Stage Profiling Breakdown Chart
Run the Stage Profiling Breakdown Chart MicroSim Fullscreen
You can include this MicroSim on your website using the following iframe:
1 2 | |
About This MicroSim
You have a working spectrum analyzer and it is too slow. Where do you start?
The wrong answer is wherever your intuition points. Intuition is famously bad at this — programmers routinely optimize the code they find most interesting rather than the code that is actually expensive.
The right answer is: measure first. This chart is the measurement.
| Stage | Share of frame time |
|---|---|
| Capture | 1% |
| Compute (FFT) | 66% |
| Draw (SPI to OLED) | 33% |
What the Numbers Decide
Capture is not worth touching. It is 1%. Even reducing it to literally zero buys you 1% of frame time. Any effort spent here is effort not spent where it matters.
Compute is the bottleneck, and it is not close. Two thirds of every frame. A 25% improvement to the FFT saves 16.5% of total frame time — more than eliminating capture and halving draw combined.
Draw is worth revisiting second. A third of the frame is real, and it is not nothing. But per unit of effort, compute pays roughly twice as well.
This is Amdahl's law in its most practical form: the speedup available from optimizing any stage is capped by that stage's share of the total. You cannot get more than 1% back from capture no matter how brilliant your optimization is.
How to Use
- Click each segment in turn and read its verdict.
- Compute the maximum possible speedup from perfectly optimizing each stage.
- Switch to the Pie chart view. Same data, and the same conclusion — the view does not change the answer, which is the point.
- Ask: if you had time to optimize exactly one stage, which one, and what is the best case you could hope for?
Lesson Plan
Grade Level
Undergraduate (college junior/senior)
Duration
8-10 minutes
Prerequisites
- The pipeline has capture, compute, and draw stages
- Percentages of a total
Learning Objective
Students will be able to judge, from a stacked breakdown of stage times, which stage should be prioritized for optimization, using measured percentages rather than intuition.
Activities
- Rank by payoff (3 min): Students rank the three stages by maximum achievable savings.
- Amdahl bound (4 min): Students compute the best-case overall speedup from making compute infinitely fast (about 3× — the other 34% remains).
- Argue against intuition (3 min): Students describe a plausible wrong guess about the bottleneck and what measurement would have corrected it.
Assessment
Ask: "A colleague spends a week hand-optimizing the capture routine and reports a 40% speedup in that stage. How much faster is the overall frame?" (0.4% — a week for four parts in a thousand.)
Technical Details
- Library: Chart.js 4.4.0
- Views: horizontal stacked bar and pie, toggled without changing the data
- Interaction: click any segment for its share and verdict
Related Resources
References
- Amdahl's law — the bound on speedup from optimizing one part.
- Profiling (computer programming) — why measurement precedes optimization.