Put every thread under a steady load straight from the browser, watch the throughput hold or fall away, and hear whether the cooling keeps up. It loads your processor — it does not benchmark it.
This page gives your processor something heavy to do and shows you whether it keeps doing it. That is a smaller claim than most CPU pages make, and it is deliberate.
The workers here run an arithmetic loop as fast as the browser allows and count how many passes they complete. That count depends on your browser, its version, what else the machine is doing, and how the engine decided to compile the loop today — so it is not a score, and it does not belong next to anyone else's. What it is good for is comparing this machine against itself, minute by minute, under a load you control.
Start a run and the throughput climbs for a few seconds, then settles. If it holds there for five minutes, the machine is sustaining the load. If it steps down and stays down, something is clocking the chip back — heat, a power limit, or a battery-saver profile. The page says so once it has enough history to be sure it is not looking at one noisy second.
No temperatures, no clock speeds, no fan RPM, no processor model, no core count you can fully trust. None of it is exposed to a web page on any platform, and a site that claims otherwise is guessing from your user agent. The one fact available before the test is the thread count the browser reports — and Safari and privacy modes round even that down. For temperatures, run a local monitor next to this page.
All threads is the setting for testing cooling: the whole package heats up, the fans have to answer, and a marginal cooler gives itself away within minutes. One thread does the opposite — the rest of the chip stays idle, so that core reaches its highest boost clock and holds it. Running one thread at a time is also how you find a single core that fails while the others are fine.
Full load is not always the thing you need to reproduce. A machine that is stable flat out but stutters at half load has a different problem, and a fan curve is far easier to hear at a steady 40% than with everything spinning at once. The slider duty-cycles the work inside each worker — computing for part of every 100 ms, idling for the rest — so the cores are genuinely loaded, just not continuously.
Yes. Nothing a web page does can push a processor past its own limits: thermal and power protection lives in the hardware and the firmware, below anything a browser can reach, and it clocks the chip down or shuts the machine off long before damage. What to expect is a hot machine, loud fans and, on a laptop, a battery going down fast. If the machine switches itself off during a run, that is a finding — the cooling or the power supply is not coping.
The quickest way to hear a cooling system under real load, and to find out whether a fan that never spins up is broken or simply never asked to work.
Run the same preset before and after. A machine that used to fall away after three minutes and now holds level for five has genuinely been fixed.
Twenty minutes in front of the seller is enough. A machine that throttles hard in the first five, or switches itself off, has a cooling problem you would be buying along with it.
Load every thread before trusting a new machine with real work. A cooler that was not seated properly, or a power supply that cannot hold up, shows itself under load and nowhere else.
Faults that only appear under load are miserable to chase. The endurance preset reproduces the conditions on demand, so you can watch what happens instead of waiting for it.
Hold the intensity at a steady 40 or 50% and listen. A fan that hunts up and down at partial load, or roars at a load the machine should handle quietly, points at the curve or a sensor rather than the fan.