Thermal Throttling

Every chip that computes generates heat in the process. Beyond a certain temperature threshold — for modern processors usually between 90 and 105 degrees Celsius — the chip automatically switches into a slower operating mode. It performs fewer calculations per second, thereby generating less heat and cooling itself down. This process is called thermal throttling. It happens completely automatically and without any user intervention. The device is not damaged in the process — that is, after all, precisely the point of the mechanism.

Why chips have to slow themselves down

Silicon, the material chips are made of, can only tolerate heat up to a certain point. If a chip stays too hot for too long, it ages faster and, in the worst case, can fail permanently. Thermal throttling is therefore not a defect but a deliberately built-in safety feature — comparable to a thermostat that shuts off the heating before it gets too hot.

The problem arises because performance and heat are directly linked. A chip that computes quickly needs more electrical power. More power means more waste heat. If the cooling can’t keep up — because it’s built too weakly or the surrounding air is too warm — the temperature keeps rising. Without thermal throttling, the chip would simply keep computing until it destroyed itself.

This is especially relevant for compact devices like laptops or smartphones. There’s little room for cooling systems there, meaning large fans or thick metal heat sinks. As a result, the chips are far more often at their thermal limit than in a spacious desktop PC.

How throttling unfolds inside the chip

Every modern processor contains tiny temperature sensors built directly into the chip. They continuously measure how hot individual areas currently are. If the temperature exceeds a defined threshold, a control unit steps in and lowers the so-called clock speed — that is, the number of computing steps per second. A chip that normally runs at 3.5 gigahertz can be throttled down to 1.5 gigahertz or less in this process.

The throttling happens gradually and dynamically. If the temperature drops again, the chip raises its clock speed once more. This control loop runs many times per second. Users usually don’t notice it directly, but rather indirectly: a game suddenly starts stuttering, a video export takes longer than expected, or a laptop becomes noticeably slower after running under full load for twenty minutes.

A common misconception is to mistake thermal throttling for a software bug. It is not a crash, not a broken app, and not a virus. It is physics: the chip cannot output more performance than the cooling system can carry away. Anyone who wants to solve the problem permanently needs to improve the cooling — for example by cleaning the fans, replacing the thermal paste between the chip and the heatsink, or operating the device in a cooler room.

Thermal throttling in AI chips and in practice

In the AI and tech world, thermal throttling is an ongoing topic especially for graphics cards. These chips — called GPUs, or graphics processing units — handle the bulk of the computations during the training and operation of AI models. In doing so, they work under full load for hours and generate enormous amounts of waste heat. Data centers therefore invest substantial sums in cooling systems to avoid thermal throttling — because every instance of throttling means less computing power per hour and thus higher costs.

In everyday life, the term frequently comes up in reviews of laptops and smartphones. Tech publications specifically test whether a device throttles under sustained load and by how much. A laptop that’s fast in a short benchmark but throttles heavily after ten minutes is, in practical terms, worth less for tasks like video editing or long render jobs than its test results would suggest.

Apple has gained a lot of attention with its M-series chips for MacBooks by almost entirely avoiding thermal throttling under everyday conditions — not because the chips don’t generate heat, but because they are built so efficiently that the cooling can easily keep up. This shows that thermal throttling is not an unavoidable fate, but a design problem that can be reduced through smarter engineering.