top of page

DLC: liquid cooling for performance and sustainability

  • Cedric KTORZA
  • Nov 17
  • 7 min read
ree

DLC—when liquid cooling becomes a lever for performance and sustainability. This article explains how Direct Liquid Cooling can unlock higher compute density, lower energy use, and practical pathways to heat reuse—all without compromising reliability or maintainability.

 

At a glance

  • Direct Liquid Cooling (DLC) enables high-density AI/HPC workloads while reducing cooling energy, fan noise and thermal hotspots.

  • Warm-water loops and heat reuse turn “waste heat” into a resource, improving sustainability KPIs beyond PUE.

  • Proven technologies include direct-to-chip cold plates, immersion cooling, and rear-door heat exchangers—each with clear use cases.

  • A phased roadmap (pilot → hybrid rows → scale) limits risk and avoids disruption to business operations.

  • At Score Group, we integrate DLC end-to-end: data center design (Noor ITS), energy systems and heat recovery (Noor Energy), and smart controls/IoT/AI (Noor Technology).

 

What DLC means in practice

Liquid cooling uses water or dielectric fluids to extract heat much closer to the heat sources than traditional air-based systems. Because liquids conduct heat far more efficiently, DLC can handle higher power densities with lower energy overhead and fewer airflow constraints.

DLC is not a niche technology for labs—it is fast becoming the default for AI clusters and modern HPC.

 

Direct-to-chip cold plates

  • Cold plates mount directly on CPUs/GPUs and other hot components.

  • A facility or in-row Coolant Distribution Unit (CDU) manages flow, temperature, and pressure.

  • Suited to mixed environments: liquid removes most heat; residual heat can be handled by existing air systems.

 

Immersion cooling (single- or two-phase)

  • Entire servers are immersed in dielectric fluid; heat is removed either via pumped fluid (single-phase) or controlled boiling/condensation (two-phase).

  • Delivers maximum density and simplified airflow paths; ideal for homogenous AI/HPC pods.

 

Rear-door heat exchangers (RDHx)

  • Liquid-cooled doors mount on rack rears to capture and remove heat from exhaust air.

  • A strong retrofit option to increase density in existing rooms without fully changing IT hardware.

For design guidance and standards, see ASHRAE TC 9.9 and the Open Compute Project’s Advanced Cooling Solutions:

 

Why liquid now: density, resilience, ESG

 

Performance impact you can measure

  • Higher sustained clocks: Lower junction temperatures allow turbo modes to hold longer, improving throughput for AI training and HPC workloads.

  • Tighter SLAs: Better, more predictable thermals reduce throttling risk and hot-zone incidents.

  • Capacity per square meter: DLC supports high and ultra-high rack densities, reducing the footprint of compute clusters.

  • Energy savings: Fan power falls substantially; chiller/compressor hours drop with warm-water loops and economization.

  • Resilience: Independent liquid loops, CDUs, and leak detection support fault containment and maintenance without impacting entire rooms.

 

Sustainability gains beyond PUE

  • PUE and WUE: DLC improves energy efficiency and can reduce water consumption when paired with dry coolers or closed-loop systems. For water guidance, see LBNL’s resources: https://datacenters.lbl.gov/resources/water-efficiency-data-centers

  • Heat reuse: Warm return temperatures (e.g., 40–60°C) enable recovery to heat offices, nearby buildings, or processes via heat exchangers/heat pumps—turning a liability into a resource.

  • Fewer refrigerants: Reduced reliance on compressor-based cooling cuts refrigerant use and associated emissions risks.

  • Grid-friendliness: Lower peak demand and the ability to pre-heat thermal storage can support demand-response strategies.

 

Air vs. liquid cooling: choosing the right fit

Title: Cooling strategies compared for density, efficiency, and retrofit fit

Dimension

Optimized Air

Rear-Door Heat Exchanger (RDHx)

Direct-to-Chip (DTC)

Immersion

Practical rack density

Low–moderate

Moderate–high

High–ultra-high

High–ultra-high

Energy efficiency

Good, limited by airflow

Better (captures exhaust heat)

Excellent (heat captured at source)

Excellent (minimal air handling)

Retrofit complexity

Low

Moderate

Moderate–High

High (separate pods/IT form factors)

Water usage

Possible if using adiabatic

Closed-loop variants available

Closed-loop; warm-water capable

Closed-loop; warm-water capable

Heat reuse potential

Limited

Moderate

High

High

Ideal use cases

Mixed legacy IT

Incremental density boost

AI/HPC clusters, mixed estates

Dedicated AI/HPC blocks

Note: Specific densities and savings depend on site design, climate, and IT load profile. Consult standards such as ASHRAE TC 9.9 and OCP ACS for reference designs.

 

Design and integration considerations

 

Hydraulics and temperatures

  • Warm-water supply (often 30–45°C) enables dry coolers and reduces or eliminates chillers.

  • Decouple IT and facility loops via CDUs for pressure control, filtration, and redundancy (N+1 or better).

 

Materials, safety, and maintenance

  • Use compatible metals, corrosion inhibitors, and quick-disconnects designed for IT.

  • Implement continuous leak detection, drip trays, and containment; define “wet work” procedures.

 

Controls and telemetry

  • Integrate valve control, pump speed, and CDU logic with BMS/DCIM.

  • Use sensors for flow, ΔT, differential pressure, humidity, and leak detection; stream data to analytics.

For reference architectures and KPIs, see:

 

Migration paths that work

  1. Baseline and model: Map heat distribution, airflow, and energy use; identify “hot” rows and AI/HPC islands.

  2. Target a pilot: Start with a high-density pod or a single row using RDHx or DTC with a CDU.

  3. Validate and measure: Track PUE, WUE, rack density, component temperatures, and availability for at least one seasonal cycle.

  4. Extend to hybrid rows: Combine DLC racks with optimized air-cooled racks to balance capacity and risk.

  5. Integrate heat reuse: Add heat exchangers/heat pumps to capture useful temperatures for building heating or nearby loads.

  6. Automate: Tie DLC controls to BMS/DCIM, add AI-driven setpoint optimization, and formalize maintenance playbooks.

 

How Score Group helps you adopt DLC

At Score Group, we bring energy, digital, and new tech together—so your cooling strategy advances performance and sustainability in tandem.

  • Noor ITS (data centers): Site assessment, capacity planning, reference architectures, DLC-ready designs (RDHx, DTC, immersion), and DCIM/BMS integration.

  • Noor Energy (energy systems): Hydronic engineering, CDUs and heat exchangers integration, free cooling, heat pumps, heat reuse to buildings/processes, and energy performance monitoring.

  • Noor Technology (smart tech): IoT sensors, telemetry, anomaly detection, and AI-driven control to optimize setpoints and detect faults early.

  • Program governance: Roadmapping, vendor-neutral selection, pilot-to-scale management, training, and M&V aligned with your ESG reporting.

Discover how this integrated approach helps you move faster with less risk at Score Group.

 

Governance, KPIs and reporting

  • Efficiency: PUE (ISO/IEC 30134-2), partial PUE, and distribution of cooling power.

  • Water: WUE (site and source water), makeup water quality and drift management.

  • Carbon: Emission factors for electricity/heat, avoided emissions from heat reuse.

  • Resilience: Redundancy (N, N+1, 2N) at CDU/loop levels, MTTR, incident logs.

  • Density: kW/rack progress, outlet temperatures, and thermal compliance bands per ASHRAE classes.

For sector trends and benchmarking, consult the Uptime Institute annual surveys: https://uptimeinstitute.com/research-and-reports/annual-survey-of-it-and-data-center-managers

 

FAQ

 

What is the difference between direct-to-chip and immersion cooling?

Direct-to-chip (DTC) uses cold plates mounted on specific components (CPUs/GPUs) and circulates a coolant through a closed loop controlled by a CDU. Airflow still handles residual heat from memory, VRMs, and storage. Immersion cooling places entire servers in a dielectric fluid, removing the need for most internal fans and simplifying airflow. DTC is often easier to integrate into mixed estates and with standard servers, while immersion shines for dense, homogenous AI/HPC pods where maximum heat capture and density are priorities.

 

Can liquid cooling reduce my overall energy consumption?

Yes, in most cases. By moving heat into liquid, you can cut fan energy substantially, raise supply water temperatures, and reduce or eliminate compressor-based cooling for many hours of the year. Combined, these effects can lower total facility energy and improve PUE. Actual savings depend on your climate, load profile, and loop design. Warm-water DLC with dry coolers and economization typically delivers the largest gains, especially for continuous AI/HPC workloads that benefit from stable, lower junction temperatures.

 

Is DLC safe for production environments?

Modern DLC systems are engineered with quick-disconnects, drip trays, leak detection, and pressure management to minimize risk. Separation between facility and IT loops via CDUs adds additional protection. Proper material selection, routine inspection, and well-defined maintenance procedures are essential. Many operators deploy DLC first in a pilot pod, validate controls and workflows, then scale to hybrid rows—proving reliability before broad rollout. Following ASHRAE and OCP guidance further supports safe, production-grade adoption.

 

How does DLC help with sustainability targets?

DLC improves energy efficiency by capturing heat at the source, reducing fan and chiller energy. Warm-water loops enable free cooling, fewer refrigerants, and practical heat reuse to buildings or processes—supporting carbon and energy KPIs, not just PUE. With careful design, WUE can improve through closed-loop systems and reduced evaporative cooling. These measures align with frameworks like the EU Code of Conduct and can help demonstrate progress on ESG goals, especially when paired with robust measurement and verification.

 

Do I need to replace all my servers to adopt liquid cooling?

Not necessarily. Many organizations start with targeted deployments: rear-door heat exchangers on selected rows, direct-to-chip kits for AI/HPC nodes, or self-contained immersion pods. This hybrid approach minimizes disruption and capital intensity. Over refresh cycles, you can increase the share of DLC-ready hardware. The key is to model heat loads, select the right technology per use case, and ensure facility-side readiness (CDUs, piping, controls) that can scale as your DLC footprint grows.

 

Key takeaways

  • DLC enables high density and stable performance for AI/HPC while cutting cooling energy.

  • Warm-water loops open the door to free cooling and meaningful heat reuse.

  • A phased, standards-aligned roadmap lowers risk and accelerates time-to-value.

  • Hybrid estates are practical: mix RDHx, DTC, and immersion where each fits best.

  • Robust telemetry and AI-driven controls sustain efficiency and reliability over time.

  • Ready to explore a DLC roadmap tailored to your site and workloads? Start the conversation with Score Group.

 
 
bottom of page