenzise i-GPU nge-PyTorch noma i-TensorFlow ku-Windows, qiniseka ukuthi uyayifaka izakhiwo ezihambisanayo ngenguqulo yakho ye-CUDA noma sebenzisa i-WSL eneziqukathi ze-NVIDIA.
• Izinguqulo ezindala ezithileAmanye amathuluzi adinga izinhlanganisela ezifana ne-CUDA 10.1 ne-cuDNN 7.6.4. Uma kunjalo, faka lezo zinguqulo eziqondile bese ubeka ifayela I-DLL ye-cuDNN kufolda yomgqomo wekhithi yamathuluzi ehambisanayo, ukugwema ukuba nama-cuDNN amaningi ngesikhathi esisodwa.

Uma ubheke ukufaka i-CUDA ku-Windows futhi usheshise umsebenzi wakho ngomhlahlandlela ophelele, izinyathelo nezincomo ezingenhla zizokusiza ukuthi wenze konke. Ilingana njengegilavu. kusukela ekuqaleni.