Read the metric before naming the limit
A frame-time graph shows how long each frame took, so it reveals delivery problems that an average FPS value hides. Read the vertical spikes, the stable baseline and the timing budget together. A 60 FPS target allows 16.7 milliseconds per frame, while 144 FPS allows only 6.9 milliseconds.
A metric is useful only when the capture window, scene and warm-up state are documented. First hypothesis to test: A one-off asset load can create an isolated spike.
Create a repeatable performance capture
Use the first two controls in order: Capture the same sixty-second route three times; then Remove menus and loading screens from the sample. This establishes whether the first observation survives a controlled comparison.
Run the checks in this order
- Capture the same sixty-second route three times.
- Remove menus and loading screens from the sample.
- Mark the target frame-time line before comparing runs.
- Match each spike with clocks, utilization and memory activity.
Repeat until the normal run-to-run range is visible; a claimed change must be larger than that range.
Match the trace to the controlled change
Result, control and meaning
| Observation | Controlled change | Likely interpretation |
|---|---|---|
| Single isolated spike | Repeat after warm-up | Compilation or asset loading is more likely |
| Sustained high baseline | Lower rendering load | A clear response supports GPU pressure |
| Flat ceiling | Check caps and V-Sync | The limiter may be intentional |
When the observation is “Single isolated spike,” apply “repeat after warm-up.” The evidence table reads the response as follows: Compilation or asset loading is more likely.
Capture conditions that can distort the reading
Compare two early explanations: (1) A one-off asset load can create an isolated spike; and (2) A repeating sawtooth can indicate pacing or synchronization behavior. They can resemble one another in a headline result but predict different responses to the checks below.
Alternative causes to inspect
- A one-off asset load can create an isolated spike.
- A repeating sawtooth can indicate pacing or synchronization behavior.
- A rising baseline can follow heat, power or scene complexity.
- Regular flat steps can be caused by a frame cap.
Keep the sampling tool and route unchanged so a difference in the trace reflects the tested variable rather than a new capture method.
Set a measurement-based action rule
Choose the next test from the shape of the trace. Repeating high frame time needs a throughput test; isolated spikes need an event correlation. Save the raw capture so a driver, setting or hardware change can be compared against exactly the same route.
Test record
Record sample duration, warm-up pass, route, cap and every reported unit beside the result.
- capture duration
- millisecond axis
- spike timestamp
- baseline band
- target budget
- warm pass
- present interval
- event marker
Checks that can overturn the decision
- Treating one spike as a permanent bottleneck.
- Comparing traces from different scenes.
- Reading FPS and milliseconds as interchangeable.
Decision boundary: The decision above should also survive the most likely shortcut: treating one spike as a permanent bottleneck.
Related measurement guides
Build the cluster one verified step at a time
Compare this result with PC Benchmark Run-to-Run Variance Explained, CPU Time vs GPU Time: Find the Slower Frame Stage, and Average FPS vs 1% Lows: What Each Metric Means; keep the same workload and baseline while testing the competing explanations.
Technical sources
References for the measurement method
