CPU Stress / Throttle Test
Watch whether your machine slows down under a minute of sustained load — and stays down.
How this benchmark works
A steady workload runs for one minute while the completed rate is sampled every two seconds, producing a curve you can see as it happens. The verdict compares the median rate over the opening quarter of the run against the median over the closing quarter. Medians on both ends rather than a peak against a final sample, deliberately: one fast interval at the start and one slow one at the end would otherwise manufacture a forty per cent drop out of ordinary scheduling noise, and telling someone their laptop is throttling when it is not sends them to a repair shop for nothing. A drop of twenty per cent or more that persists is reported as throttling, eight to twenty as mild, and anything below that as steady. This is the one benchmark on the site that genuinely diagnoses something, because a machine compared against itself sixty seconds earlier is a controlled comparison in a way that a score compared against strangers' machines never is.
What the results mean
- Performance drop
- How far the rate fell from the opening quarter of the run to the closing quarter. Under 8% is steady, 8 to 20% is the ordinary settling of a boost clock, and above 20% sustained is throttling.
- Slowdown began
- When the rate first fell below 85% of the opening rate. A drop that starts at 20 to 40 seconds is the classic thermal pattern; one that starts immediately is more likely to be other software competing.
- Rate over time
- The live curve. A staircase down that then flattens is a throttle. A jagged line at a steady average is other software competing for the processor, which is a different finding.
- Steady
- The rate at the end matched the rate at the start. Under a minute of single-threaded load this machine did not slow down — which does not rule out throttling under a heavier all-core load.
Common problems and fixes
- The machine is quick for a minute then becomes sluggish
- That is exactly the pattern this test is built to show, and a throttled result confirms it rather than leaving it as a feeling. Start with airflow: a laptop on a bed or cushion blocks its intake, and that alone accounts for a large share of cases.
- It throttles on battery but not on mains
- Several manufacturers cap performance hard on battery power. Run it once on each to confirm, then check the power mode — this is a setting rather than a fault, and it is usually adjustable.
- The curve is jagged rather than stepping down
- That is contention, not throttling. Something else is taking the processor in bursts. Close other applications and run it again — a curve that flattens out identifies the cause.
- You want to know the temperature it reached
- No web page can tell you. Core temperature, fan speed and power limits are not exposed to a browser at all. A native monitoring tool run alongside this test will show them, and the two together are much more informative than either alone.
Frequently asked questions
Is it safe to run this?
Yes. It runs one thread of ordinary JavaScript for one minute and stops on its own, and there is a stop button throughout. Every device is designed to protect itself by slowing down long before anything is at risk — that protection is precisely what this measures. It will make a thin laptop warm, which is why it asks before starting.
Why compare medians instead of the fastest and slowest samples?
Because the fastest and slowest samples are noise. Scheduling on a real machine produces occasional outliers in both directions, and building a verdict from two of them would report throttling on a machine that is fine. Medians at each end mean the finding has to be sustained to be reported at all.
Can it tell me whether the thermal paste needs replacing?
No. It can show that performance falls away under load and roughly when — the shape of the problem. Dried paste, a blocked heatsink, a failing fan, dust and an aggressive power profile all produce the same curve, and separating them needs temperature readings a browser cannot get.
Why single-threaded when throttling is worse with all cores loaded?
Because a single thread produces a clean, interpretable curve, and running every core from a browser measures how many workers the browser will schedule as much as it measures the machine. A single-threaded throttle is a strong finding: if one thread is enough to trigger it, an all-core load will be considerably worse.