Thermal throttling

Every modern CPU and GPU carries an on-die temperature limit, commonly around 95-105 °C for a CPU junction and 83-90 °C for a GPU core. When a sensor reaches that limit the chip drops frequency and voltage within microseconds. Nothing is broken when this happens; the mechanism is protective and designed-in. What matters is whether it happens at the temperature you expected and how far the clock falls.

Throttling is the single most common reason a machine feels fast for the first ten seconds of a workload and mediocre afterwards. A short benchmark run finishes before the heat has propagated from the die into the heatsink, so it reports the boost clock. A long export, compile, or game session runs almost entirely in the throttled state. This is why peak numbers and sustained numbers can differ by 30% or more on the same laptop.

SystemCheck cannot read your temperature sensors. No browser can. What it can do is measure the consequence. The 60-second sustained multi-core CPU load records throughput continuously, and the decay figure in your report is the drop from the level it settled at after its boost to the end of the run. The early boost step is shown on its own and not counted. A large decay on a load that never changes is consistent with throttling, whether from heat, a power limit, or a current limit.

If you need the actual temperature that triggered it, pair SystemCheck's decay measurement with HWiNFO64, Core Temp, or your motherboard vendor's monitoring utility on Windows, or `sensors` on Linux. SystemCheck tells you that performance is falling and by how much; a hardware monitor tells you which limit caused it.