liveBook · Manning

Mean power

As explained in Quantization and Fast Inference

Mean power is the average electrical power consumed by a device or workload over a measurement interval.

Mean power is the average power drawn over a measurement interval. For edge workloads, it is a sustained operating constraint: continuous power above a phone’s roughly 4–6 W thermal budget can cause latency to fall as the device throttles. section 11.1

A 30-second steady-state window is used to compare mean power across configurations. Treat the reading as one axis of a deployment budget alongside user-visible latency; a fast cold start does not establish that a workload can run sustainably. section 11.1.1 section 11.6

Device-specific tradeoffs

For EfficientNet-Lite0, the M3’s best configuration runs at 0.36 ms and 6.3 W, while the Pixel’s runs at 4.09 ms and 1.75 W. The faster result draws more power; which configuration fits depends on the deployment’s power budget. section 11.1

Figure 11.1 Cross-modal design space on MacBook Air M3 (laptop) and Google Pixel 10 Pro (phone). Marker shape encodes modality (circle = vision, triangle = audio, square = LLM). Twelve points populate the chart. Five M3 vision variants sit at 4.4–6.6 W, two M3 audio variants at 5.7–6.7 W, Pixel vision at 1.75 W, Pixel audio at 3.33 W, three Pixel LLM points at 5.51 W. The iPhone 16 is absent. iOS exposes no third-party API for instantaneous power (MetricKit reports daily-aggregated CPU and GPU time, not a milliwatt figure at the cadence this axis needs). M3 LLM is also absent. MLX publishes neither TTFT nor power telemetry, so the cell cannot land on either axis. Latency is p50 for vision and audio; TTFT for LLM. Power is powermetrics on macOS and BatteryManager on Android, both over a 30 s window.

Compare measured latency and mean power across device and modality; readings use powermetrics on macOS and BatteryManager on Android over 30 seconds. section 11.1.1

How Quantization and Fast Inference covers it

How deep it goes
Defines a 30-second steady-state measurement window but does not derive power from electrical fundamentals.
Examples
Compares EfficientNet-Lite0 power on M3 and Pixel devices and charts vision, audio, and LLM measurements.
What you can do after
Evaluate edge configurations against sustained power and latency budgets before selecting a deployment target.
How it connects
Connects mean power to latency, thermal throttling, and energy per inference.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Quantization and Fast Inference. The section links point to the parts of the book it draws on.