Multi-Misali GPU Rarraba
Multi-Instance GPU (MIG) fasaha ce ta NVIDIA wacce ke yanka GPU na zahiri guda ɗaya zuwa ɓangarorin keɓancewar kayan masarufi da yawa.
Dubawa
It matters because it lets one expensive accelerator serve many small workloads at once without them interfering with each other.
Zurfafa nutsewa
An gabatar da shi tare da NVIDIA A100 (Ampere) kuma ya ci gaba akan H100 da sabbin GPUs na cibiyar bayanai, MIG ya zana GPU zuwa lokuta masu zaman kansu har zuwa bakwai. Ba kamar yankan lokaci na software ba, MIG yana ba da keɓewar kayan aiki na gaskiya: kowane misali yana samun nasa keɓaɓɓun na'urorin sarrafawa masu yawa (SMs), yankan cache na L2, masu sarrafa ƙwaƙwalwar ajiya, da ƙayyadadden yanki na ƙwaƙwalwar bandwidth mai girma. Ana iya raba A100 tare da 40GB zuwa lokuta bakwai na 5GB, ko kaɗan mafi girma. Kowane bangare yana aiki kamar ƙaramin GPU mai zaman kansa, don haka aikin hayaniya ko faɗuwa a wani misali ba zai iya yunwa ko lalata wani ba. Wannan ingantaccen ingancin sabis yana sa MIG ya zama manufa don ƙaddamar da sabis, gungu masu haya da yawa, da mahallin ci gaba inda yawancin masu amfani ke raba kati ɗaya.
Fahimtar Fasaha
MIG yana aiki ta hanyar shigar da mashigin giciye na GPU ta jiki don haka kowane misali yana da madaidaiciyar hanya zuwa yanki na ƙwaƙwalwar ajiya da SMs. NVIDIA tana bayyana bayanan martaba azaman juzu'i kamar 1g.5gb (yanki guda ɗaya, 5GB) har zuwa 7g.40gb. Misalin GPU yana adana ƙwaƙwalwar ajiya da SMS; A cikin sa Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar Ƙididdigar a cikinsa tana ƙara rarraba SMs. Saboda ɓangarorin an tilasta musu kayan aiki, kurakurai, kurakuran ECC, da bandwidth na ƙwaƙwalwar ajiya sun kasance suna iyakance ga misali guda.
Dabarun Tasiri
Kudin da kasafin kuɗi
Hukunce-hukuncen gine-gine suna haifar da aiki da tsadar aiki na shekaru.
Shawarwari masu haske
Ilimin fasaha yana taimaka wa ƙungiyoyi su zaɓi tari mai kyau, ba kawai sabon abu ba.
Kula da inganci
Zaɓuɓɓukan injiniya mafi kyau suna rage abin dogaro a cikin samarwa.
Makomar Rarraba Multi-Misali GPU
Yayin da GPUs ke girma zuwa 80GB, 141GB, da kuma bayan haka, rarrabuwa ya zama mafi ban sha'awa saboda ƙirar mutum ɗaya da wuya yana buƙatar cikakken katin don tunani. Yi tsammanin kubernetes mai ƙarfi da haɗin gajimare, rarrabuwar kawuna mai ƙarfi ba tare da ɓata kumburi ba, da ingantaccen bayanan martaba. Dillalai masu gasa suna bin irin wannan SR-IOV-style GPU kamanceceniya, kuma dandamalin ba da izini na uwar garken yana ƙara dogaro da rarrabuwa don tattara samfuran da yawa da kuma yanke sharar gida.
Aiwatar da Gaskiyar Duniya
Mai ba da girgije yana raba A100 guda ɗaya zuwa yanayi bakwai don haka abokan ciniki bakwai kowanne ya sami garanti, yanki na GPU keɓe don ra'ayi.
Rukunin bincike na jami'a yana ba kowane ɗalibin PhD misali na 10GB MIG don yin samfuri maimakon sarrafa duka katunan.
Sabis na ƙididdigewa yana tattara ƙananan harsuna da ƙirar hangen nesa a kan H100 guda ɗaya, kowanne a cikin ɓangarensa tare da jinkirin da ake iya faɗi.
Tarin Kubernetes yana tallata misalan MIG azaman albarkatun da za'a iya tsarawa don haka kwas ɗin suna buƙatar 'nvidia.com/mig-1g.5gb' kamar kowane hanya.
Hatsari & Tsare-tsare
Haɓaka ma'auni ɗaya na iya ɓoye manyan raunin tsarin.
Sau da yawa ana raina kayan more rayuwa da kuma kuɗin kulawa.
Tsaro da gibin lura na iya girma yayin da tsarin ke ƙara haɓaka.
Taswirar Hanya
Ƙayyade latency, inganci, da maƙasudin farashi kafin aiwatarwa.
Alamar ma'auni a ƙarƙashin ainihin kaya da yanayin bayanai.
Kula da kayan aiki don kurakurai, ɗigo, da tasirin mai amfani.
Shirya bijirowa da hanyoyin mayar da martani kafin sikeli.
Ci gaba da Bincike
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Multi-Instance GPU Partitioning quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Jagora na gaba
Koyon Ayyuka da yawa
Tambayoyin da ake yawan yi
What is Multi-Instance GPU Partitioning?
Multi-Instance GPU (MIG) fasaha ce ta NVIDIA wacce ke yanka GPU na zahiri guda ɗaya zuwa ɓangarorin keɓancewar kayan masarufi da yawa. Yana da mahimmanci saboda yana ƙyale mai haɓaka mai tsada ɗaya yayi hidimar ƙananan ayyuka da yawa a lokaci ɗaya ba tare da sun shiga tsakani ba.
Wane irin keɓewa MIG ke bayarwa tsakanin al'amura?
MIG yana tilasta warewa a cikin kayan aiki, keɓe SMs, yankan cache L2, da ƙwaƙwalwar ajiya ga kowane misali don haka nauyin aiki ba zai iya tsoma baki ba.
A kan wanne gine-ginen NVIDIA aka fara gabatar da MIG?
MIG yayi muhawara tare da tushen A100 na Ampere kuma ya ci gaba akan H100 da GPUs na cibiyar bayanai daga baya.
Menene matsakaicin adadin lokuta da GPU mai iya MIG za a iya raba?
GPU mai iya MIG kamar A100 ana iya raba shi zuwa lokuta masu zaman kansu har zuwa bakwai.
A cikin bayanin martaba na MIG kamar '1g.5gb', menene '5gb' ke wakilta?
Bayanin bayanan yana ɓoye ƙididdige yanki (1g) da keɓaɓɓen ƙwaƙwalwar ajiya (5GB) da aka ba misalin.
Wane nauyin aiki ne MIG ya fi dacewa da shi?
MIG yana haskakawa lokacin da ƙananan ƙananan ayyuka keɓance (kamar ra'ayi) suna raba kati ɗaya tare da ingantaccen ingancin sabis.