Updated 2026-10-08

Glossary of liquid cooling terms

Definitions are self-contained — written to be quotable. Terms alphabetized; each term is anchorable (/glossary#term-slug).

TermDefinition
1.1× ruleThe room-cooling sizing rule for liquid-to-air cabinets: room AC capacity ≈ 1.1 × IT load, the extra 10% covering pumps, losses and commissioning margin.
Adiabatic assistBoosting a dry cooler in hot weather by pre-cooling incoming air with water evaporation — recovers capacity at the cost of water consumption and treatment.
Approach temperatureThe temperature difference between the coolant and the air (or facility water) rejecting its heat. Liquid-to-air capacity scales with this difference — smaller approach means less cooling capacity.
ASHRAE W-classesCoolant temperature classes for liquid-cooled IT: W17 through W45, where the number is the maximum facility supply temperature in °C. Higher classes enable chiller-free or high-efficiency operation if equipment supports them.
Blind-mateA connection design where components engage automatically as they slide together — servers mate to the rack manifold by being pushed in, with no hand-plugged couplings. Specified for racks by OCP’s ORv3 blind-mate manifold specification.
CDUCoolant Distribution Unit — the component that pumps the server coolant loop and transfers its heat to a final sink (facility water or room air). Contains pumps, heat exchanger, filter and controls. Comes in in-rack, in-row and facility sizes.
Capture ratioThe share of server heat a cooling architecture removes through liquid: typically 60-80% for direct-to-chip, near 100% for immersion. Drives flow, CDU capacity and the residual air load.
Cold plateA metal block with internal fluid channels bolted onto a CPU or GPU so heat conducts directly into the circulating coolant. The defining component of direct-to-chip cooling.
CommissioningThe staged acceptance sequence for a liquid loop: dry pressure test, fill and purge, flow and ΔT verification against design, pressure survey, dew-point check, leak-system test, load ramp, baseline snapshot.
Condensation marginThe gap between the coolest wetted surface (coolant supply) and the room dew point. Negative margin means water forms on live hardware — the reason coolant temperature has a floor, not just a ceiling.
COPCoefficient of Performance — cooling delivered divided by electricity consumed by the refrigeration unit. A COP-3 chiller moves 3 kW of heat per 1 kW drawn; the bridge between cooling tons and electrical loads.
Cooling towerA heat rejection device that evaporates water to chill its loop toward the ambient wet-bulb temperature. Efficient in dry climates; output floors at wet-bulb + roughly 5-8 K of tower approach.
DeratingThe reduction of a cooling system’s capacity as operating conditions worsen — most importantly rising ambient temperature. Capacity-vs-ambient curves state it; very few vendors publish them.
Dew pointThe temperature at which air becomes saturated and water condenses. In liquid cooling it bounds how cold coolant supply may run without sweating pipes and plates in a humid room.
DLCDirect Liquid Cooling — liquid cooling where coolant touches a plate on the chip (as opposed to immersing whole servers). The mainstream architecture; captures 60-80% of server heat.
Dry coolerA heat rejection device that exchanges heat with ambient air without evaporating water. Output floors at dry-bulb + roughly 8-12 K of approach; no water use, worse performance in hot climates.
ΔT (delta-T)A temperature difference. In loop design, the supply-return coolant difference that sets flow for a given heat load (flow = kW × 60 / (4.186 × ΔT)); in rooms, the air temperature rise across equipment.
Fluid analysisPeriodic coolant sampling for pH, inhibitor level, conductivity and biological growth — the early-warning system for corrosion and biofouling in water-glycol loops, typically semi-annual.
Floor loadingStructural pressure under equipment: filled mass × 9.81 ÷ footprint, in kN/m². Office floors typically design for 2.5-3.5; computer-room and industrial slabs for 7-12+. Concentrated, not averaged, load governs.
Free coolingRejecting data center heat without mechanical refrigeration, using ambient air (dry coolers) or evaporation (towers). Viable hours are set by the site’s dry-bulb/wet-bulb statistics versus the required coolant temperature.
Friction factor (Darcy)The dimensionless resistance coefficient in the Darcy–Weisbach pressure-drop equation — about 0.03-0.04 for smooth hose at typical coolant flows. Enters as f × (L/D) × ρv²/2.
FWSFacility Water System — the building’s water loop that ultimately carries heat away. The CDU isolates the clean server loop (TCS) from it.
Heat capture splitHow a rack’s heat divides between liquid and air: P_liquid = capture × P_IT. The air share (20-40% in DLC) still needs cabinet airflow and room AC.
Heat exchangerA device transferring heat between two fluids without mixing them — the core of every CDU (technology loop ↔ facility side) and of dry coolers.
Immersion coolingSubmerging entire servers in a dielectric fluid (oil or fluorinated chemistry). Captures nearly all heat but changes hardware, service and fluid management compared to direct-to-chip.
InhibitorsChemical additives in water-glycol coolant that suppress corrosion and biological growth. Their depletion — tracked by fluid analysis — is the main driver of fluid replacement intervals.
kN/m²Kilonewtons per square meter — the unit of structural floor load. 1 kN/m² ≈ 102 kg/m². A 1,200 kg rack on a 600 × 1,200 mm footprint prints about 16.4 kN/m².
kW vs kVAReal power vs apparent power in AC systems; they differ by the power factor (< 1). Cooling and IT budgets are kW conversations; UPS, feeds and switchgear are kVA — convert explicitly, early.
L2A / L2LLiquid-to-Air vs Liquid-to-Liquid — how a CDU rejects heat. L2A blows it into the room (self-contained, capacity derates with room temperature); L2L exchanges with facility water (higher capacity, needs building plumbing).
ManifoldThe rack-level supply/return spine distributing coolant to each server position. Good manifolds are blind-mate so servers connect on insertion.
Micro data centerA self-contained compute deployment (one cabinet to a small row) in a room not built as a data center — offices, clinics, wiring closets, light-industrial spaces.
N+1Redundancy scheme with one spare unit beyond the required count — e.g., two pumps where one suffices. In cooling, apply it to pumps, fans and AC units; the failure question belongs in every RFQ.
Nameplate capacityThe rating on the brochure, quoted at standard conditions (typically 25 °C air / 45 °C coolant). Actual capacity at your room temperature is lower — demand the derating curve.
OCPOpen Compute Project — the industry body whose specifications (UQD couplings, ORv3 racks and manifolds) make liquid cooling parts interoperable across vendors.
ORv3Open Rack v3 — OCP’s current rack standard family, including the blind-mate manifold specification (Rev 1.0, 2024) that lets compliant servers and racks interoperate.
PFASPer- and polyfluoroalkyl substances — the regulated chemical family behind fluorinated coolants. Global tightening and 3M’s 2025 manufacturing exit pushed buyers toward water-glycol and defined the modern fluid decision.
PG25A coolant of 75% water and 25% propylene glycol — the default fluid for direct-to-chip systems. Non-flammable, non-hazardous to ship, PFAS-free; needs corrosion inhibitors and periodic checks.
Pressure drop (ΔP)The head a loop consumes as fluid moves: friction in hoses (Darcy–Weisbach), fittings and couplings, in kPa. If total ΔP exceeds the CDU’s available head, operating flow — and capacity — fall.
PUEPower Usage Effectiveness — total facility energy ÷ IT energy. Air-cooled sites typically run ~1.5; liquid-cooled data centers reach 1.1-1.2. Assumes facility-scale cooling; a single cabinet in a room depends on that room’s AC.
Pump headThe pressure a CDU’s pump can supply, in kPa, as a function of flow. The loop’s resistance curve intersecting the pump curve sets real operating flow — read the intersection, not the peak.
RDHxRear-Door Heat Exchanger — a water-cooled door replacing a rack’s rear panel, capturing exhaust air heat into a liquid loop without touching the servers inside. An air-to-liquid retrofit technology.
Refrigeration tonA cooling unit: 1 ton = 3.517 kW = 12,000 BTU/h — historically the heat to melt one short ton of ice per day. Building HVAC still quotes in tons; divide kW by 3.517 to speak it.
Specific heatHeat per kilogram per degree — water’s 4.186 kJ/(kg·K) is the constant behind the flow formula. Glycol lowers it ~10%, which is why pure-water math slightly undersizes PG25 loops.
SS 726:2026Singapore’s national standard for liquid cooling in tropical data centers — the first of its kind. Covers direct-to-chip systems inside data centers; excludes indirect approaches and standalone edge cabinets.
TCSTechnology Cooling System — the clean, closed coolant loop circulating through server cold plates, kept separate from facility water by the CDU.
TDPThermal Design Power — a chip’s worst-case heat output rating, in watts. The unit of rack power budgeting: sum GPUs and CPUs, add ~20% overhead, multiply by node count.
UQDUniversal Quick Disconnect — OCP-standardized dripless coolant couplings. Compliance means different vendors’ parts connect and seal; the basis of open, mix-and-match liquid racks.
UQD sizesThe UQD family from UQD04 to UQD40, where the number ≈ nominal flow in L/min. Single plates run UQD04-06, accelerator nodes 08-12, rack and CDU headers 32-40.
Velocity (coolant)Flow speed in m/s = flow ÷ pipe area. Enters pressure drop squared; above roughly 3 m/s, erosion and noise make stepping up the diameter cheaper than fighting physics.
Wet-bulb temperatureThe lowest temperature achievable by evaporative cooling at a site’s humidity. High wet-bulb (tropics) disables free cooling and forces mechanical refrigeration into the chain.
Wet-bulb, designThe wet-bulb exceeded only ~1% of hours at a site — the value free-cooling designs must survive. Tropical coastal sites design at 27-30 °C, Central Europe at 20-22 °C.