Liquid Cooling in Data Centers: Which Specifications Make Comparisons Possible?

“Liquid cooling” does not, by itself, identify a complete system. It can refer to technologies that bring liquid to components, to fluid circulating through a facilities loop, or to equipment that transfers heat between circuits. A coolant distribution unit (CDU), for example, is a distribution and heat-transfer component; it should not automatically be treated as synonymous with the data center’s entire cooling system. ASHRAE’s resource page and its technical committee documentation on data centers bring together material specific to this field, but the resources consulted here do not provide a standardized specification sheet for comparing two specific installations. ASHRAE and ASHRAE TC 9.9 are starting points, not a universal product classification. The first question should be where the boundary of the published figure begins and ends.

In practice, a figure might describe only the CDU, the liquid loop serving a row of racks, or a broader infrastructure that includes heat rejection. Comparing the thermal capacities of equipment with different scopes can appear conclusive while failing to answer the purchasing decision. Record which equipment is included, which fluid circulates in each loop, where transferred heat is measured, and which external components are excluded. This boundary-setting is a way to assess the information; it does not mean that every manufacturer publishes these fields or that the evidence available here is sufficient to claim one configuration is superior to another.

Commercial terminology can also obscure architectural differences. Direct-to-chip cooling, liquid-to-liquid loops, liquid-to-air solutions, and immersion are not interchangeable labels. Before comparing proposals, request a flow diagram and confirm whether liquid reaches the servers, whether it exchanges heat with another loop, and what part of the thermal load still depends on air. Without that description, “cooling capacity” is not a sufficient basis for comparison.

A technical specification makes sense only alongside its conditions. For thermal capacity, it matters which inlet and outlet temperatures, flow rate, and ambient conditions apply; for flow rate, which fluid and pressure; and for efficiency, which energy uses are included. These fields are a recommended checklist, not confirmed data for any product in this research. LiquidStack’s guide to selecting a CDU and Chilldyne’s article on ordering liquid-cooling systems are manufacturer sources: they can help shape questions, but their recommendations and commercial claims are not, in themselves, independent results. LiquidStack and Chilldyne should be read in that context.

To compare two proposals, record every value together with its unit, measurement point, load profile, and test conditions. If one of those elements is missing, mark the figure as not comparable rather than filling the gap with an assumption. It is also important to distinguish rated capacity from capacity available in the installation: the latter may depend on the loop, configuration, and operating margins assumed by the supplier. The sources provided do not include equivalent test results for several systems or a series of specification sheets with consistent conditions. It is therefore not possible to produce a ranking, performance table, or responsible quantitative conclusion.

Auxiliary power deserves separate treatment. Pumps, fans, and other infrastructure components may require electricity; but an isolated figure cannot attribute the data center’s total energy use to liquid cooling, nor can it compare systems if one figure includes more components than another. Request electrical power under defined conditions, system boundaries, workload, and calculation method. If a percentage improvement is cited, ask for the reference scenario, measurement period, equipment included, and whether the result was measured or modeled. A percentage without an explicit baseline and boundary says little about the savings another installation can expect.

Comparison does not end with the equipment specification sheet. A proposal may require changes to pipework, connections to existing infrastructure, space, commissioning procedures, and coordination with servers. To evaluate it, document who supplies each section, what infrastructure is assumed, and which tests are required before system acceptance. Also request instructions for inspection, intervention, circuit isolation, and response to a leak. These are technical due-diligence criteria; the available material does not support presenting specific installation or maintenance requirements as universal.

Fluid quality and compatibility with materials are relevant questions for an operating plan, but do not infer values or service intervals without documentation from the supplier and the connected components. ASHRAE TC 9.9 lists technical publications related to system preparation and coolant integrity; its page can help locate specialized material, although the information available in this research does not provide the complete contents of those documents. ASHRAE TC 9.9 index. Useful evidence for a purchase should cover both planned operation and the tasks needed to sustain it.

For redundancy, expansion capacity, and service, it is not enough to accept terms such as “modular” or “fault tolerant.” Ask for the exact configuration that supports each property: installed units, operating mode during an outage, capacity limits, and shared dependencies. If the supplier does not provide those details, the feature should remain an assertion awaiting verification. Product documentation does not replace installation design, nor can it by itself establish continuity of service for an entire data hall.

A robust review distinguishes four types of evidence: technical standards and guidance; manufacturer specifications and manuals; independent measurements; and testimonials or promotional publications. ASHRAE offers data-center resources and maintains technical committee TC 9.9; its pages are institutional references for locating documentation, but a figure not present on an index page should not be attributed to that page. Chilldyne’s and LiquidStack’s articles, meanwhile, are publications by companies with an interest in the market. They can help identify selection questions, but a seller’s recommendation is neither a neutral comparison nor proof of performance.

The academic articles included in the research address general questions of energy and cooling in data centers, but the available extracts do not allow specific results for a particular CDU or conditions for comparing products to be verified here. Review of data center energy consumption. To use independent research in a decision, check that the study covers a relevant architecture, describes its method and limitations, and publishes data that can be related to the scenario of interest. A modeling study is not equivalent to a field measurement; a measurement at one installation does not prove that the same result will be reproduced at another.

A checklist can reduce ambiguity before requesting proposals. Ask for: (1) a diagram and system boundary; (2) capacity, flow rate, and temperatures with test conditions; (3) itemized auxiliary power and measurement method; (4) connection and commissioning requirements; (5) a maintenance plan and compatible fluids; and (6) test documentation and references for a comparable scenario. Not every field needs to appear on a single specification sheet, but each response should be traceable to specific documents. If a claim cannot be linked to a source, record it as unverified rather than incorporating it into a score.

The evidence consulted supports a limited conclusion: ASHRAE has technical resources devoted to data centers, and manufacturers publish guidance on selecting liquid cooling and CDUs. It does not support a system ranking, comparative energy-use figures, or a purchase recommendation for a particular installation. The available extracts do not include consistent specifications, shared protocols, or verified results under equivalent conditions. For now, the responsible decision is to compare complete technical dossiers, not isolated promises.

A specific comparison would require identified models, official documentation for each unit, matching operating conditions, and data clarifying which energy uses and components are included. It would also be necessary to define the load profile, the project’s physical boundaries, and continuity and maintenance requirements. Without that information, any product table would imply a level of precision the sources do not provide. The absence of comparable data does not prove that the solutions are equivalent; it only prevents establishing a substantiated difference.

Readers can use this framework to prepare a request for information and identify gaps before a promotional figure becomes a selection criterion. Treat manufacturer claims as hypotheses that need documentation; require units, conditions, and scope; and distinguish measured data from calculated data. Temperature, auxiliary power, and installation requirements are useful fields, but they become valuable for comparison only when their definitions match. This guide does not evaluate specific products or replace engineering design for an installation.