What makes a device throttle, and how do you spot it?
Throttling is a device deliberately slowing itself down — for heat, for power, or for the battery. Here is what causes it, how it looks in a measurement, and why a benchmark partly causes the thing it reports.
Throttling is a device choosing to run slower than it can. That is a feature, not a fault: the alternative is overheating, shutting down, or drawing more current than the battery can safely give. Every phone and laptop does it. What differs is when it starts and how far it goes.
Four separate causes, often confused
- Thermal. Silicon has a temperature limit. As the chip approaches it, clock speeds come down. A case, direct sun, or charging while gaming all bring this on sooner.
- Power. Sustained work is capped at a lower level than brief work, on purpose. A processor may run at its maximum for a few seconds and then settle to a level it can hold indefinitely.
- Battery. An older cell cannot deliver sharp current spikes. The system caps the processor to stay inside what the battery can supply, so an ageing battery slows a device that is otherwise perfectly healthy.
- Software. Battery-saver modes cap performance deliberately, and often switch on automatically at a charge threshold you have forgotten about.
These produce the same symptom and need different fixes, which is why guessing between them is unhelpful. The sequence of the slowdown is the clue: thermal throttling builds gradually as heat accumulates, a power cap appears abruptly a few seconds in and then holds steady, and a battery-saver cap is there from the first second.
How it looks in a measurement
A benchmark that reports one number cannot show throttling at all — the number is already an average of the fast part and the slow part. What shows it is watching the same work over time.
- Speed under load. The share of its starting speed a device is still holding as the run continues. A flat line means it held; a curve bending downward is throttling as it happens.
- Pass spread. Each subtest here runs three times. If the later passes come in materially below the first, the device slowed while being measured.
- 1% low against average. Thermal throttling tends to drag both down together. Stutter from background software drops the 1% low while leaving the average largely intact.
The benchmark causes some of what it reports
This is worth being honest about. A full run here is roughly thirty seconds of deliberately heavy load, and thirty seconds of heavy load is enough to warm a phone measurably. The second half of a run is therefore measuring a hotter device than the first half.
That is unavoidable in any sustained benchmark, and it is why the test runs three shorter passes rather than one long one, and reports how far they differed rather than hiding the variation inside an average. It is also why a result taken immediately after gaming will be lower than one taken from cold — both are true measurements of the device in that state.
What cannot be measured in a browser
There is no web API for processor or battery temperature. None. Any web page showing you a temperature in degrees, or a heat map of where a chip is hot, is drawing rather than measuring — this project shipped exactly that mistake once and deleted it.
What can be measured is the effect: work completed per unit of time, and how that changes across a run. Slowdown under sustained load is real and observable. The temperature causing it is not, and calling an inferred slowdown a temperature would be inventing a figure.
Reducing it
- Take the device out of its case before a long session.
- Do not charge and game at the same time; both generate heat in the same place.
- Keep it off soft surfaces — a bed or a cushion insulates a laptop’s vents.
- Check whether battery saver has switched itself on.
- Let it cool fully, then run the test and compare against a run taken while warm.
Measure the device you are holding
About a minute, nothing installed, and it works on a device you do not own.
Run the test