Cooling

Thermal Protection: Throttle to Shutdown

The staged defense a miner runs before it faults, and the two numbers that pin down where the heat is.

The miner protects itself in stages

A modern ASIC does not just die when it gets hot. It steps down through a defense ladder, each stage buying the boards more time. Read the stage off the logs and dashboard before you touch anything - it tells you how much margin is left.

Stage	Trigger	What happens
Throttle	Ambient past ~40 °C	Frequency backs off (S21: 490 -> 440 MHz), fans ramp toward 100%
Over-temp fault	~95 °C chip temp	Bitmain logs 'ERROR_TEMP_TOO_HIGH: temp3 > 95 C, power off!'; Whatsminer raises its 300-360 codes
Protective shutdown	~47 °C ambient (S21)	Unit powers off to save the boards

Where the ceilings actually sit

Chip temp is what kills hashboards over time. Ambient is what you can control on the floor.

Longevity target: keep chip temps below ~85 °C. Above that you trade board life for hashrate.
Whatsminer ambient ceilings are model-dependent, not a flat number - M50S tolerates ~45 °C, the newer M60S wants ~35 °C.
A throttled unit that is still hashing is NOT healthy - it is stage one of the ladder. Find the heat before it climbs to a fault.

Do not clear a throttle or over-temp alarm and walk away because the miner is 'still hashing.' Fans pinned at 100% and a dropped frequency mean the machine is already defending itself. Fix the heat source or the next stage is a shutdown.

Delta-T and approach: the two diagnostic numbers

These two readings, taken together, tell you whether a hot unit is the machine, the flow, or the heat rejection.

DELTA-T is the temperature RISE the load puts across the miner or a stage. Air: exhaust minus intake. Hydro: return-coolant minus supply, ~10-20 °C across a hydro unit.
APPROACH is how close the cold side gets to the ultimate heat sink: supply-coolant minus ambient dry-bulb at the dry cooler.
High delta-T at one unit -> the machine (that unit is making more heat or moving less air/coolant through itself).
Low delta-T but the unit is hot -> the flow (air or coolant is not moving through it).
High approach -> the heat rejection. The dry cooler or facility cannot dump the heat, so everything downstream runs hot.

One hot miner in a cool row is a machine or flow problem. A whole row creeping up together with a high approach is heat rejection - stop chasing individual units and look at the dry cooler.

The miner protects itself in stages

Stage	Trigger	What happens
Throttle	Ambient past ~40 °C	Frequency backs off (S21: 490 -> 440 MHz), fans ramp toward 100%
Over-temp fault	~95 °C chip temp	Bitmain logs 'ERROR_TEMP_TOO_HIGH: temp3 > 95 C, power off!'; Whatsminer raises its 300-360 codes
Protective shutdown	~47 °C ambient (S21)	Unit powers off to save the boards

Where the ceilings actually sit

Chip temp is what kills hashboards over time. Ambient is what you can control on the floor.

Longevity target: keep chip temps below ~85 °C. Above that you trade board life for hashrate.

Whatsminer ambient ceilings are model-dependent, not a flat number - M50S tolerates ~45 °C, the newer M60S wants ~35 °C.

A throttled unit that is still hashing is NOT healthy - it is stage one of the ladder. Find the heat before it climbs to a fault.

Delta-T and approach: the two diagnostic numbers

These two readings, taken together, tell you whether a hot unit is the machine, the flow, or the heat rejection.

DELTA-T is the temperature RISE the load puts across the miner or a stage. Air: exhaust minus intake. Hydro: return-coolant minus supply, ~10-20 °C across a hydro unit.

APPROACH is how close the cold side gets to the ultimate heat sink: supply-coolant minus ambient dry-bulb at the dry cooler.

High delta-T at one unit -> the machine (that unit is making more heat or moving less air/coolant through itself).

Low delta-T but the unit is hot -> the flow (air or coolant is not moving through it).

High approach -> the heat rejection. The dry cooler or facility cannot dump the heat, so everything downstream runs hot.

One hot miner in a cool row is a machine or flow problem. A whole row creeping up together with a high approach is heat rejection - stop chasing individual units and look at the dry cooler.