Thermal Resistance

Thermal resistance quantifies how strongly a material, joint, or path opposes the flow of heat, exactly as electrical resistance opposes current. It is expressed in degrees Celsius per watt (°C/W): the number tells you how many degrees hotter one side of the path will sit above the other for each watt of heat forced through it. A junction dissipating 20 W across a total path of 1.5 °C/W will run 30 °C above ambient. Lower thermal resistance means heat escapes more easily, and the same silicon runs cooler at the same power — which in mining translates directly into stability, efficiency, and lifespan.

The junction-to-ambient chain

Heat leaving an ASIC crosses a series of stages, and thermal resistances in series add just like resistors. Engineers write the chain as junction-to-case (through the chip package), case-to-sink (through the thermal interface material), and sink-to-ambient (from the heatsink into moving air). Each stage contributes its own °C/W, and the total determines chip temperature for a given load. The interface layer's share depends on the material's conductivity, the contact area, and the bond-line thickness: a thin, well-spread, conductive layer over full contact area beats a thick blob every time, which is why application technique matters as much as the paste chosen — see thermal paste.

Reading the numbers like a miner

The series model makes miner thermals predictable. A hashboard dissipating on the order of a kilowatt spreads that load across dozens of chips, so each chip's local path might carry 10–40 W. Add even 0.5 °C/W of avoidable resistance per chip — dried paste, a warped sink, poor clamping pressure — and those chips run 5–20 °C hotter. Hotter chips are worse in three compounding ways: firmware throttles or drops frequency (an autotuner recalculates operating points at runtime as thermals degrade, so hashrate quietly sags), leakage current rises so efficiency in J/TH worsens, and long-term heat accelerates aging of solder joints and silicon alike.

What creeps up in the field

Diagnosing a hot miner is, in effect, finding which resistance in the chain has grown. The usual suspects: interface material that has pumped out or dried after years of heat cycling (case-to-sink rises), dust matting the fin stack or a dying fan (sink-to-ambient rises), and mechanical problems — stripped standoffs, a bowed board, uneven clamping — that shrink true contact area. Repeated expansion-contraction cycles also crack solder under chips, the failure mode covered under thermal fatigue; a chip whose thermal pad connection has degraded both runs hotter and reads erratically. Board-level symptoms follow a pattern: one hashboard reporting temperatures well above its siblings, temperature spread widening across a single board, or thermal shutdowns that begin in summer.

Why it pays to think in °C/W

You rarely get to measure °C/W directly in the field, but firmware telemetry gives workable proxies. Compare per-board and per-sensor temperatures at matched power: a board running persistently hotter than its siblings under identical load and airflow has extra resistance somewhere in its chain. Watch the spread between inlet air temperature and reported chip temperatures — a rising delta at constant wattage over months is the signature of degrading interfaces or clogging fins, distinct from a hot summer raising everything together. Logging those deltas turns thermal maintenance from guesswork into trend-following: you service the board whose curve is bending, before it starts throwing errors.

Every fix targets a specific link: fresh interface material attacks case-to-sink, cleaning fins and restoring airflow attacks sink-to-ambient, and immersion or hydro cooling replaces the weakest stages entirely. Thinking in resistances turns "it runs hot" into a measurable, attackable quantity — the same discipline a repair bench applies when triaging boards. If a machine is throttling and a paste-and-clean pass has not cured it, deeper board-level diagnosis is the next step: start a repair and let the chain be measured stage by stage.