Heat is the number-one killer of mining hardware. A blocked filter, a failed fan or one hot afternoon can push chip temperatures past safe limits — and by the time you notice by hand, a hashboard may already be damaged. This guide covers what temperatures each Antminer generation actually tolerates, the three ways farms overheat in practice, and how to automate protection so a hot miner puts itself to sleep before it becomes a repair ticket.
Why temperature matters more than hashrate
A miner running a little slow costs you a little revenue. A miner running too hot costs you the machine. Most Antminers are rated for chip temperatures up to roughly 85–90 °C; sustained operation above that range degrades the silicon permanently, and the damage is cumulative — a summer of 90 °C afternoons shows up as failed chains in winter. Heat damage is also asymmetric: hashrate lost to a conservative work mode comes back with a click, while a cooked hashboard is a soldering-station job or a warranty argument. Every threshold in this guide follows from that asymmetry — err toward cool.
Safe operating temperatures by model
"Normal" depends on generation and cooling type. These are chip-temperature bands you want to see in a ~25 °C room, and the point where you should intervene:
| Model family | Healthy range | Investigate above | Alert / act at |
|---|---|---|---|
| S19 family (air) | 60–78 °C | 80 °C | 85 °C |
| S21 / S21 Pro (air) | 55–75 °C | 78 °C | 85 °C |
| T21 (air) | 60–78 °C | 80 °C | 85 °C |
| S21 Hyd. / S23 (hydro) | 45–65 °C | 70 °C | 80 °C + check loop |
| L7 / L9 (Scrypt, air) | 60–78 °C | 80 °C | 85 °C |
Two readings matter, not one: chip temperature tells you about the silicon, PCB temperature about the board around it. A widening gap between them usually means airflow across the heatsinks has degraded — dust, a slowing fan — even while absolute numbers still look acceptable. Hydro units are their own case: chip temperature barely moves until coolant flow fails, and then it moves fast, so loop health (inlet/outlet spread, pump behaviour) is the real early signal.
The three ways farms actually overheat
1. Airflow decay. The slow one. Dust loads filters, intake screens clog, hot-aisle recirculation creeps in as seals age. Chip temps drift up a degree a week until one warm day crosses the line. The tell: efficiency (J/TH) worsening across many units at once while ambient feels normal.
2. Fan failure. The fast one. An Antminer that loses a fan either throttles hard or stops with a fan error; the neighbouring units then inherit its heat. Fans are consumables — three to five years of continuous duty — and on fleets older than that (most S19s by now), failures come in clusters.
3. Ambient spike. The calendar one. The first heat wave of summer, an HVAC failure in a container, a ventilation exhaust blocked by snow, ice or a parked truck. Every unit in the room climbs together — which is exactly when per-miner babysitting fails and automation earns its keep.
Automate the response: alert + auto-sleep
The fix is not watching a dashboard all day; it is making the platform act while you sleep. In Alerts, two rules cover heat end to end. First, a high-temperature alert at 85 °C chip temp (80 °C for hydro): you get an email naming the miner the moment any unit crosses the line, with a 30-minute per-miner cooldown so one hot afternoon does not flood your inbox. Second — the one that saves hardware — add an auto-sleep action: when a unit crosses the threshold, AntminerTools switches it to sleep mode over the same tunnel, power drops to a trickle, and the silicon stops cooking within seconds. A slept miner is earning nothing, but it is also not dying; you wake it remotely once you have dealt with the cause. Alerts auto-resolve when temperatures recover, so the incident record stays truthful without housekeeping.
If you run the hashrate-zero auto-restart rule as well (recommended — see the remote monitoring guide), the two compose sensibly: a rebooted miner gets a ten-minute grace before hashrate judgement, so a heat-slept unit waking up does not trigger a false restart.
The summer checklist
Before the first hot week: wash or replace intake filters and check every fan spins up to speed (the dashboard shows all fan RPMs per unit — sort by lowest). Verify hot-air exhaust actually leaves the building rather than looping back. Confirm your high-temp rules are enabled and the alert email is one you read on your phone. Then, during heat waves, watch the fleet's hottest-unit tile rather than individual miners — it tells you when the room is losing, not just one machine. If ambient regularly pushes units into the 80s, consider scheduled sleep for the peak hours: four slept hours a day costs ~17 % of revenue and can be the difference between a fleet that lasts five years and one that lasts two.
First five minutes after a temperature alert
One unit hot, neighbours normal: it is that machine — check its fans and filters, reboot if a fan reads zero RPM, sleep it if it keeps climbing. Several units in one rack: airflow problem in that row — look for a blocked intake or a failed exhaust fan behind them. Everything climbing together: ambient event — ventilation, HVAC, weather — and the move is bulk sleep on the hottest units while you fix the room, not heroics per miner. The per-chain temperature view separates a failing hashboard (one chain hot) from a failing environment (all chains hot) in one glance.
Container, garage and bedroom-wall farms
Where the miners live changes the heat math. Containers concentrate risk in the ventilation system: one failed exhaust fan or a filter wall loading up with cottonwood fluff can take a 40-unit container from healthy to critical in under an hour, so alert coverage on every unit is not optional — the fleet-wide pattern is the ventilation monitor. Garages and sheds ride the weather; the same S19 that idles at 68 °C in April will flirt with 85 °C in an August afternoon, and scheduled sleep for peak hours is usually cheaper than air-conditioning a space that leaks like a shed. Home setups add a constraint the datasheet ignores: the intake air you give a miner is the exhaust air of the room it is in. A closed 10 m² room gains heat from a single S21-class unit far faster than most people expect — duct the exhaust out, or the room becomes the reason for the alert.
Fans: the consumable everyone forgets
Every temperature problem that is not weather eventually turns out to be a fan. Antminer fans run at full duty around the clock, and their bearings age accordingly — the failure curve steepens sharply after year three. The dashboard's fan-speed readings make the pre-failure pattern visible: a fan that used to hold its target RPM and now oscillates or reads consistently low is announcing its retirement weeks in advance. Keep two spare fans per twenty air-cooled units, replace on the first symptom rather than the final failure, and remember that a miner running with one weak fan silently overworks the remaining ones — fan failures cluster for a reason. After any fan swap, watch the unit's chip-to-PCB temperature gap for a day; if the gap does not shrink, the fan was a symptom and the heatsink airflow path (dust mat, blocked fins) was the disease.
Frequently asked questions
Is 80 °C chip temperature dangerous? Not acutely — but it leaves no headroom. Treat it as the investigate line: find out why the unit runs hotter than its peers before summer answers for you.
Why does my hydro S21 show 0 °C in some tools? Hydro units report chip temperature in a different sensor field than air-cooled models. AntminerTools reads the correct sensor automatically; a tool showing 0 °C is reading the wrong one.
Auto-sleep or auto-restart for heat? Sleep. A restart puts the miner right back to full power in a hot room; sleep removes the heat source and waits for you.
Do I lose the alert history when things recover? No — alerts auto-resolve but stay logged, so you can spot patterns like the same rack alerting every hot afternoon.
What about immersion systems? Immersion changes the numbers (coolant temperature governs) but not the logic: alert on the tank, sleep units when the loop fails, and watch efficiency for early drift.
Can I run hotter in winter to heat a space? Cold intake air buys you real headroom — units that struggle in July often sit 15 °C cooler in January — and mining-as-heating works. Keep the same 85 °C alert though: a blocked filter does not care what month it is, and the rule costs nothing while it is quiet.
Set it up once, stop worrying
Create a free account, connect your farm with one command — no port forwarding, as covered in the remote monitoring guide — and add the two heat rules in under a minute. The next failed fan becomes an email and an automatic nap instead of a dead hashboard.