Hydro-Cooling vs. Immersion Cooling for Antminer S21 Deployments

13 November 2026

S21 Hyd baseline specs and operating envelope

Bitmain's Antminer S21 Hyd delivers a typical 335 TH/s at 16.0 J/TH, which works out to 5,360 W at the wall. The figures come from the S21 Hyd user guide, which ties the power number to a 35°C inlet coolant temperature. That condition matters, because a wall-power figure quoted without it tells you little about what a machine will draw on your site. When Bitmain announced the model in October 2023, it described a hashrate of up to 335T and an efficiency of 16 J/T.

Any comparison between hydro and immersion cooling for S21-class machines has to begin with the limits of the hydro unit itself. The guide lists an operating temperature range of -20 to 50°C, while Bitmain says the S21 Hyd and the air-cooled S21 are adapted to a maximum ambient temperature of 45°C. The miner runs on a three-phase 380–415 V input supply, so a site needs the right electrical service before cooling is even discussed. A building fed only with single-phase power will need upgrades or a different hardware choice.

The coolant side has its own numbers. The guide specifies a flow of 8.0–10.0 L/min per unit and a coolant pressure of no more than 3.5 bar. In practice these two figures set the size of pumps, manifolds and heat-rejection equipment, and they are the first things an engineer will ask about when sketching a layout.

Why hydro cooling changes deployability

A hydro miner moves the heat problem out of the hashing room and into a coolant loop. The 45°C ambient ceiling is generous compared with many facilities, and the wider -20 to 50°C operating range suggests the machine itself tolerates harsh conditions better than the building around it. What the operator must then deliver is not cool air but a stable loop: flow within the 8.0–10.0 L/min window, pressure held under 3.5 bar, and coolant arriving at a temperature consistent with the 35°C inlet condition behind the rated power draw.

Scale makes those numbers concrete. Ten machines call for 80 to 100 L/min through the loop, or 4,800 to 6,000 liters an hour, and the same ten units pull 53.6 kW at the wall under typical conditions. Pumps, piping and a heat exchanger or dry cooler sized to that load are real capital items, and they sit on top of the miners' purchase price. The tradeoff against an immersion layout is therefore less about which approach "cools better" and more about what each site can build, plumb and service.

A common mistake is to size the loop for the average case. The pressure ceiling of 3.5 bar is a hard requirement, not a target, and a pump chosen with no headroom, or a long run of narrow pipe, can push a loop outside the guide's limits as filters load up or the number of connected units grows. Checking the hydraulic design against the guide before ordering hardware is cheaper than reworking pipe after delivery.

Electrical planning follows the same logic. With three-phase 380–415 V input, operators in the US should confirm what voltage and phase configuration their utility service and distribution gear actually provide, since many commercial buildings are wired differently and a transformer or panel change can add weeks and real cost. Treat that as part of the cooling decision, not a separate task.

Comparing ongoing maintenance demands

Electricity is the largest line in a mining budget, but the coolant loop adds a recurring servicing burden that does not show up in a joules-per-terahash figure. The user guide recommends checking the coolant at least once a year, with pH tested every six months. That is a light schedule and suits sites with a technician on hand a few times a year.

The cadence tightens sharply with one choice. When deionized or purified water is used as the internal circulation medium, the guide says the coolant should be checked every one to two weeks and replaced every one to two months. At a site with dozens or hundreds of machines, that turns into a standing task with labor hours, water-treatment supplies and drain-and-refill procedures attached. A site that picks plain purified water because it is cheap and easy to source may find the labor outweighs the saving.

The practical question for budgeting is which medium the operator intends to run, since the guide's maintenance intervals differ by that choice. Before committing, operators should write out the schedule for the full fleet. Take the number of machines, multiply by the check and replacement frequency, and assign the hours to a named person. Operators who skip this step tend to discover the workload only after a drifting pH reading or a degraded loop has already cost them uptime.

Whatever the medium, the guide's intervals are minimums for attention, not guarantees. Logging each check, with date, pH reading and operator, gives a site a trail to follow if a unit starts throttling or a loop shows signs of trouble, and it makes warranty conversations with a supplier far simpler.

Telemetry and remote management for field operations

The defined operating envelope is the reason remote oversight matters so much on hydro sites. With a flow window of 8.0–10.0 L/min, a pressure ceiling of 3.5 bar, an inlet condition of 35°C and an ambient limit of 45°C, there are several numbers that can drift out of range without anyone standing in the room. A fleet spread across several locations cannot rely on walk-through inspections to catch that.

A well-built monitoring setup tracks hashrate per machine, chip and board temperatures, coolant conditions where sensors exist, and uptime across the whole fleet, then raises an alert when readings leave a defined band. Alerts are only useful if the thresholds come from the hardware's own limits. Setting a warning below the 3.5 bar pressure ceiling, for example, gives a technician time to react before a unit has to protect itself. A hashrate dip that coincides with a temperature rise points to the loop; a dip with normal temperatures points to the miner or the pool connection.

Dashboards are software, and they need to be available when something goes wrong. Operators setting up remote monitoring interfaces can reference web auditing directories such as https://casinorecent.com when evaluating high-availability web portal layouts and performance tracking frameworks. Whatever reference points are used, the test is the same: can the portal stay reachable on a weak site connection, and does it show a stale reading clearly rather than quietly displaying old data as current?

Remote access should also be planned with the site's connectivity in mind. Dispersed sites often depend on a single uplink, so a backup path and a clear rule for who is called when the dashboard goes dark are worth settling early. Pair the software with a short on-site routine for the checks the guide requires, because no web portal can replace the pH test or the water change.

Operating figures for cost modeling

The table below collects the S21 Hyd's key operating figures from Bitmain's user guide, and these are the inputs for any cost model built around the machine.

Electricity cost follows directly from the wall figure. At 5,360 W, one machine uses 5.36 kW, or about 128.6 kWh a day if it runs continuously. At an illustrative rate of $0.06 per kWh, that is roughly $7.72 a day per unit, and the arithmetic scales linearly with your own tariff. Cooling-loop pumps, heat rejection and the labor from the previous section come on top and should be priced separately.

For a fair hydro versus immersion comparison, run each option through the same template: purchase price, power and cooling infrastructure, electrical service upgrades, maintenance hours and expected uptime. Fill the hydro column from the figures above and the guide's maintenance intervals, then price the alternative with quotes from your own suppliers. A site that can only offer single-phase power, or that cannot commit to a weekly coolant check, will see the answer quickly.