Cooling: Air, Hydro & Immersion

The three cooling methods at a glance

Every ASIC turns almost all the power it draws into heat. The whole job of a cooling system is to move that heat off the chips fast enough that the boards never throttle or cook. There are three ways the industry does it, and they behave very differently at the machine.

Air cooling: fans, heatsinks, and airflow

Air is still the most common method in the world's mining farms. Each machine pulls cool air across aluminum heatsinks bonded to the hash chips and blows the heated air out the back. The building's job is to feed every intake with cool air and get the hot exhaust out without letting the two mix.

• Each unit has its own fans (usually two, sometimes four) that force air over the heatsinks; the firmware ramps fan speed against onboard temperature sensors.
• Hot aisle / cold aisle containment is the core layout: racks face each other so cold intake air and hot exhaust never blend. It's the cheapest efficiency upgrade a site can make.
• Airflow is run either horizontally (racks along the wall, cold plenum behind) or vertically. Either way, the goal is positive cold-side pressure and a sealed hot side.
• Temperature sensors live on the hashboard (PCB temp) and, on some models, inside the chip die. An air-outlet sensor crossing the firmware limit throws an over-temperature warning.

Air cooling: dust is the enemy

In an air-cooled machine the fans double as an unfiltered vacuum, pulling every bit of dust in the room through the heatsinks and mesh. A dust blanket on the fins acts like insulation, temps climb, fans spin up to compensate, and the machine either throttles or fails. Dust and heat, not age, kill most air-cooled boards.

• Site air intakes should have washable pre-filters (furnace-grade or dedicated ASIC screens); swap or clean them on a monthly cadence.
• Use the approved cleaning method after the required isolation and cooling. Compressed-air cleaning requires pressure regulation, effective chip guarding and PPE under the site procedure; follow equipment-specific fan and ESD precautions.
• A fan-speed error means the tach reads a fan below its minimum RPM (often around 2,400 RPM depending on model), so the firmware halts hashing to protect the chips.
• Sustained overheating carbonizes thermal paste and fatigues the BGA solder joints under the chips, damage that needs a rework station, not a field fix.

Hydro (direct water) cooling: how the loop works

A hydro miner deletes the fans entirely. Instead, a cold plate (water block) is clamped directly onto the hash chips, and facility water is pumped through it to carry heat away. The heat rides the water out to a CDU and dry cooler on the roof or pad, dumps into the ambient air, and the cooled water comes back around. Because the water touches a sealed plate and never the electronics, it's a closed loop.

• Cold water enters a supply manifold at the head of the rack, splits to each miner's cold plate, absorbs heat, and returns through a hot manifold.
• The CDU (Coolant Distribution Unit) is the pump-and-heat-exchange heart of the loop; it pushes coolant through the miners and hands the heat to the dry cooler. Bitmain's own CDU line runs roughly 50 kW to 400 kW per unit.
• You generally need a CDU once you're running three or more hydro miners; below that a single external radiator can carry the load.
• Miners connect with factory quick-disconnect couplers so a unit can be pulled without draining the whole facility loop.
• Coolant is treated water: deionized/distilled water plus a corrosion inhibitor, often with antifreeze added for cold sites. Conductivity gets monitored and the loop gets topped off for evaporation.

Hydro: flow, temperature, and pressure limits per maker

The numbers below are the guardrails each maker publishes. Miss the minimum flow or run the inlet too hot and the machine will down-clock or stop. Every maker differs, so read the manual for the exact unit, but these are representative.

Hydro install and service review

Use the exact model and facility procedure. This checklist is a review for authorized personnel, not a universal installation sequence.

1. Confirm authorization, the equipment procedure, approved coolant and lifting arrangements before work.
2. Before connecting, opening, bleeding or removing fittings, isolate the relevant electrical and fluid sources, control stored pressure, allow cooling, and verify safe conditions under that procedure.
3. Authorized personnel complete the specified connections and checks. Use the manufacturer's fluid-specific chemistry, pressure and flow limits; do not substitute a generic coolant recipe.
4. Before return to service, complete the required leak, flow and air-removal checks under the approved commissioning procedure; clear personnel and obtain restart clearance before enabling hashing.
5. For removal, repeat the isolation, depressurization and cooling procedure before disconnecting couplers; drain, contain and store fluid as specified by the manufacturer and SDS.

Immersion cooling: single-phase vs two-phase

Immersion drops the whole board, fans and heatsinks stripped off, straight into a tank of dielectric (non-conductive) fluid. The fluid touches the chips directly and carries heat away far better than air. There are two flavors, and they behave very differently.

Coolants and dielectric fluids you'll handle

Different systems use very different fluids, and how you handle them changes with each. The one rule that never changes: immersion fluid is dielectric (won't conduct), while hydro water absolutely will, which is exactly why hydro water stays sealed behind a cold plate and never touches a board.

Zero hashrate: check cooling evidence before diagnosing boards

This trips up techs coming off air-cooled fleets, so burn it in: hydro and immersion units have no fans. Don't write up a missing-fan fault or go looking for a fan header, there isn't one. Heat leaves through water or fluid, period.

• The flip side is a hard safety interlock: if flow drops or the coolant is out of range, the firmware halts hashing to protect the boards.
• A protection stop may produce zero hashrate; use corroborating alarm and log evidence rather than assuming the boards are healthy.
• On hydro, a common cause is coolant that's too cold or too low to start (the log will say something like 'too low water temperature'); the fix is restoring flow and temperature, not swapping boards.
• Review cooling telemetry and logs early when all boards read zero. External observation is distinct from fitting or bleeding work, which requires authorized isolation and pressure/heat controls.

Leak response and PPE

Liquid cooling means liquid can leak, and the two systems fail differently. Hydro leaks are water under pressure near live electronics; immersion leaks are fluid loss that can eventually run a tank dry. Know the response before it happens.

1. Spot it early: run leak sensors and a secondary containment tray under tanks and CDUs so a drip is caught and alarmed, not discovered later.
2. Keep clear of liquid near potentially energized equipment and report the leak. Authorized personnel must establish electrical, pressure and thermal safety before fitting work or cleanup; follow the fluid SDS.
3. Contain and absorb with pads kept near the equipment; keep spare fittings and couplers on hand for a fast reseat.
4. Wear the PPE: gloves, safety glasses, and closed-toe shoes when handling coolant or fluid; avoid skin and eye contact and follow the fluid's safety data sheet.
5. After any fluid work, dry and inspect every fitting and connector before you re-energize, and log the incident so a slow recurring leak gets caught.