引導式測驗 · 中等 等級
AI 功能與手機電池壽命測驗
了解為什麼手機 AI 電池使用量因工作負載、設備、處理路徑、感測器、網路和後台調度而異;了解如何測量它。
問題 1 的 8
Why can two AI features have different battery impact?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- Why can two AI features have different battery impact?
- On-device inference always consumes less battery than cloud inference
- They can differ in compute duration, sensors, display use, network, and scheduling
- Battery use is determined only by the model’s parameter count
- A cloud request requires all model inference to run locally
- Which testing design helps compare an AI task’s battery effect?
- Use the same device and task while also changing screen brightness and network conditions
- Treat response time alone as a direct energy measurement
- Compare one run with uncontrolled background apps against another
- Repeat a defined test against an idle or non-AI baseline on the same device
- Android 分析能幫開發者檢查測試裝置上的哪些項目?
- System activity such as CPU, memory, timing, and power behavior
- A universal energy cost per model call, independent of device and workload
- The factual accuracy of generated model responses
- The remote data center’s power draw for a cloud inference request
- 哪一組資料能提供最完整的手機端電池比較,針對雲端 AI 請求?
- Server compute and model size, without phone activity
- Radio and display time alone
- Network, screen time, request duration, and any local work
- Local inference only, excluding network and display activity
- Android 從待機桶限制描述了什麼?
- A fixed energy cost for each model token
- How system state can restrict scheduled app work; they are not AI energy estimates
- A guarantee that an app’s background job runs immediately
- The maximum size of a model loaded into RAM
- Which statement about an NPU and CPU is supported by this guide?
- An NPU always uses less energy for every task
- A CPU cannot run model inference
- An NPU removes sensor and network costs
- Efficiency depends on workload, chip, model, acceleration, and duration
- What should a report include when it claims an AI feature drains a certain amount?
- Device and workload, but no software version or measurement method
- Software and network conditions, but no device or task definition
- Device, workload, software, network conditions, and measurement method
- A single run without stating the phone, workload, or measurement method
- If battery use rises during a feature, what is a useful next diagnostic step?
- Check diagnostics for foreground compute, sensors, radio, and background work
- Assume background sensors caused the drain without checking diagnostics
- Disable every permission before identifying the source
- Replace the battery before comparing the feature against a baseline