Kvíz spojený s průvodcem · Střední úroveň

Co je NPU v kvízu o telefonu nebo notebooku

Pochopte roli NPU, kompatibilitu softwaru, umístění provádění, hodnocení TOPS a kontroly potřebné pro užitečnou akceleraci AI na zařízení.

Otázka 1 z 8

Který popis nejlépe vystihuje roli NPU v telefonu nebo notebooku?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. Který popis nejlépe vystihuje roli NPU v telefonu nebo notebooku?
    • Specialized hardware for supported neural-network computation
    • A complete trained model stored as a processor
    • A guarantee that every application runs offline
    • A replacement for all general application software
  2. Notebook obsahuje NPU. Co by měl uživatel zkontrolovat, než se domnívá, že určitá aplikace pro přepis má z toho výhody?
    • Only the laptop’s screen resolution
    • Whether that application and its model use the supported NPU execution path
    • Only the size of the app icon
    • Whether the app has AI in its name
  3. Proč by se jeden modelový graf mohl spustit na více než jednom backendu?
    • Each backend must always receive an equal share
    • Model files cannot contain more than one operation
    • Different backends may support different nodes or subgraphs
    • An NPU automatically sends every operation to the cloud
  4. Co popisuje hypotetická hodnota 10 TOP hardwaru?
    • Ten trillion complete predictions each second
    • Ten seconds per model inference
    • Ten terabytes of model memory
    • Ten trillion counted operations per second under stated assumptions
  5. Jaká kontrola je nutná po převodu modelu na nižší numerickou přesnost pro zrychlení?
    • Evaluate task quality as well as performance on the intended device
    • Assume every answer stays identical
    • Check only that the file became smaller
    • Stop measuring resource use
  6. A video effect has a fast isolated model benchmark. Which additional measurement helps assess the actual application?
    • Only the advertised chip rating
    • Complete frame latency, power, sustained behavior, and visual output
    • Only the model filename
    • Only the number of processor brands in the device
  7. Why does local NPU inference alone not prove that an application keeps all data on the device?
    • NPUs always upload their inputs
    • Local hardware cannot process private data
    • Other application features may upload logs or synchronize results
    • Privacy depends only on the TOPS rating
  8. Which information makes an NPU benchmark easier to interpret?
    • Only the highest observed operation count
    • Only whether the model launched successfully
    • Only the device’s marketing category
    • Model version, precision, input shape, software versions, and measurement conditions