Why 72 Hours Matters for ASIC Thermal Review
Short burn-in tests miss the thermal creep that appears after heat sinks fully saturate. Here is what we see between hour 24 and hour 72.
When a new Antminer or Whatsminer arrives at a farm, the instinct is to plug it in, confirm hash rate, and rack it. Five minutes of stable output feels like confirmation. In our chamber, that confidence usually evaporates somewhere between hour eighteen and hour forty.
The Saturation Problem
Heat sinks and thermal interface materials do not reach steady-state temperature in minutes. A cold heat sink absorbs energy rapidly at first, keeping junction temperatures artificially low. As the sink saturates, the temperature gradient between the chip and ambient shrinks, and junction temps climb even if hash rate stays constant.
We see this pattern on nearly every air-cooled ASIC rated above 0.05 TH/s. The first six hours look fine. By hour twenty-four, chip three on hash board two is often 8–12°C hotter than the same probe at hour six.
What Changes Between Day One and Day Three
Fan curves shift. Firmware adjusts fan speed based on averaged temperature readings. A unit that ran fans at 60% on day one may hit 85% by day two as thermal mass builds.
Throttle events cluster. Most throttling we document occurs after hour thirty. A unit that never throttled in a two-hour bench test may throttle four to six times during a 72-hour run.
Power draw drifts. As fans work harder and VRMs heat up, total wattage can increase 3–5% even at constant hash rate. Efficiency calculations based on short tests underestimate real operating cost.
Why We Chose 72 Hours
Seventy-two hours captures two full diurnal cycles in our chamber plus enough post-saturation time to see whether temperatures plateau or continue climbing. Units that are still rising at hour seventy-two are flagged as unsuitable for continuous deployment without cooling modifications.
Shorter windows—24 or 48 hours—catch obvious failures but miss the slow drift that causes warranty disputes three months later.
Practical Takeaway for Operators
If you cannot send a unit to a review lab, run your own candidate hardware for at least 48 hours in your actual deployment environment before committing to a bulk order. Log chip temperatures, not just hash rate. The number that matters is junction temp at hour forty-eight, not hash rate at minute five.