It's underappreciated how much thermal capacity varies from machine to machine, and over time. It makes benchmarking hard.<p>Clock speeds, because they're crystal-controlled, don't vary by more than several PPM between machines. Thermal capacity varies by ±10% easily.<p>The thermal resistance depends on a bunch of mechanical factors that are hard to control in manufacturing, like how well the thermal grease fills the tiny pores in the metal of the heat sinks, and how well the air flows through the ventilation ducts. And these usually get worse over time as bubbles form in the grease and dust sticks to the fan blades. And of course it depends on room temperature and pressure.<p>It means that benchmarks will have to be done under more tightly controlled conditions. The automotive world has standards for this, where horsepower measurements are adjusted by a correction factor that depends on temperature and pressure.