Nkewa GPU nke ọtụtụ ihe atụ
Multi-Instance GPU (MIG) bụ teknụzụ NVIDIA nke na-egbutu otu GPU anụ ahụ n'ime ọtụtụ akụkụ ngwaike dịpụrụ adịpụ.
Nchịkọta
It matters because it lets one expensive accelerator serve many small workloads at once without them interfering with each other.
Ime miri emi
Ewebata ya na NVIDIA A100 (Ampere) wee gaa n'ihu na H100 na GPU nke etiti data ọhụrụ, MIG na-ese GPU n'ime ihe ruru ugboro asaa nọọrọ onwe ha. N'adịghị ka slicing oge ngwanrọ, MIG na-enye ezigbo ikewapụ ngwaike: ihe atụ ọ bụla na-enweta multiprocessors gụgharia raara onwe ya nye (SMs), Mpekere cache L2, ndị na-ahụ maka ebe nchekwa na ebe nchekwa dị elu nke ebe nchekwa bandwidth dị elu. Enwere ike kewaa A100 nwere 40GB n'ọnọdụ asaa 5GB, ma ọ bụ nke pere mpe. Nkebi nke ọ bụla na-eme ka obere GPU kwụ ọtọ, yabụ ọrụ mkpọtụ ma ọ bụ na-akụda n'otu oge enweghị ike ịnwụ ma ọ bụ mebie ọzọ. Ọrụ a na-ekwe nkwa na-eme ka MIG dị mma maka ozi ntinye aka, ụyọkọ ndị nwe ụlọ, na gburugburu mmepe ebe ọtụtụ ndị ọrụ na-ekerịta otu kaadị.
Nghọta nka nka
MIG na-arụ ọrụ site n'iji anụ ahụ na-eme ka ihe dị n'ime GPU dị n'ime ya ka ihe atụ ọ bụla nwee ụzọ edoziri na ebe nchekwa nke ya na SMS. NVIDIA na-akọwapụta profaịlụ dị ka ụmụ irighiri ihe dị ka 1g.5gb (otu mpekere compute, 5GB) ruo 7g.40gb. Ihe ngosi GPU na-edobe ebe nchekwa na SMS; N'ime ya, Compute Interstance na-ekesa SMS n'ihu. N'ihi na akụkụ ndị ahụ bụ ngwaike-mmanye, mmejọ, njehie ECC na bandwit ebe nchekwa nọrọ naanị otu ihe atụ.
Mmetụta atụmatụ
Ọnụ ego na mmefu ego
Mkpebi ihe owuwu ụlọ na-akwalite arụmọrụ yana ọnụ ahịa ọrụ ruo ọtụtụ afọ.
Mkpebi doro anya
Nkà mmụta nka na-enyere ndị otu egwuregwu aka ịhọrọ nchịkọta ziri ezi, ọ bụghị naanị nke kachasị ọhụrụ.
Quality akara
Nhọrọ injinia ka mma na-ebelata ihe omume ntụkwasị obi na mmepụta.
Ọdịnihu nke nkewa GPU ọtụtụ ihe atụ
Ka GPU na-eto ruo 80GB, 141GB, na karịa, nkewa na-adọrọ mmasị karị n'ihi na ụdị onye ọ bụla anaghị achọ kaadị zuru ezu maka ntinye aka. Na-atụ anya Kubernetes siri ike na njikọta ígwé ojii, nbibi dị ike na-enweghị igbapu ọnụ ọnụ, yana profaịlụ mara mma. Ndị na-ere ahịa na-asọmpi na-agbaso ụdị SR-IOV-ụdị GPU virtualization, na ikpo okwu na-enweghị ihe nkesa na-adabere na nkewa iji kwakọba ọtụtụ ụdị ma bepụ ihe mkpofu na-abaghị uru.
Mmejuputa n'ezie n'ụwa
Onye na-eweta ígwé ojii na-ekewa otu A100 n'ime oge asaa ka ndị ahịa asaa ọ bụla na-enweta nkwa nke GPU dịpụrụ adịpụ maka ntinye.
Otu ụyọkọ nyocha nke mahadum na-enye nwa akwụkwọ PhD ọ bụla ihe atụ 10GB MIG maka imepụta ihe kama ịchịkọta kaadị niile.
Ọrụ inference na-achịkọta ọtụtụ obere asụsụ na ụdị ọhụụ n'otu H100, nke ọ bụla na nkebi nke ya nwere ike ịnwụ anwụ.
Otu ụyọkọ Kubernetes na-akpọsa oge MIG dị ka akụrụngwa nwere ike ịhazi ya mere pods rịọ 'nvidia.com/mig-1g.5gb' dị ka akụrụngwa ọ bụla ọzọ.
Ihe ize ndụ & okporo ụzọ nche
Ịkwalite otu akara ngosi nwere ike zoo adịghị ike sistemụ sara mbara.
A na-eledakarị ihe akụrụngwa na ụgwọ ọrụ anya.
Ọdịiche nchekwa na nleba anya nwere ike itolite ka sistemu na-adịwanye mgbagwoju anya.
Map mmejuputa
Kọwaa latency, ịdịmma na ebumnuche ọnụ ahịa tupu mmejuputa ya.
Benchmark n'okpuru ibu dị adị na ọnọdụ data.
Nleba anya akụrụngwa maka mperi, ịkpafu na mmetụta onye ọrụ.
Kwadebe ụzọ nzaghachi azụghachi azụ na ihe omume tupu ịchachaa.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Ịmụta ọtụtụ ọrụ
Ajụjụ a na-ajụkarị
What is Multi-Instance GPU Partitioning?
Multi-Instance GPU (MIG) bụ teknụzụ NVIDIA nke na-egbutu otu GPU anụ ahụ n'ime ọtụtụ akụkụ ngwaike dịpụrụ adịpụ. Ọ dị mkpa n'ihi na ọ na-ahapụ otu ngwa ngwa ngwa ngwa dị oke ọnụ na-eje ozi ọtụtụ obere ọrụ n'otu oge na-enweghị ha na-etinye aka na ibe ha.
Kedu ụdị ikewapụ MIG na-enye n'etiti ihe atụ?
MIG na-akwado ikewapụ na ngwaike, na-arara SMs, Mpekere cache L2 na ebe nchekwa na oge ọ bụla ka ibu ọrụ enweghị ike igbochi ya.
Na nke NVIDIA architecture ka MIG bu ụzọ webatara?
MIG gbara arụkwaghịm na A100 dabere na Ampere wee gaa n'ihu na H100 na GPU nke etiti data emesia.
Kedu ọnụ ọgụgụ kachasị nke GPU nwere ike kewaa MIG?
Enwere ike kewaa GPU nwere ike MIG dị ka A100 n'ime ihe ruru asaa nọọrọ onwe ya.
Na profaịlụ MIG dị ka '1g.5gb', gịnị ka '5gb' na-anọchi anya?
Profaịlụ ahụ kpuchiri mpekere (1g) yana ebe nchekwa raara onwe ya nye (5GB) enyere na ọmụmaatụ.
Kedu ọrụ MIG dabara adaba maka?
MIG na-enwu mgbe ọtụtụ obere ọrụ dịpụrụ adịpụ (dị ka ntinye) na-ekerịta otu kaadị nwere ogo ọrụ.