Benchmarks that run for thirty seconds tell you about a laptop’s peak. Work that runs for thirty minutes tells you about its cooling. The gap between those two figures is thermal throttling, and it is the main reason identical chips behave differently across machines.
Processors are designed to boost aggressively when thermal headroom exists and to pull back when it does not. When silicon approaches its temperature limit, firmware reduces clock speed and voltage to stay within safe bounds. This is protective and normal. The question is not whether a laptop throttles, but how far it falls and how quickly.
A well designed machine settles into a sustained state that is meaningfully below peak but still strong. A poorly designed one drops sharply within a couple of minutes and stays there.
A processor can only dissipate as much heat as the cooling system can move. That depends on heatsink mass, heat pipe layout, fan size and airflow path, and how much of the chassis can act as a thermal mass. Thin, sealed, fanless designs have a hard physical ceiling regardless of the chip inside.
Manufacturers also set their own sustained power limits. Two vendors using the same processor may configure very different long-duration power targets, producing very different sustained performance from nominally identical hardware.
Throttling often gets blamed on software. A machine that feels quick in the morning and sluggish during a long export is usually thermally limited, not infected or badly configured. Fan noise ramping then settling while performance drops is the clearest signal.
Airflow is the cheapest fix. Most laptops intake from underneath, so operating on a soft surface that blocks vents has a direct and measurable cost. A hard surface or a simple stand helps.
Dust accumulation in fins genuinely degrades cooling over years of use. Periodic cleaning restores some headroom on older machines.
Many manufacturers ship performance profiles that trade noise for sustained clocks. If you need sustained throughput and can tolerate fan noise, the high performance profile is usually worth selecting deliberately rather than leaving on balanced.
Repasting can help on older machines where the original thermal compound has degraded, but it carries warranty risk and is rarely transformative on a well designed modern laptop.
A short benchmark rewards machines that boost hard and cool poorly. Because that is not how most real work behaves, our bench holds load long enough for the chassis to reach thermal equilibrium before recording a result. The methodology is documented in how we test.
Browse tested machines in our laptops section.
What we tested this week, what failed, and what is actually worth your money.
Fixed bench, published scoring weights, dated corrections. Every commission disclosed.
Checked against 90 days of price history before we call it a deal.
Charge cycles matter less than heat and sustained time at full...
ReadSpec sheets are written by marketing departments. Here is which numbers...
Read
Same scenes, same light, same processing settings across every handset.
Read
Where the compromises actually bite, and which models avoid the worst...
ReadSend the counter-evidence. If it holds up we retest and publish a dated correction on this page.