Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo Cyinshi cya GPU Ikibazo

Koresha imyirondoro ya MIG, GPU-na-kubara urugero rw'ibikoresho, hamwe no kwerekana ibikoresho, harimo n'impamvu ibyuma bishyigikiwe bigena guhitamo ibice.

Inzira zijyanye no kuyoboraIbice byinshi GPU Gutandukana
Ikibazo 1 Bya 8

Which resource boundary distinguishes separate MIG GPU Instances from software time slicing?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Which resource boundary distinguishes separate MIG GPU Instances from software time slicing?
    • Only a different process priority on one shared execution queue
    • A guaranteed absence of every host or device failure
    • A separate network cable for every CUDA kernel
    • Dedicated hardware compute and memory resources for each GPU Instance
  2. Ni ubuhe bwubatsi bwa NVIDIA MIG yatangijwe bwa mbere?
    • Volta
    • Ampere (A100)
    • Turing
    • Pascal
  3. On an A100 40 GB using supported small-instance profiles, how many GPU Instances can a layout provide?
    • Exactly sixteen on every supported GPU
    • Exactly two regardless of profile
    • Up to seven
    • An unlimited number with no memory limits
  4. Mu mwirondoro MIG nka '1g.5gb', '5gb' igereranya iki?
    • Ububiko bwagenewe urwo rugero
    • Umubare wa SMs
    • Umuyoboro wa PCIe
    • Umuvuduko w'isaha
  5. Nuwuhe murimo MIG ubereye cyane?
    • Amahugurwa manini manini akeneye GPU yose
    • Multi-tenant umwanzuro ukorera hamwe nubukererwe buteganijwe
    • CPU ihujwe namakuru mbere yo gutunganya
    • Kubika igihe kirekire
  6. Nigute cluster ya Kubernetes ishobora kwerekana MIG ingero kuri pod?
    • Nka disiki mbisi
    • Binyuze mu ntoki SSH
    • Nkibikoresho byateganijwe nka 'nvidia.com/mig-1g.5gb'
    • Binyuze mu nyandiko za DNS
  7. Two Compute Instances belong to the same MIG GPU Instance. Which resources do they share?
    • Their parent GPU Instance’s memory resources
    • Only a remote CPU’s filesystem
    • Separate complete copies of physical GPU memory
    • Every compute slice on unrelated GPUs
  8. A team wants to deploy a named MIG profile on a new GPU model. What should it check first?
    • That every NVIDIA GPU supports identical profiles
    • That the model parameter count is an even integer
    • That the Kubernetes pod name contains the profile text
    • That the GPU supports MIG and that exact profile layout