I-GPU Engcono Kakhulu Yokufunda Komshini

Ezweni lanamuhla eliqhutshwa idatha, ukufunda komshini kanye nokufunda okujulile kuye kwaba yizingxenye ezibalulekile zokusungula izinto ezintsha zanamuhla - kusukela ekucubungulweni kolimi lwemvelo (i-NLP) kuya embonweni wekhompyutha kanye nezinhlelo ezizimele. Enhliziyweni yalezi zindlela eziyinkimbinkimbi kukhona ingxenye ebalulekile: i-GPU (iyunithi yokucubungula ihluzo). Ngenkathi ama-CPU enza izibalo ezijwayelekile, ama-GPU asheshisa ukuqeqeshwa kwamamodeli okufunda komshini ngokwenza imisebenzi eyizinkulungwane ngesikhathi esisodwa.

Ukukhetha i-GPU engcono kakhulu yokufunda komshini akuseyona indaba ebaluleke kakhulu kubacwaningi kuphela. Kuthinta:

• Ukufakwa kwe-AI yebhizinisi (isb., ukuqagela okukhulu kwemodeli)
• Izinkampani ezintsha ze-AI ezihlose ukuthuthukisa amapayipi okuqeqesha
• Ososayensi bedatha kanye nonjiniyela be-ML: Ukwakha amamodeli okuhlola
• Abacwaningi bezemfundo bandisa imingcele yobuhlakani bokwenziwa
• Abathandi bezinto zokuzilibazisa kanye nabathandi belebhu yasekhaya bacwaninga amanethiwekhi ajulile e-neural

Ukukhetha i-GPU efanele yokufunda komshini kuhilela okungaphezu nje kokuhlola amashadi okusebenza. Ukwakheka, umthamo wememori, ukuhambisana nezinhlaka ezifana ne-TensorFlow noma i-PyTorch, kanye nokusekelwa kwezici ezifana ne-ANDERS noma i-ROCm konke kunegalelo ekunqumeni ukusebenza kwe-GPU ye-AI kanye nemithwalo yemisebenzi yokufunda okujulile.

Imininingwane ebalulekile

Izakhiwo ze-GPU zanamuhla

• I-NVIDIA Ampere (A100, RTX 3090): Yaziwa ngomklamo wayo oqinile we-Tensor Core kanye nezici ze-MICH
• I-NVIDIA Hopper (H100, H200): Ingeza ukwesekwa kwe-RP8 futhi ithuthukisa i-bandwidth nge-watt ngayinye.
• I-Blackwell (B100, B200): Ukwakhiwa kwesizukulwane esilandelayo se-NVIDIA kuthembisa ukuthuthuka okukhulu ekusebenzeni kwekhompyutha ye-AI
• I-AMD CDNA (MI300X): Incintisana ne-NVIDIA ngokunikeza inkumbulo ye-bandwidth ephezulu (i-HBM3) kanye nokuhambisana kwe-ROCm.

Ukuhambisana kwesistimu yokusebenza kwesofthiwe

I-GPU inhle kuphela njengohlelo lwayo lwe-ecosystem. Ama-GPU e-NVIDIA abusa ngamalabhulali avuthiwe e-ANDERS kanye ne-cuDNN, kuyilapho i-AMD iqhubeka nokuthuthukisa ukwesekwa kwe-ROCm kwamathuluzi omthombo ovulekile.

Ukuhambisana nezinhlaka ezivamile ze-ML ezifana nalezi:

• I-TensorFlow
• I-PyTorch
• I-JAX
• Isikhathi sokusebenza se-ONNX

kuqinisekisa ukuhlanganiswa okungenamthungo kanye nokusetshenziswa okuphezulu kwemisebenzi ye-GPU ngesikhathi sokuqeqeshwa kwemodeli kanye nokucabanga.

Ngamafuphi, i-GPU engcono kakhulu yokufunda okujulile kufanele ilinganisele amandla ekhompyutha eluhlaza, ukwakheka kwememori, kanye nokusekelwa kwesofthiwe, okwenza ifaneleke emisebenzini efana ne-NLP, umbono wekhompyutha, noma ukufunda kokuqinisa kuwo wonke amazinga ahlukahlukene abasebenzisi.

Imakethe ye-GPU ngo-2025 izohlinzeka ngezinketho ezahlukahlukene ezenzelwe imisebenzi ehlukene yokufunda komshini - kusukela ekuqeqesheni amamodeli olimi amakhulu kuya ekuqondeni kwe-AI ngesikhathi sangempela. Ama-GPU alandelayo avelele ngenxa yokwakheka kwawo, amandla okukhumbula, kanye namakhono okwenza ngcono i-AI, okwenza kube ukukhetha okuphezulu kososayensi bedatha, abacwaningi be-AI, kanye nokuthunyelwa kwamabhizinisi.

1. I-NVIDIA H100 / H200 (ukwakheka kwe-Hopper)

Lawa ma-GPU ayindinganiso yegolide yokuqeqeshwa kwamamodeli amakhulu, ngokunemba kwe-FP8, inkumbulo ye-HBM3 engu-80–141 GB kanye nokusekelwa kwama-GPU amaningi (i-MIG). Alungele ama-LLM, ikhompyutha yesayensi, kanye namaqoqo okuqeqeshwa kwama-GPU amaningi.

2. I-NVIDIA A100 (Ukwakhiwa kwe-Ampere)

Isetshenziswa kabanzi kumapulatifomu e-cloud GPU, i-A100 inikeza ibhalansi phakathi kwezindleko, ukusebenza, kanye nokutholakala. Njengoba inememori ye-HBM2e efika ku-80 GB, ifanelekela ukuqeqesha amanethiwekhi ajulile ezinzwa, amamodeli okubona ikhompyutha, kanye nemisebenzi ye-NLP.

3. Isizukulwane se-NVIDIA L40S / RTX 6000 Ada

Njengoba ziqondiswe ezindaweni zokusebenza ze-AI futhi zinikeza imodeli yebhizinisi, lawa ma-GPU asebenzisa ukwakheka kwe-Ada Lovelace ukuze kusebenze kahle ukusebenza kwe-inference, ukulandelela imisebe kanye nokubala okusebenzisa amandla ngendlela eyongayo.

4. I-NVIDIA RTX 4090/3090 Ti

Lawa ama-GPU angcono kakhulu abathengi bonjiniyela be-ML nabacwaningi abadinga ukusebenza okuphezulu ngaphandle kwamanani ebhizinisi. Ngememori ye-GDDR6X engu-24GB, asekela iningi lamafreyimu e-ML, kufaka phakathi i-TensorFlow ne-PyTorch, futhi enza kahle emisebenzini efana nokuhlukaniswa kwezithombe, ukuqeqeshwa kwe-GAN, kanye nokulungiswa kahle kwemodeli ye-NLP.

5. I-AMD Instinct MI300X / MI250

I-MI300X ye-AMD inikeza imemori ye-HBM3 engu-128 GB, ukwakheka kwe-CDNA 3, futhi isekela i-ROCm yezinhlaka zokufunda komshini ezivulekile. Ingumncintiswano onamandla ezindaweni zocwaningo ze-HPC kanye ne-AI ezidinga i-bandwidth enkulu yememori.

I I-SIN-3042-H110 ihlinzeka ngezinhlelo zokuthutha ezisebenza kahle kakhulu kanye nokuhlunga amaphasela:

• Ukufinyelela idivayisi ye-Multiprotocol: Njengoba inamachweba e-USB ayi-6, ingaxhumanisa izikena zebhakhodi, izikali ze-elekthronikhi, kanye nezinzwa. Uma ihlanganiswe ne-Intel Gigabit Ethernet, ivumela ukutholwa kwedatha ngesikhathi sangempela kanye nokulayisha.
• Isitoreji esiguquguqukayo: Ukusekelwa okubili kwe-HDD/SSD okungu-2.5-intshi kanye nokukhipha okubili kokubonisa (VGA + HDMI) kwenza kube lula ukuqapha uhlelo kanye nokukhipha ividiyo.
• Ukusekelwa kohlelo lwe-AGV: Ngesikhala se-Mini-PCIe, idivayisi ingahlanganiswa nezinhlelo zokuhlela ze-AGV, ithuthukise isivinini sokuhlunga kanye nokusebenza kahle kokuzenzakalela ezindaweni zokugcina izinto ezihlakaniphile.

Uma uhlola i-GPU engcono kakhulu yokufunda komshini, kubalulekile ukudlulela ngale kwezincazelo ezibalulekile nokuqonda ukuthi i-GPU ngayinye, emisebenzini ehlukene ye-AI, osayizi bamamodeli, kanye nezimo zokufakwa, izici ezifana ne-bandwidth yememori, umthamo we-VRAM, kanye nokwakheka okuyinhloko kuthinta ngqo ikhono layo lokusebenzisa kahle amamodeli okufunda okujulile ayinkimbinkimbi.

Izilinganiso zokuqhathanisa ezibalulekile

Sebenzisa ukufaniswa kwezinhlamvu

• I-NVIDIA H100/H200: Yakhelwe amamodeli amakhulu olimi, ukuqeqeshwa kwamamodeli ayisisekelo, kanye nokucabanga kwe-multi-GPU. Ilungele amalebhu ocwaningo kanye nabahlinzeki bengqalasizinda ye-AI.
• I-NVIDIA A100: Ukukhetha okuguquguqukayo kwezinhlaka zokufunda okujulile njengeTensorFlow neJAX, ikakhulukazi ezimweni ze-cloud GPU.
• I-RTX 4090/3090 Ti: Kuhle kakhulu konjiniyela be-ML ngabanye futhi kunikeza ukusebenza okuphezulu kokwenza amamodeli, ama-GAN, kanye nokuqagela kwesikhathi sangempela.
• I-AMD MI300X: Ngememori enkulu ye-HBM3, iphatha osayizi abakhulu be-batch kanye nokucutshungulwa kwezithombe ezinesinqumo esiphezulu, ifanelekela imithwalo yemisebenzi yesayensi ye-ML.

Okunye okucatshangelwayo

• I-MIG kanye ne-NVLink zibalulekile ekwabiweni kwe-GPU kwabaqashi abaningi kanye ne-bandwidth ye-inter-GPU kumaqoqo ebhizinisi.
• Ukusatshalaliswa kwamandla okushisa (i-TDP) kanye nokuhambisana kokunikezwa kwamandla kufanele kucatshangelwe lapho kwakhiwa izindawo zokusebenza ze-AI zasendaweni.
• Ukusekelwa kwe-Software stack (isb., i-CUDA vs. i-ROCm) kunquma ukuhambisana kohlaka.

Kafushane: I-GPU efanele kumele ifane nosayizi wemodeli yakho, isikhathi sokuqeqeshwa, ipayipi ledatha, kanye nendawo yokufakwa.

Ukukhetha i-GPU engcono kakhulu yokufunda komshini kuncike kakhulu esimweni sakho sokusetshenziswa, isabelomali, kanye nezidingo zobuchwepheshe. Kungakhathaliseki ukuthi ungumuntu osafufusa, isikhungo socwaningo, noma unjiniyela ozimele, ukufanisa amandla e-GPU nomsebenzi wakho womsebenzi kuqinisekisa ukusebenza okuhle kakhulu kanye nokubuyiselwa kwemali ekutshalweni kwezimali.

Kwezinkampani kanye nama-laboratory ocwaningo

Ama-GPU anconyiwe:

• I-NVIDIA H100 / H200
• I-AMD Instinct MI300X
• I-NVIDIA A100

Kungani:

Lawa ma-GPU anikeza ukucubungula okuhamba phambili okumangalisayo, inkumbulo ye-bandwidth ephezulu (i-HBM3), kanye nokwehluka kwe-multi-GPU (nge-NVLink, i-MICH, noma i-PCIe Gen5). Alungele:

• Ukuqeqeshwa kwamamodeli ezilimi ezinkulu (ama-LLM)
• Amapayipi e-AI akhiqizayo
• Iqoqo le-GPU labasebenzisi abaningi
• Ikhompyutha yesayensi ethuthukisiwe

Kwamabhizinisi amasha kanye nokuthuthukiswa kwe-AI ebangeni eliphakathi

Ama-GPU anconyiwe:

• Isizukulwane se-NVIDIA RTX 6000 Ada
• I-NVIDIA L40S
• I-NVIDIA A100 (isibonelo samafu)

Kungani:

Lawa ma-GPU anikeza ukusebenza okufanayo kanye nentengo efanayo. Anikeza amandla e-Tensor computing aqinile, i-VRAM enkulu (kufika ku-48-96 GB) kanye nokuhambisana nezinhlaka ezidumile ezifana ne-TensorFlow, i-PyTorch, kanye ne-ONNX runtime.

Kwabathuthukisi ngabanye kanye nabantu abathanda ukuzilibazisa

Ama-GPU anconyiwe:

• I-NVIDIA RTX 4090/3090 Ti
• I-RTX 4070 / 4080 (evumelana nesabelomali)

Kungani:

Lawa ma-GPU abathengi anikeza ukusebenza okuhle kakhulu kwe-FP32/FP16, i-VRAM eyanele (24 GB), kanye nokusekelwa okuqinile kwe-CUDA ngentengo engabizi kakhulu. Kuphelele ku:

• Imodeli yokulinganisa
• Ukuqeqeshwa kwe-GAN kanye ne-CNN
• Ukulungiswa kwe-NLP
• Ukuhlolwa kwe-AI ekhaya

Uma usebenzisa izindlela zokufunda komshini, esinye sezinqumo ezibaluleke kakhulu zengqalasizinda ukuthi uzosebenzisa ama-GPU asekelwe efwini noma utshale imali endaweni yokusebenza ye-GPU yendawo. Indlela ngayinye inikeza izinzuzo ezahlukene kanye nokushintshana, kuye ngokuthi imodeli yakho ye-AI iyinkimbinkimbi kangakanani, isabelomali, kanye nosayizi weqembu.

Umhlinzeki:

• Izinsizakalo Zewebhu ze-Amazon (AWS)
• Ipulatifomu Yamafu ka-Google (i-GCP)
• I-Microsoft Azure
• AmaLabda Labs, izicubu eziyinhloko, umkhakha wephepha

Izibonelo ezidumile:

• I-NVIDIA A100 / H100 / L40S
• I-AMD MI300X (evelayo)

Izinzuzo:

• Ukukwazi ukukhula: Ukukala okulula kuma-GPU amaningi ukuze kuqeqeshwe amamodeli amakhulu
• Ukuzivumelanisa nezimo: Qasha ama-GPU afunwayo ngaphandle kwezindleko zehadiwe kusengaphambili.
• Ukufinyelela komhlaba wonke: Amaqembu angasebenzisana kuzo zonke izifunda.

Imikhawulo:

• Izindleko zesikhathi eside: Amamodeli akhokhelwayo njengoba usebenza angabiza kakhulu ngokuhamba kwesikhathi.
• Ukubambezeleka: Kuphakeme kakhulu ekucabangeni kwesikhathi sangempela
• Ukuphepha kwedatha: Idatha ebucayi kumele ilayishwe kumaseva ezinkampani zangaphandle.

Izindawo zokusebenza ze-GPU zasendaweni

Ihadiwe eyabiwe:

I-NVIDIA RTX 4090, i-RTX 6000 iyatholakala, i-3090 Ti iyatholakala

Isiteshi sokusebenza sakhiwe nge-AMD Threadripper noma i-Intel Xeon

Izinzuzo:

• Izindleko zesikhathi esisodwa: Kuyonga kakhulu ekusetshenzisweni kwesikhathi eside
• Ukulawula okugcwele: Phatha ukwakheka kokushisa, ukuthuthukiswa, kanye nenkumbulo.
• Ukuvikelwa kwedatha: Gcina amasethi edatha namamodeli endlini.

Imikhawulo:

• Ukutshalwa kwezimali kusengaphambili: Izindleko eziphezulu zokuthenga ihadiwe kanye nokunikezwa kukagesi
• Ukukhuliswa okulinganiselwe: Kunzima kakhulu ukufinyelela umthamo ofanayo wefu.
• Ukuguga kwehadiwe: Ukuphelelwa yisikhathi okusheshayo emakethe ye-GPU esheshayo

Khetha inketho efanele

Ekugcineni, ukukhetha kwakho kuzoncika kusayizi wemodeli, imvamisa yokusetshenziswa, ukuphathwa kwedatha, kanye nezindleko zokusebenza eziphelele.

Ngaphambi kokutshala imali ku-GPU engcono kakhulu yokufunda komshini, kubalulekile ukuhlola ukuthi ihadiwe ihambisana kahle kangakanani nezidingo zakho zokumodela, i-software stack, kanye nemigomo yokukhulisa isikhathi esizayo. Ukukhetha nje i-GPU enamandla kakhulu akuqinisekisi ukusebenza kahle noma ukusebenza kahle kwezindleko—ikakhulukazi uma ingahambisani nedatha yakho noma indawo yokuthuthukiswa.

1. Ukuhambisana kwesofthiwe

• I-CUDA vs. i-ROCm: Ama-GPU e-NVIDIA asekela i-ANDERS, i-cuDNN, kanye ne-NCCL - asetshenziswa kabanzi ku-TensorFlow, i-PyTorch, kanye ne-JAX. Ama-GPU e-AMD, yize ethuthukiswa ngaphezu kwe-ROCm, asantula ukuhambisana okugcwele nawo wonke amalabhulali okufunda okujulile.
• Ukusekelwa kohlaka: Qinisekisa ukuthi amafreyimu akho e-ML alungiselelwe i-GPU ekhethiwe. Ezinye izici ezintsha (njenge-FP8 precision noma i-GPU Multi-Instance (MIG)) zitholakala kuphela kuzakhiwo ezintsha ze-NVIDIA Hopper kanye ne-Blackwell.

2. I-VRAM kanye nosayizi wemodeli

Amamodeli amakhulu okufunda okujulile (isb., ama-LLM, ama-transformer, ama-GAN) adinga inkumbulo ye-GPU eyengeziwe. Bamba:

• Ifanele amamodeli ayisisekelo e-ML, ama-CNN amancane, i-prototyping
• 24–48 GB: Ilungele ukuqeqesha amanethiwekhi ayinkimbinkimbi anosayizi abakhulu
• 80 GB+ (HBM3): Kuyadingeka ekuqeqeshweni okukhulu, i-AI yezindlela eziningi, noma ukubala kwesayensi

3. Ukuhlanganiswa kwezinhlelo kanye nengqalasizinda

• Izidingo zokupholisa kanye nokunikezwa kwamandla: Ama-GPU aphezulu njenge-RTX 4090 noma i-H100 adinga ugesi oqinile (kufika ku-600 W) kanye nokupholisa okuthuthukisiwe.
• Ukusekelwa kwemizila ye-PCIe kanye nebhodi lomama: Qinisekisa ukuthi uhlelo lwakho lungayisebenzisa ngokugcwele i-PCIe Gen4/Gen5 ukuze kube nomkhawulokudonsa ophezulu.
• Ukusethwa kwe-NVLink/Multi-GPU: Uma uhlela ukukala, khetha i-GPU esekela ukuxhumana kanye nokufinyelela imemori okwabelwana ngayo.

4. Indlela yokuqina nokuthuthukisa

Cabanga ngomjikelezo wokuphila we-GPU kanye neshejuli yokusekela. Ukutshalwa kwezimali ezakhiweni zamanje ezifana ne-Ada Lovelace, i-Funnel, noma i-CDNA 3 kunikeza ukubaluleka kwesikhathi eside njengoba imithwalo yemisebenzi yokufunda komshini iba nzima kakhulu.

Ukukhetha i-GPU efanele kudinga ubudlelwano obulinganiselayo phakathi kokusebenza, ukuhambisana, kanye nokulungela ingqalasizinda - hhayi idatha eluhlaza kuphela.

Isimo se-GPU sokufunda komshini sishintsha ngokushesha, siqhutshwa yizidingo ezikhulayo ezivela kumamodeli amakhulu olimi (ama-LLM), i-AI enqenqemeni, kanye namapulatifomu e-AI-as-a-Service. Njengoba ubunzima kanye nobukhulu bezinhlelo zokusebenza ze-AI bukhula, abakhiqizi behadiwe bacindezela imingcele ekwakhiweni kwe-GPU, ukwakheka kwememori, kanye nobuchwepheshe bokusheshisa i-AI.

1. Ukwakhiwa kwe-NVIDIA Blackwell (B100/B200)

Ngemva kwempumelelo ye-Funnel (H100/H200), ama-NVIDIA's Blackwell GPU akulungele ukuchaza kabusha ukusebenza kokufunda okujulile. Intuthuko ebalulekile ifaka:

• Ukuthuthukiswa kokuphuma kwe-FP8/FP4 tensor core
• Phinda kabili umkhawulokudonsa wesitoreji nge-hopper
• Usekelo lwe-Greater NVLink 5.0 lokuxhumana kwe-multi-GPU
• Ukusebenza kahle kwamandla okusebenzisa i-AI

Lawa ma-GPU aklanyelwe ukuhlungwa kahle kwe-LLM, ukuqeqeshwa kwe-AI yama-node amaningi, kanye nokubala kwe-exascale.

2. Ukwandiswa kwe-AMD: i-CDNA 3 nangaphezulu

I-MI300X ye-AMD, eyakhelwe phezu kwesakhiwo se-CDNA 3, imele intuthuko ebalulekile futhi inikeza:

• Imemori ye-HBM3 engu-128 GB
• Umkhawulokudonsa wesitoreji ongu-5.2 TB/s
• Ukusekelwa kwemvelo kwezinhlaka ze-ROCm kanye ne-ML zomthombo ovulekile

Njengoba kwanda ukwamukelwa kwabasebenzi abasebenzisa ubuchwepheshe obuphezulu kanye nezikhungo zesayensi, i-AMD iziveza njengomncintisani wangempela ekubalweni kwe-AI.

3. Ukwanda kwezisheshisi ze-AI ezenziwe ngokwezifiso

Ngaphandle kwama-GPU endabuko, izinkampani zitshala imali kuma-accelerator aqondene nesizinda se-AI:

• I-Google TPU v5e/v6
• Ama-chip e-AWS Trainium kanye ne-Inferentia
• Injini Yesikali Se-Cerebras Wafer
• Ama-NPU e-Groq kanye ne-Tensorrent

Lokhu kulungiselelwe imisebenzi ethile, njenge-transformer inference, i-video processing, kanye nama-graph neural networks, okunikeza i-throughput ephezulu ngokusetshenziswa kwamandla okuphansi.

4. I-AI emaphethelweni

Lindela ukukhula kuma-GPU anamandla aphansi aklanyelwe ukuqagela okungaphezulu kuma-robotics, i-IoT, kanye nezinhlelo zokucubungula izithombe ngesikhathi sangempela. I-Jetson Music, i-Intel Havana, kanye ne-NVIDIA IGX ziyizibonelo ezihamba phambili.

Ukukhetha i-GPU efanele yokufunda komshini kuncike ezintweni eziningana - ubunzima bemodeli, isabelomali, ubude bomsebenzi, nokuthi ngabe usebenzisa izindawo zamafu noma ezisendaweni. Lokhu kubhekisela okusheshayo kuklanyelwe ukukusiza ukuthi wenze lula inqubo yakho yokwenza izinqumo ngokusekelwe esimweni sakho esithile sokusetshenziswa.

Isinyathelo 1: Chaza umthwalo wakho womsebenzi

Isinyathelo 2: Linganisela izidingo zesitoreji

Kufanelekela ukuqeqeshwa kwasekuqaleni noma isiphetho

• 16–24 GB: Iphatha ama-CNN ajwayelekile, ama-GAN, kanye nemisebenzi yokulungisa kahle
• 48GB+ / HBM3: Kuyadingeka ku-AI ye-multimodal, ukuqeqeshwa kweqembu elikhulu, noma ividiyo enesinqumo esiphezulu

Isinyathelo 3: Lungisa ingqalasizinda

• Abasebenzisi bokuqala kwefu: Khetha ama-instance e-H100 , L40S , noma e-MI300X nge-AWS, GCP, noma i-Azure.
• Abakhi bezindawo zokusebenza zasendaweni: Khetha i-RTX 4090, 6000, noma i-A100 PCIe.
• Abasebenzisi be-hybrid: Sebenzisa i-GPU yendawo ukuze uthuthukise futhi ulinganisele amafu ukuze uthole imfundo.

Isinyathelo 4: Cabanga ngesabelomali kanye nokusebenza

Ngokuvumelanisa imininingwane yehadiwe, ukwesekwa kwesofthiwe, kanye nokonga izindleko, lo mhlahlandlela wokuthatha izinqumo usiza ukuthola i-GPU engcono kakhulu yokufunda okujulile ehambisana nezidingo zakho zobuchwepheshe nezokusebenza - kungakhathaliseki ukuthi usebenzisa i-PC yezimboni ene-GPU, usebenzisa ikhompyutha ye-AI, wenza ngcono ikhompyutha enqenqemeni yezimboni, ulungiselela i- i-PC efakwe ezimbonini , noma ukufaka ikhompyutha ye-rackmount yezimboni.