A cold plate project for an AI server passes through three sets of hands. Thermal engineers set the performance targets, the data center team plans the loop and picks the coolant, and procurement sends out RFQs and compares quotes. All three use the same liquid cooling terms, but they don’t always mean the same thing. This liquid cooling glossary is meant to close that gap.
When a thermal engineer says “0.02 °C/W,” is that measured from the junction or from the cold plate base? Does the facilities team’s pressure drop budget cover one cold plate or the whole run from CDU supply to return? If one quote was tested with PG25 and the other doesn’t name a coolant, can you compare them at all?
Mismatched terminology leads to quotes you can’t compare, samples built in the wrong direction, and rounds of back-and-forth that push out the schedule. Below are 69 cold plate and liquid cooling terms, grouped the way a project unfolds. Each entry opens with a one-sentence definition, and where APALTEK has a full guide on the topic, we link to it.
What Is a Liquid Cooling Glossary?
A liquid cooling glossary is a set of standardized definitions for the terms used to design, source, and operate liquid-cooled systems. For cold plate projects, the essential terms fall into eight groups: system architecture, cold plate anatomy, thermal performance, hydraulic performance, materials and manufacturing, coolant chemistry, connectors and testing, and sourcing.
The 10 Terms You’ll Hear Most
| Term | Quick Definition |
| Cold plate | A metal heat exchanger that mounts on a chip and carries coolant through internal channels. |
| Direct-to-chip cooling | Liquid cooling in which cold plates sit directly on the CPU or GPU. |
| CDU | The unit that conditions and circulates coolant for the IT loop, controlling its temperature, flow, and pressure. |
| TCS | The loop that runs from the cold plates to the CDU, through the IT equipment. |
| FWS | The facility-side loop that carries heat from the CDU to chillers or dry coolers. |
| Thermal resistance | Temperature rise per watt of heat (°C/W). Lower is better. |
| Pressure drop | The pressure lost as coolant flows through the cold plate. |
| Flow rate | The volume of coolant moving through the cold plate per minute, in LPM or GPM. |
| TIM | The material that fills microscopic gaps between the chip and the cold plate. |
| Quick disconnect | A fluid fitting you can connect and disconnect without draining the loop. |
How to Use This Glossary
Terms are grouped by project stage, and the A–Z index at the top of the page jumps straight to any entry. Linked terms point to APALTEK guides that go deeper.
Liquid Cooling System Architecture & Loops
Direct-to-Chip Cooling (D2C / DLC)
Direct-to-chip cooling is a liquid cooling method in which cold plates mount directly on CPUs, GPUs, and other high-power chips, and coolant flowing through the plates carries the heat away. DLC (direct liquid cooling) is the broader term, but in data centers the two usually mean the same thing. Learn more: What is a direct-to-chip cold plate?
Single-Phase Liquid Cooling
Single-phase liquid cooling is a system in which the coolant stays liquid through the entire loop and absorbs heat only by warming up. Most data center cold plate loops today are single-phase and run on water-based coolants such as PG25.
Two-Phase Liquid Cooling
Two-phase liquid cooling is a system in which the working fluid partially boils inside the cold plate and absorbs heat through the phase change. It can move a lot of heat at low flow and keeps chip temperatures more uniform, but the system has to manage vapor and tighter pressure control.
Technology Cooling System (TCS)
The technology cooling system (TCS) is the closed loop that runs through the IT equipment, also called the secondary loop. The OCP Cold Plate Sub-Project scopes it as the technology cooling system from the cold plate to the CDU, covering cold plates, tubing, manifolds, quick disconnects, and the CDU. It has the strictest water quality and filtration requirements in the system.
Facility Water System (FWS)
The facility water system (FWS) is the building-side water loop, also called the primary loop, which moves heat out of the CDU to chillers, dry coolers, or cooling towers. The FWS and TCS exchange heat inside the CDU, but their fluids never mix.
Coolant Distribution Unit (CDU)
A coolant distribution unit (CDU), also called a cooling distribution unit, is the device that circulates and conditions coolant for the IT loop, controlling its temperature, flow, and pressure. A liquid-to-liquid (L2L) CDU isolates the TCS from facility water through a heat exchanger; a liquid-to-air (L2A) CDU rejects heat to room air instead, with no facility water connection. CDUs are installed either in-rack or in-row. Learn more: What is a CDU?
Manifold
A liquid cooling manifold is a component that splits one coolant stream into many and collects the returns, from the rack level down to the cold plates inside each server. Learn more: What is a liquid cooling manifold?
Approach Temperature
Approach temperature is a measure of heat exchanger performance in a liquid-to-liquid CDU: the difference between the coolant temperature the CDU supplies to the IT side and the facility water temperature entering the CDU. For cold plate sizing, it tells you how much warmer than facility water your inlet coolant will be.
ASHRAE Liquid Cooling Classes (W17–W45, W+)
ASHRAE liquid cooling classes are categories defined by ASHRAE TC 9.9 according to the maximum facility water supply temperature. The fifth edition of its Thermal Guidelines replaced the old W1–W5 classes with W17, W27, W32, W45, and W+, and added a new W40 class; each number marks the upper supply temperature in °C. Warmer facility water means warmer inlet coolant and a tougher thermal resistance target for the cold plate.
Liquid-Assisted Air Cooling (LAAC)
Liquid-assisted air cooling (LAAC) is a rack-level approach in which a sealed liquid loop pulls heat off chips through cold plates, then rejects it to room air through a rear door heat exchanger, with no tie-in to facility water. Learn more: What is liquid-assisted air cooling?
Rear-Door Heat Exchanger (RDHx)
A rear-door heat exchanger (RDHx) is a liquid-to-air coil on the back door of a rack that cools server exhaust air before it returns to the room. Passive designs rely on server fans; active designs add their own fans for higher rack power.
Immersion Cooling
Immersion cooling is a method that cools servers by submerging them in a non-conductive dielectric fluid, either single-phase (the fluid stays liquid) or two-phase (it boils on hot surfaces). Learn more: Direct-to-chip vs. immersion cooling
Open-Loop vs. Closed-Loop Liquid Cooling Module
At the module level, a closed-loop (all-in-one, or AIO) module has its own pump and radiator and ships pre-filled and sealed, while an open-loop module connects through tubing and quick disconnects to an external CDU or heat rejection unit to form a complete loop. Don’t confuse the latter with facility-level “open-loop” cooling towers. Learn more: open-loop and closed-loop modules.
Cold Plate Anatomy
Cold Plate
A cold plate, also called a liquid cold plate or water block, is a metal heat exchanger with one face against a heat source and coolant channels inside. Heat conducts through the metal and passes into the flowing coolant. Server CPU and GPU cold plates are usually copper. Learn more: What is a cold plate and how does it work?
Base Plate & Cover
The base plate is the part of a cold plate that contacts the chip and carries the fins or channels; the cover seals those channels and holds the ports. They’re joined by brazing, friction stir welding, laser welding, or O-rings and screws. APALTEK’s server and GPU cold plates use brazed copper construction, forming one permanent, sealed assembly.
Inlet / Outlet Ports & Fittings
Inlet and outlet ports are the openings where coolant enters and leaves a cold plate, along with the threaded fittings, barbs, or quick disconnects that connect to them. Port location dictates how the internal channels can be routed, so decide early which positions can move.
Skived Fin
A skived fin is a thin fin sliced up from a solid metal block, usually copper, with a precision tool, producing dense rows of fins that are one piece with the base. With no bonded interface between fin and base, skived fins are a common choice for AI server cold plates. Learn more: Cold plate manufacturing process
Folded Fin
A folded fin is a thin metal strip folded into a corrugated shape and brazed to the base plate. It’s low-cost with lots of surface area, but it adds a brazed interface between fin and base.
Pin Fin
A pin fin structure is an array of small posts that coolant flows around from multiple directions. Mixing and heat transfer are strong, and so is pressure drop. Pin fins are common in power electronics and EV cold plates.
Microchannel
Microchannels are dense, parallel flow channels typically well under a millimeter wide. They pack more heat transfer area into the same volume but are more sensitive to particles and biofouling, so filtration matters more.
Jet Impingement
Jet impingement is a cold plate design in which coolant is forced through nozzles to strike the base plate directly above the chip’s hot spot. The thin boundary layer gives very high local heat transfer, at the cost of more complexity and higher pressure drop.
Flow Path Configurations (Serpentine, Parallel, Split Flow)
Flow path configuration is the layout coolant follows inside a cold plate. Serpentine routes all coolant along one long channel, with thorough heat transfer but high pressure drop. Parallel splits flow across many short channels, cutting pressure drop but requiring even distribution. Split flow enters at the center and exits both sides, so the coldest coolant hits the central hot spot first. Learn more: Cold plate design guide
Mounting & Retention Hardware
Mounting and retention hardware is the backplates, spring-loaded screws, and clips that hold a cold plate against the chip at a specified pressure. Mounting pressure sets TIM thickness and contact resistance, so follow the chip vendor’s mechanical specs.
Thermal Performance Terms
TDP / TGP
TDP (thermal design power) is the chip power a cooling solution must handle continuously, while TGP (total graphics power) covers an entire graphics or accelerator board, including memory and power delivery. Size cold plates to maximum sustained power, not typical power.
Heat Flux & Hot Spots
Heat flux is the power dissipated per unit area, usually in W/cm². Uneven power inside a chip creates hot spots, and at the same total power, higher heat flux makes a chip harder to cool.
Junction Temperature (Tj)
Junction temperature (Tj) is the temperature of the silicon inside the chip, and the number thermal design ultimately protects. Past the vendor’s maximum, the chip throttles, and sustained overheating shortens its life.
Case Temperature (Tcase)
Case temperature (Tcase) is the temperature at the surface of the chip package, such as the integrated heat spreader. Because it can be measured directly, it’s often used as a boundary condition in design and testing.
Thermal Resistance (°C/W)
Thermal resistance is the temperature difference divided by the heat load, expressed in °C/W; lower is better. Inside a cold plate, it’s the sum of conduction through the metal, convection at the channel walls, and the coolant’s temperature rise, acting in series. Suppliers don’t all measure it between the same two points:
| Reference basis | Temperature difference used |
| Case-to-inlet | Chip case temperature minus coolant inlet temperature |
| Surface-to-fluid | Cold plate surface temperature minus local fluid temperature |
| OCP test specification | Cold plate base temperature minus coolant inlet temperature, divided by applied power |
At the chip level, junction-to-case thermal resistance (θJC) describes heat flow inside the package. Before comparing numbers, confirm the reference temperatures, flow, coolant, and TIM match. Learn more: How to specify comparable performance targets in an RFQ
Thermal Interface Material (TIM)
A thermal interface material (TIM) is the paste, phase-change material, pad, or liquid metal that fills microscopic gaps between chip and cold plate, replacing air, which conducts heat poorly. The TIM layer is often a meaningful share of total thermal resistance.
Bond Line Thickness (BLT)
Bond line thickness (BLT) is the actual TIM thickness after the cold plate is clamped down. Thinner means lower interface resistance; BLT depends on mounting pressure, surface flatness, and the TIM’s viscosity and filler size.
Flatness & Surface Roughness
Flatness is the deviation between the highest and lowest points of the cold plate contact surface, and surface roughness (such as Ra) describes its fine texture. Poor flatness leaves the TIM layer thick and uneven, which raises contact resistance.
Coolant Temperature Rise (ΔT)
Coolant temperature rise (ΔT) is the outlet temperature minus the inlet temperature. More flow means less rise, and a lower-specific-heat coolant warms more. In series layouts, one plate’s rise becomes the next plate’s higher inlet temperature.
Hydraulic Performance Terms
Flow Rate (LPM / GPM)
Flow rate is the volume of coolant passing through a cold plate per minute, in liters (LPM) or US gallons (GPM); 1 GPM is about 3.785 LPM. Specs should reflect the flow each plate actually receives, not total loop flow.
Pressure Drop (ΔP)
Pressure drop is the pressure lost as coolant flows through a cold plate, usually including its fittings. It rises faster than flow (roughly with its square in turbulent flow), and the combined drop across plates, manifolds, quick disconnects, and tubing must fit within the pump or CDU head.
P-Q Curve (Flow Impedance Curve)
A P-Q curve is a plot of pressure drop measured across a range of flow rates. Overlay it on the pump curve to find the real operating point; a single data point can’t predict behavior anywhere else.
Flow Balancing
Flow balancing is getting the intended flow to each branch when several cold plates are fed at once. Series layouts are simple but feed downstream plates warmer coolant; parallel layouts need matched flow resistance or flow limiters to keep branches even.
Operating, Proof & Burst Pressure
Maximum operating pressure is the pressure a cold plate sees in long-term service; proof and burst pressures are test levels above it. Proof testing holds the plate above operating pressure without leaks or deformation; the OCP cold plate test specification holds the plate at maximum operating pressure for 5 minutes, then at 3x that pressure for 2 minutes. Burst testing is destructive, raising pressure until failure or a defined ceiling to confirm the safety margin.
Flow Velocity & Erosion
Erosion is the gradual wear of copper channel walls caused by coolant moving too fast, which also adds pressure drop and noise. Design and validation both set a maximum velocity, so check it whenever you add flow to chase performance.
Cold Plate Materials & Manufacturing
Copper Grades (C11000, C12200, C10200)
C11000, C12200, and C10200 are the copper grades most often discussed for cold plates. C11000 is electrolytic tough pitch (ETP) copper, at least 99.90% pure with trace oxygen; it conducts well and costs less but risks hydrogen embrittlement when brazed in hydrogen-bearing atmospheres. C12200 is phosphorus-deoxidized (DHP); it brazes well, but residual phosphorus lowers conductivity somewhat. C10200 is oxygen-free (OFC), at least 99.95% pure, and well suited to high-temperature brazing and welding.
Aluminum Alloys (6061, 3003)
6061 and 3003 are the aluminum alloys most common in cold plates. Aluminum conducts heat about half as well as copper but weighs about a third as much, so it’s widely used in EV and energy storage cold plates. 6061 is strong and machinable; 3003 forms easily and is a common brazing sheet core.
CNC Machining
CNC machining is cutting channels, fins, and features out of solid stock on a computer-controlled mill. It handles complex shapes and tight tolerances and is the go-to process for prototypes and low-to-mid volumes. Learn more: Cold plate manufacturing process
Vacuum Brazing
Vacuum brazing is a joining process in which the assembly is heated in a vacuum furnace until a lower-melting filler metal liquefies and is pulled into the joints by capillary action, bonding the parts. The vacuum prevents oxidation, so flux usually isn’t needed. It can join dense internal fins in one cycle, but batch cycles are long and high temperatures can soften copper.
Friction Stir Welding (FSW)
Friction stir welding (FSW) is a solid-state joining process in which a rotating tool softens and stirs the metal into a weld without melting it or adding filler. Welds are dense and near base-metal strength, a good fit for simple perimeter seams between cover and base.
Diffusion Bonding
Diffusion bonding is a joining process in which high temperature and pressure make atoms diffuse across mating surfaces, forming a metallurgical bond without filler. It suits stacked-layer microchannel structures, but equipment and cycle costs are high.
Nickel Plating
Nickel plating is a nickel layer applied to copper to improve corrosion resistance and reduce copper leaching into the coolant. Weigh it against the loop’s other wetted materials and coolant chemistry.
Additive Manufacturing (Metal 3D Printing)
Additive manufacturing is building parts layer by layer from fused metal powder, which enables conformal 3D channels that machining can’t produce. It’s mostly used for prototypes; cost, internal roughness, and density limit volume production.
Liquid Cooling Coolants & Fluid Chemistry
Deionized (DI) Water
Deionized (DI) water is high-purity water with dissolved ions removed. It offers the best heat transfer of the common coolants, but keeping corrosion and biological growth in check takes ongoing treatment.
PG25
PG25 is a coolant of 75% water and 25% propylene glycol, a concentration that keeps biological growth in check, usually with corrosion inhibitors added. It’s the most common data center cold plate coolant. Being more viscous and lower in specific heat than water, it produces more temperature rise and pressure drop at the same flow, so performance data should always name the test coolant.
Corrosion Inhibitors
Corrosion inhibitors are coolant additives that form a protective film on metal surfaces. Formulations vary by supplier and must match every wetted metal, so an RFQ should name the specific coolant product, not just “PG25.”
Conductivity & pH
Conductivity and pH are the two most common coolant quality checks. Rising conductivity usually signals dissolved ions or corrosion byproducts; drifting pH can speed corrosion or deplete inhibitors.
Filtration (Micron Rating)
A filter’s micron rating is the smallest particle it captures, measured in microns (µm). Fine cold plate channels make TCS filtration far stricter than the facility side’s; OCP’s modular TCS white paper includes a dedicated discussion of 25 µm filtration in TCS loops.
Wetted Materials & Galvanic Corrosion
Wetted materials are everything in the loop that touches the coolant, and galvanic corrosion is what happens when dissimilar metals such as copper and aluminum share a conductive fluid: the more active metal corrodes faster. That’s why material selection needs the loop’s full bill of materials. Learn more: What is a cold plate and how does it work?
Dielectric Fluid
A dielectric fluid is a non-conductive coolant, such as a synthetic hydrocarbon or fluorinated fluid, used mainly in immersion cooling. A leak won’t short a circuit, but heat transfer is lower than water-based coolants and cost is higher.
Liquid Cooling Connectors, Testing & Reliability
Quick Disconnect (QD)
A quick disconnect (QD) is a fluid fitting that connects and disconnects while the loop is filled. Dry-break designs release virtually no fluid; others leave slight wetting. Sizes vary across the loop, and the CDU-to-manifold connection is usually where the largest quick disconnects are used.
Universal Quick Disconnect (UQD)
The universal quick disconnect (UQD) is an open OCP standard for hand-mated, drip-free liquid cooling connectors. It defines interface dimensions so that conforming plugs and sockets of the same nominal size are interchangeable, along with requirements for service life, fluid loss, and pressure ratings. See the UQD specification.
Blind-Mate Connection
A blind-mate connection is a quick disconnect that couples automatically as a server slides into the rack. Blind-mate fittings must tolerate some misalignment and are common in dense, frequently serviced racks.
Leak Testing (Pressure Decay & Helium)
Leak testing is the factory verification that a cold plate is sealed. Pressure decay testing fills the plate with gas and watches for pressure loss; helium leak testing uses a tracer gas to find extremely small leak paths. At APALTEK, test items are defined by each customer’s requirements rather than one fixed checklist. Learn more: Cold plate manufacturing process
Thermal Test Vehicle (TTV)
A thermal test vehicle (TTV) is a heated fixture with built-in sensors that mimics a chip’s footprint and power map, used to measure thermal resistance when real silicon isn’t available. The closer it matches the real chip, the more useful the data.
Thermal Cycling & Vibration Testing
Thermal cycling and vibration testing are reliability tests that expose a cold plate to repeated hot-cold extremes, shock, and vibration, then retest pressure integrity and thermal performance to confirm joints and seals hold up over the product’s life.
Leak Detection
Leak detection is real-time sensing of leaks during operation, using sensor cables under cold plates and manifolds wired to the CDU or server management system. Leak testing is a factory checkpoint; leak detection is a safeguard in the data hall.
Cold Plate Sourcing & Project Terms for Buyers
RFQ (Request for Quotation)
An RFQ (request for quotation) is the package a buyer sends so a manufacturer can evaluate and price a custom cold plate: drawings, thermal targets, coolant conditions, test requirements, and volumes. A complete RFQ means a more accurate quote and less sample rework. Learn more: How to prepare an RFQ for a custom cold plate
NDA (Non-Disclosure Agreement)
An NDA (non-disclosure agreement) is a confidentiality agreement signed before chip thermal data or drawings change hands. Server and accelerator projects typically require one before full drawings are shared.
DFM (Design for Manufacturability)
DFM (design for manufacturability) is a review of a design before it’s locked to confirm it can be built economically: channels can be machined, tools can reach every seam, tolerances are affordable, and the part can be inspected. Bringing your cold plate supplier into this review before the design is frozen is the cheapest way to avoid rework.
NRE (Non-Recurring Engineering)
NRE (non-recurring engineering) is the one-time cost of design, simulation, dedicated fixtures and tooling, and first-build validation, usually quoted separately from unit price.
Prototype, EVT, DVT & PVT
EVT, DVT, and PVT are the validation stages a hardware design passes through after the prototype. EVT (engineering validation) confirms the design meets its targets, DVT (design validation) runs full reliability and compliance testing on near-production parts, and PVT (production validation) confirms yield and consistency on the production line.
FAI (First Article Inspection)
FAI (first article inspection) is a documented, feature-by-feature inspection of the first production parts against the drawing, and a key gate between samples and volume production.
Golden Sample
A golden sample is a customer-approved reference part used throughout production as the benchmark for appearance, dimensions, and performance.
OEM vs. ODM
An OEM (original equipment manufacturer) builds to the customer’s design, while an ODM (original design manufacturer) handles design and production and the customer sells under its own brand. The cold plate SKUs on APALTEK’s website are capability examples; actual products are custom-built around each project’s chips, space, and loop conditions.
Lead Time & MOQ
Lead time is the time from order confirmation to delivery, and MOQ (minimum order quantity) is the smallest accepted order. Both shift between prototype, pilot, and volume stages because process routes and tooling differ.
Frequently Asked Questions
What’s the difference between TCS and FWS?
The TCS is the closed loop from the cold plates to the CDU, running through the IT equipment, with the strictest cleanliness requirements. The FWS is the facility loop from the CDU to chillers, dry coolers, or cooling towers. In a liquid-to-liquid CDU they exchange heat through a heat exchanger, but their fluids never mix, so IT equipment never touches untreated facility water. Learn more: What is a CDU?
Why can’t you compare cold plate thermal resistance numbers across suppliers?
Because suppliers use different reference temperatures: case minus inlet, surface minus local fluid, or cold plate base minus inlet. Differences in flow rate, coolant, heated area, and TIM widen the gap further. Ask for data on the same basis and test conditions, or specify the outcome instead: under defined conditions, chip temperature must stay below a set limit.
What is PG25, and why is it so common in cold plate loops?
PG25 is 25% propylene glycol and 75% water. That concentration suppresses biological growth, and with inhibitors it protects copper cold plates and CDUs with far less maintenance than pure DI water. It transfers heat slightly less well than water, so performance data should state which coolant was used.
What do EVT, DVT, and PVT mean in a cold plate project?
They’re the validation stages on the way to production: EVT checks the design meets its targets, DVT completes reliability and compliance testing on near-production parts, and PVT verifies yield and consistency on the real line. Telling a supplier which stage you’re in leads to a more accurate quote.
Put These Terms to Work in Your Next Cold Plate Project
You don’t need every number on day one. Spell out a few key parameters in the terms above, and an engineering team can point you in the right direction much faster: chip TDP and hot spot locations, coolant type and inlet temperature, the flow and pressure drop budget per cold plate, your temperature target and how it’s measured, and the mounting interface.
Send those parameters to APALTEK’s engineering team and request a cold plate design review. We’ll evaluate channel structure, materials, and process route based on your chip, space, and loop conditions. Or browse APALTEK’s cold plate lineup to see the range of custom work we do.