A hot CPU rarely crashes, it quietly drops clocks until the game stutters. Read the right sensor, separate thermal from power limits, then fix the cause.
Watch **per-core temperatures and clocks at the same time** under a repeatable all-core load. Clocks falling at 95 degrees or above is thermal; a flagged power or current limit is the board's configuration.
Use a tool that shows per-core temperature, package power and the current clock side by side. A single CPU temperature from the firmware is an idle reading and says nothing about a loaded chip.
HWiNFO64 > Sensors-onlyRun a ten minute all-core test with the window visible. The first minute shows the boost clock, the tenth shows what the cooler can actually hold; only the second number matters for a long session.
Prime95 Small FFTs / Cinebench multi-coreThermal throttling begins at the chip's throttle point, usually 95 to 100 degrees. Power and current limits cut clocks earlier and appear as separate flags, so a chip sitting at 75 degrees can still be limited.
HWiNFO: Core Thermal Throttling / Package Power LimitCheck that the front intake is not blocked by a filter mat, that the cooler fan blows toward the rear exhaust, and that the case has as much intake as exhaust. Ten degrees often live in the case, not in the heatsink.
A curve that only reacts at 80 degrees is already late. Start the ramp around 55 and reach full speed by 80; the noise difference is small and the clock difference is not.
BIOS > Hardware Monitor > Fan CurveA lower long-duration power limit costs a few percent of multi-core performance and buys back most of the temperature. Change one limit, then re-run the same ten minute test and compare the clock floor.
BIOS > PL1 / PL2 (Long / Short Duration)Background load, a pump that is not spinning, or a fan curve with a zero-rpm stop. Check the pump RPM before anything else.
Uneven paste spread or a cooler tightened on one side. Re-mount it with even, alternating pressure.
Dust in the fins and a shared heat pipe. Clean the vents and raise the rear of the chassis before you consider repasting.
A curve that starts too high, or a firmware update that replaced your profile. Reapply the curve and save it to a slot.
Below the throttle point it is within spec. It costs clock speed, so it matters for performance rather than for the chip's survival.
Both read the same sensor. Software under load is the only reading worth judging.
Restore the default fan curve and the stock power limits in the firmware, then save. Temperatures return to what they were before the changes; nothing on the disk is touched.
Free · no account needed · tracks your progress