ASIC Immersion Cooling: Design the Whole Heat-Removal Loop

Decide from constraints, not from the tank

Immersion is usually a good design answer only when it solves a measured constraint. Write down the problem first: occupied-space fan noise, dust, hot ambient air, rack density, stable thermal control, or a useful liquid-heat destination.

Before committing, compare a duct or shroud system, a purpose-built Bitcoin heater, and the broader home-mining operating plan. For a one- or two-miner immersion build sequence, use the detailed home immersion guide. This page stays at the system boundary: it helps decide what must be designed and verified before any step-by-step build is appropriate.

Follow the heat through every boundary

In single-phase immersion the dielectric liquid remains liquid. Natural or forced convection carries heat away from electronics; a documented rejection interface then transfers that heat to another loop or sink. Two-phase systems use a dielectric fluid and thermal/safety design intended to boil at the operating condition and condense. Fluid chemistry, tanks, controls, exposure limits and service methods require exact qualification and are not transferable from single-phase systems.

1. 1. ASIC loadNearly all electrical input ultimately becomes heat inside the chosen site boundary.
2. 2. Dielectric bathCompatible fluid contacts the approved immersed components.
3. 3. Primary circulationPump or convection moves hot fluid through a controlled path.
4. 4. Rejection interfaceA heat exchanger or documented interface transfers heat to an approved secondary fluid or sink.
5. 5. Final heat sinkDry cooler, radiator, building loop, or useful-heat load rejects the energy.

The Open Compute Project (OCP) immersion requirements define single-phase immersion as dielectric liquid circulating across IT equipment and a heat-exchange approach without phase change. They also distinguish the technology cooling system from the facility-water system. A heat exchanger or documented rejection interface may transfer heat to an approved secondary fluid, ambient-air device, building loop or useful-heat load. Do not put a secondary-loop fluid into the dielectric bath unless the exact system explicitly defines it there.

Draw the real loop. Add tank volume and freeboard, expansion behavior, inlet/outlet orientation, pump curve, pressure drop, bypass paths, strainers specified by the system/fluid provider, drain and sample points, secondary-loop water quality, dry-cooler conditions, make-up strategy, containment, service lift/drip area, and every sensor and shutdown action. A parts list without these relationships is not a design.

Air, immersion, and hydro are different interfaces

Use the airflow calculator for an air-cooled room and the hosting guide when facility responsibility is the actual constraint. Cooling terminology is not interchangeable, even when product names look similar.

Size from measured heat, fluid properties, and design temperatures

There is no universal flow rate or bath temperature. The equations below are a steady-state, single-phase sensible-heat check, not a pump prescription. They cannot be copied into a two-phase boiling/condenser calculation. They may help check a hydro loop’s energy balance, but never replace its exact cold-plate coolant, pressure, flow and water-quality requirements.

Worked boundary, not a universal setting: a 4,000 W primary-loop duty, documented fluid specific heat of 2,000 J/kg·K, density of 800 kg/m³, and a 5 K design rise imply 0.40 kg/s or 30 L/min by energy balance. Do not add PSU, pump or control watts already included by the selected meter. The result does not prove that 30 L/min reaches every chip or that a selected pump can overcome tank, hose, heat-exchanger, filter, and fitting losses.

As an example of model specificity, the official S21 Imm. specifications and March 2025 user guide publish a 30-55°C inlet-oil range and 1.0 m³/h oil-flow requirement, mark those operating parameters as mandatory, and require checking flow and inlet temperature before power-on. Those values belong to that exact product and operating mode; they are not a retrofit rule for an S19, another S21 variant, or a mixed tank.

• Use wall measurements from the intended firmware profile.
• Add only separate auxiliary heat deposited inside the design boundary.
• Use fluid properties at the intended operating temperature.
• Choose and document the allowed inlet and outlet temperatures.
• Use a heat-exchanger performance curve at the real flow rates.
• Rate the final heat sink at worst-case ambient and approach temperature.
• Check pump duty at required flow and total dynamic head.
• Define the response to pump, fan, sensor and utility failure.

Credit a useful-heat load only for duty it is guaranteed to accept at the design condition; otherwise provide a sized bypass or backup heat sink. Use the heat-reuse decision workbook and miner heat-reuse hub to separate seasonal heat value from cooling capacity. Use measured profiles from the ASIC power-profile database rather than a family-level wattage assumption.

Choose fluid from a testable evidence package

“Nonconductive” is not a complete fluid specification. The OCP base specification for immersion fluids groups the evidence into thermal performance, signal integrity, material compatibility and reliability/serviceability, and environmental impact. Many properties vary with temperature, so one room-temperature number is not enough.

Canadian workplaces should evaluate the current supplier SDS and applicable WHMIS duties. Health Canada’s WHMIS guidance describes the SDS as the hazard/precaution/first-aid evidence package and requires bilingual SDS information for regulated hazardous workplace products. Home installations still need the supplier’s SDS and local fire, environmental, electrical, building, and disposal requirements even when a workplace rule does not apply.

Use the immersion-fluid reference to compare candidate data, but treat missing or vendor-claimed fields as gaps to resolve rather than blanks to fill with another product’s values.

Compatibility starts with the exact miner variant

A family name does not identify the approved heat path. Air, hydro, and immersion variants can use different PSU, chassis, connectors, firmware, operating voltages, orientation, and flow requirements.

Never infer whether the PSU belongs in the bath. Some purpose-built systems are engineered for complete immersion; some conversions keep the PSU outside; others approve a specific PSU/fluid arrangement. The answer comes from the exact server, PSU, tank, and fluid evidence package. Use the ASIC miner database, comparison tool, and firmware selector to lock model identity and control-board support before requesting conversion guidance.

For every miner revision + PSU + cable/connector + fluid + tank combination, record one evidence state: verified by exact provider documentation; tested with limits under a named method, duration, temperature and acceptance boundary; or unverified. Similar materials, related model names and short demonstrations do not promote a combination to verified.

OCP’s material-compatibility guidance and warranty guidance show why this record matters: system variation, fluid choice, deployment method, material compatibility, thermal effects, and signal integrity all influence risk. These documents do not approve an ASIC retrofit or exact pairing. Ask the seller and integrator to state coverage in writing for the exact asset bundle rather than relying on a generic claim that immersion extends hardware life.

Fluid changes cooling; it does not neutralize electrical or fire risk

The ASIC and power distribution remain energized equipment. Hot fluid and surfaces can cause thermal burns. Some dielectric fluids are combustible and others are not; use the exact SDS plus flash point, fire point, autoignition, decomposition and ventilation data. Flowing dielectric fluid may require static-charge assessment, spills create exposure and slip hazards, and a stopped heat loop will eventually overheat.

The OCP immersion requirements call for location-appropriate certification, documented electrical assemblies, regional grounding, static-charge assessment, written fire/spill procedures, fluid documentation, operator training, monitored control criteria, and containment for at least 100% of any single system, including the largest interconnected container, subject to local requirements. For production systems, instrument the boundary instead of trusting a single dashboard value:

• Dielectric-fluid inlet and outlet temperature.
• Secondary-loop supply and return temperature.
• Primary and secondary flow or verified pump state.
• Tank level and leak/containment state.
• Pressure where a closed or pumped loop requires it.
• ASIC wall power and pool-side accepted hashrate.
• Pump, dry-cooler/fan and control-panel health.
• Warning, critical alarm and controlled power-off actions.

A shutdown strategy should not depend on one unreliable sensor. Use plausibility checks and the system provider’s control design: for example, pump state plus flow plus temperature trend, with tested alarm escalation and a reachable all-power disconnect. Thermal mass may delay overheating after a cooling failure; it is not proof that the system can safely ride through one.

Commission a measured baseline before tuning

1. Freeze the evidence package. Record every miner/PSU serial and variant, firmware, tank and pump model, heat-exchanger curve, fluid product/batch, TDS/SDS, compatibility statements, electrical drawing, design temperatures, flow, load and control setpoints.
2. Inspect and clean to the approved process. Remove only materials the conversion procedure identifies. Keep water, incompatible cleaners, fibres, metal debris and unknown prior fluid out of the system.
3. Prove mechanical integrity without energized miners. Check support and floor load, containment, valves, hose/fitting retention, flow direction, drain/sample points and leaks through the intended operating range.
4. Establish flow and heat rejection. Verify pump direction, measured flow, sensor agreement, air removal where applicable, secondary-loop operation and the final heat sink before energizing the ASIC.
5. Start one known unit conservatively. Use an approved stock or low-risk profile. Trend inlet/outlet fluid temperature, chip/board readings, wall power, local hashrate, accepted shares and abnormal logs until steady state.
6. Load in stages. Add one unit or controlled load step at a time and confirm temperatures, flow distribution, pump duty, heat-exchanger approach and heat-sink capacity after each change.
7. Test the failure response. Use the system provider’s safe test method to prove high-temperature, low-flow, pump/fan fault, leak/level, network and power-loss alarms plus the controlled shutdown and restart sequence.
8. Save the baseline. Keep measurements, fluid sample/batch, photos, firmware/configuration, alarm results and acceptance limits so future drift has a known comparison.

Tuning is a separate experiment. Change one variable at a time and record wall watts, accepted hashrate, inlet/outlet fluid temperature, chip/board telemetry, ambient temperature, flow and stability duration. A higher local hashrate without pool-accepted work and thermal margin is not a successful cooling result.

Maintain fluid, heat rejection, and mining performance together

A clean-looking bath can still have changed electrical, thermal, or chemical properties. Set intervals from the exact fluid and equipment providers, operating hours, temperature, sampling history, contamination events, and warranty requirements.

The official S21 Imm. guide, for that exact model, calls for oil tracking at least annually, TAN testing every six months, and regular TAN, conductivity and related-index records. Its English TAN threshold sentence appears directionally inconsistent with the OCP explanation that rising TAN can indicate breakdown, and the guide does not publish a conductivity acceptance limit. Do not operationalize that sentence or silently substitute a generic OCP limit: obtain written clarification and exact limits from the OEM and fluid provider.

Before sending wet hardware for service, ask the repair provider for its intake process. Record the fluid, exposure duration, failure state and cleaning performed; drain and package the unit so residual liquid cannot escape in transit. Start with the ASIC symptom router, the troubleshooting library, and D-Central’s repair intake. Use the repair-cost estimator and repair-vs-replace guide when cleaning, downtime, warranty and requalification change the economics.

Move from decision to the right implementation path

Sources, review method, and limits