Why Your Laptop Throttles: Field Notes on Power Delivery and Thermal Limits
Archi2 min read
Contents
The test bench#
We took a 2023 ultrabook with a 28W-rated processor and ran a 30-minute sustained load while logging package power, core temperature, and clock speed. Within 90 seconds, the chip dropped from 4.2 GHz to 2.6 GHz. The cause was not the CPU cooler alone. Power delivery impedance and thermal interface resistance combined to cap performance.
Root cause 1: power delivery impedance#
The voltage regulator on the mainboard has to supply current across a few milliohms of PCB copper and connector pins. At 20A, a 2 mΩ path drops 40 mV and dissipates 0.8W as heat. When the VRM temperature hits 105°C, it folds back current to protect itself. We measured a 15% drop in sustained package power after 5 minutes, matching the VRM thermal foldback threshold. The fix is not a bigger heatsink on the CPU; it is lower impedance paths and better VRM cooling.
Measuring the drop#
We soldered fine wires to the VRM output and captured voltage under load. The ripple increased from 8 mV to 22 mV as the VRM heated. That ripple forces the CPU to request a higher voltage to maintain stability, which increases power and heat. The loop feeds itself.
Root cause 2: thermal interface resistance#
We replaced the stock thermal paste with a liquid metal compound and saw a 6°C drop in core temperature at the same power. The stock paste had a thermal resistance of about 0.8 °C·cm²/W; the liquid metal measured 0.1. That difference matters because the heatsink was already near its limit. The bottleneck was the interface, not the fin area. Laptop makers use paste because it is cheaper and safer to apply at scale, but it costs sustained clocks.
Root cause 3: skin temperature limits#
The chassis cannot exceed 45°C on the bottom. We measured 43°C after 20 minutes. The embedded controller throttled the CPU to keep the surface below that limit, even though the die was at 85°C. This is a hard constraint from safety standards, not a design flaw. It explains why thin laptops throttle sooner than thick ones with the same chip.
What actually helps#
Our tests showed three changes with measurable impact: a copper heat spreader over the VRM (4°C drop in VRM temperature), liquid metal on the die (6°C drop in core temperature), and a 0.5 mm copper shim between the heatsink and chassis (2°C drop in skin temperature). None of these require a new CPU. They address the power and thermal paths directly.
The takeaway for buyers#
Sustained performance depends on the whole power delivery and thermal stack. A CPU with a higher TDP rating will not help if the VRM folds back or the chassis hits its skin limit. When comparing laptops, look for teardown photos that show VRM cooling and heat pipe coverage, not just the CPU model number. Our test unit lost 38% of its peak clock after 30 minutes; a thicker model with the same chip lost 12%. The difference was in the board and the chassis, not the silicon.
More articles