Immersion cooling explained: single-phase, two-phase, and the home-scale reality
General information only, not financial advice. Mining profitability depends on electricity prices, hardware costs and network conditions that change constantly; nothing here predicts earnings.
Immersion cooling is a thermal management method where mining hardware is fully submerged in a thermally conductive but electrically non-conductive — dielectric — fluid that absorbs heat directly from every component. That one sentence contains the whole appeal and the whole catch. The appeal: no fans, no dust, even temperatures. The catch: the heat does not disappear. It moves into a liquid, and the liquid then needs a engineered path out of the building. This guide separates the two.
What immersion actually changes
In a stock air-cooled miner, high-speed fans force air over heatsinks bonded to each hashboard. In immersion, those fans are removed or disabled entirely and the whole machine sits in a tank of dielectric fluid. Because the fluid contacts the chip packages and surrounding components far more uniformly than moving air can, hot spots shrink and thermal gradients across a board flatten out. The dust ingress that slowly kills air-cooled gear is eliminated, because there is no airflow to carry dust in.
What does not change is the arithmetic of heat. A miner converts essentially all of its electrical draw into heat, immersed or not. The fluid is a transport medium: heated fluid is circulated to a heat exchanger — typically a radiator or dry cooler — where the heat is rejected to the surrounding air, and cooled fluid returns to the tank. A home immersion setup that ignores the rejection half of the loop has built a very quiet way to heat a tank.
Single-phase vs two-phase: a comparison
| Single-phase | Two-phase | |
|---|---|---|
| How heat moves | Fluid stays liquid the whole time; pumps circulate it past the hardware and through a heat exchanger | Fluid boils directly against hot components; vapour rises and condenses on a cooled coil, then drips back |
| Fluids | Mineral-oil based and synthetic dielectric fluids, including products sold for mining | Engineered fluids with low boiling points around 49–61°C, such as 3M Novec 7100 |
| Complexity | The more common, lower-complexity route | Higher: sealed tanks, specialised and costly fluids |
| Typical scale | Home and small-scale through to commercial | Primarily hyperscale data centres; not practical for most home setups |
For the home miner and small-scale operator, the technical literature consistently lands in the same place: single-phase immersion with a quality synthetic dielectric fluid is the practical route, and two-phase belongs to facilities that can amortise its fluid cost and engineering.
The full added-parts inventory
Marketing photographs show a miner in a handsome tank. The working system behind the photograph includes: the tank itself, sized for the machine plus fluid displacement; the dielectric fluid, bought by volume and treated as a consumable asset to keep clean; a circulation pump (or a convection design that genuinely works at the intended load); plumbing and fittings rated for the fluid; a heat exchanger and dry cooler with its own fans — the noise has moved outdoors, not vanished; temperature monitoring, because a silent system fails silently; and handling equipment for the day a machine must come out, dripping, for service. The honest summary from the glossary literature is that tanks, pumps, plumbing, fluid and a heat-rejection loop add significant up-front cost and operational complexity compared with plugging a miner into a wall circuit.
Servicing deserves its own sentence. An air-cooled board swap is a screwdriver job. An immersed board swap is a lift, a drip-drain wait, fluid-safe handling, and a reseal — planned maintenance windows get longer, and spare fluid and absorbent materials become shelf stock.
What immersion does not guarantee
Stable, uniform temperatures create the option of running chips harder or more efficiently; they do not themselves create hashrate. Any gain comes from tuning choices made afterwards, traded against power draw and component stress, and it varies by machine and firmware. Promotional material that quotes a fixed percentage gain from immersion alone is selling the option as if it were the outcome. Immersion also does not repeal electricity pricing: the miner’s draw, plus pump and dry-cooler draw, still arrives on the same bill, and the economics still stand or fall on the electricity-cost maths covered elsewhere on this site.
Decide: a short framework
- Noise is the binding constraint (miner near living space, no outbuilding): immersion’s strongest case — it is the only option that removes fan noise at the machine. Price the full inventory above before deciding it beats relocating the miner.
- Dust or harsh air is the binding constraint (workshop, farm building): a real case; sealed fluid excludes dust entirely.
- Cost per unit of hashrate is the binding constraint: the weakest case at small scale, because the fixed system cost is spread over few machines.
- One machine, as a learning project: legitimate if entered as a project with a budget, not as an efficiency shortcut.
Running costs nobody photographs
The purchase price of a tank is the smallest number in immersion’s ledger, and a home operator should write the rest down before committing. The fluid itself is a working asset with a lifespan: it picks up dust and degradation products over time, needs periodic checking, and eventually needs filtering or replacement — at tank volumes, that is a budgeted event, not an errand. Pumps are moving parts in a system whose entire selling point is having no moving parts at the miner; a spare pump on the shelf is the difference between a maintenance task and an emergency shutdown. The dry cooler’s fans consume power and make noise at their location, so the electricity bill gains a cooling line item and the noise problem is relocated rather than abolished — worth it beside a living room, irrelevant beside a barn, and the deciding factor is where the cooler can actually go. Heat rejection also interacts with the building: a dry cooler dumping a miner’s full draw as heat needs airflow and clearance, and in winter that same heat is an asset some operators route indoors, which is a plumbing and controls project of its own. Finally, resale cuts both ways: an immersed machine’s history is visible in its condition, and buyers of used air-cooled machines may discount a unit that has lived in fluid, while the machine itself may be in better thermal health than an air-cooled twin. None of these are arguments against immersion. They are the difference between a costed decision and an expensive surprise, and the framework above only works when they are priced in alongside the tank.
One final discipline ties the economics together: log the system as a system. Record fluid top-ups, pump hours, cooler draw and any service event alongside the miner’s own figures, so the next decision — expand, hold or return to air — is made from a ledger rather than from impressions gathered beside a quiet tank.
FAQ
What fluid is used? A dielectric (electrically non-conductive) coolant — mineral-oil based or synthetic in single-phase systems; specialised low-boiling engineered fluids in two-phase systems.
Does the miner’s heat disappear? No. The fluid carries it to a heat exchanger or dry cooler, which rejects it to the air. The rejection loop must be sized like the miner itself.
Single or two-phase for a home setup? Single-phase, on the sources’ consistent assessment: lower complexity, cheaper fluids, and practical at small scale. Two-phase is primarily a hyperscale technology.