Why room temperature does not describe the chip.
Between a workload and room temperature lie materials, thermal interfaces and time. That path matters when you want precise control.
Heat starts locally.
Electrical power in a working processor becomes predominantly heat. That heat must leave the chip through its package, across interfaces and finally into air or liquid. Every part of this path affects how temperature evolves.
A room sensor describes the temperature at its location. It does not directly measure the GPU die or its thermal margin to a manufacturer-defined limit. GPU telemetry can provide additional information; the available values depend on the device.
Time belongs in the model.
Materials store heat. A load change therefore does not reach every measurement point at the same time or with the same magnitude. A temperature reading is a snapshot; a thermal model also describes the relationships between source, storage and heat removal.
The simplest model is an energy balance: change in stored heat = input power − heat removal. Real systems require multiple coupled elements and comparison against measurements.
More context for control.
The idea behind arflo is to consider load, local temperatures and cooling state together. Room temperatures remain useful, for example as environmental context. They are part of the system picture. The accuracy of any model must be established through validation on the relevant system.
Further technical reading
NVIDIA · System Management Interface: temperature and power telemetry ↗