Modeling the Google Deschutes CDU in GT-SUITE: A Blueprint for Liquid Cooling Success

A Difficult Balance for Data Centers

As data centers push toward higher rack power densities and rapidly scaling AI workloads, liquid cooling has become essential for managing extreme thermal loads efficiently. Designing these next generation cooling systems is challenging – engineers must balance reliability, energy use, water consumption, and safety, all while navigating tight deployment timelines. Simulation plays a critical role in this environment. It enables teams to explore large design spaces, predict performance under dynamic conditions, and evaluate components and controls long before hardware exists.

Against this backdrop, Google led an Open Compute Project (OCP) initiative to define a standardized 2 MW, Next-Gen Coolant Distribution Unit (CDU) known as Project Deschutes. The project published a detailed specification and CAD reference to encourage broad industry adoption and interoperability. As a contributing OCP member, Gamma Technologies developed a complete GT-SUITE model of this CDU, faithfully reproducing the geometry and performance described in the specification. The result is a ready-to-use simulation asset that helps data center engineers evaluate system behavior, integrate the CDU into larger facility level models, and adapt the design for new products or scaling strategies.

From CAD to a High‑Fidelity 1D/3D Model

The released Deschutes model begins with a full 3D CAD assembly. Using GEM3D, the geometry is automatically converted into a GT‑SUITE simulation model, enabling accurate pressure‑drop prediction and fluid‑thermal behavior without manually extracting parameters. Engineers can visualize temperature distribution and flow variables directly on the original geometry, making results intuitive and accelerating design iteration.

Figure 1: Geometry conversion of original CAD file into a network of 1D components and visualization of results on top of original geometry

This preprocessing pipeline reduces both time and error, allowing users to move quickly from CAD to validated simulation while maintaining fidelity to the OCP reference design.

Matching OCP Performance and Enabling System Integration

The released Deschutes model reproduces the performance targets published in the OCP specification, offering a robust baseline for design studies. Because the system interfaces with chillers, buffer tanks, and IT coolant loops, the model can be directly integrated into broader facility architectures. Users can adapt component sizing, reconfigure piping, or explore alternative materials and fluids while still benefiting from the validated reference structure. This makes the model valuable not just for studying the OCP design but for developing future CDU generations.

Figure 2: Results of simulated performance compared to data from specification to validate model behavior

A Modular Simulation Environment for Technology Exploration

GT‑SUITE enables rapid comparison of components, materials, and cooling technologies. The modular environment supports evaluating single‑phase and two‑phase coolants, exploring different heat‑exchanger families, testing pumps and filters from multiple vendors, and studying the impact of dry coolers versus cooling towers. By adjusting geometry and operating conditions, teams can investigate trade‑offs related to PUE, WUE, heat‑recovery potential, and refrigerant selection. This flexibility allows engineers to tailor the CDU to site‑specific environmental conditions, sustainability targets, and operational constraints and ensure a future-proof design.

Figure 3: Illustration of modularity – heat exchangers from various suppliers can be directly compared to one another within a single simulation to optimize component selection. Other factors, such as fluid composition, can also be varied for quick and easy comparison.

Fast Optimization and Automated Design Exploration

Because data center cooling systems involve many interacting variables, optimization is essential. GT‑SUITE’s fast solvers make it possible to run extensive design‑of‑experiments studies, distributed computing sweeps, and optimization routines. In the Deschutes model, engineers can search for operating points that minimize electrical consumption while maintaining safe coolant temperatures at the rack outlets. This type of automated exploration helps identify optimal flow rates, component sizes, and control strategies, which can be performed even before physical hardware is available, thus supporting rapid product development and robust decision‑making.

Figure 4: Multi-factor, multi-objective optimization results using the built-in Design Optimizer. Primary and secondary flow rates are varied to minimize total electric power consumption and coolant return temperature from the racks.

Virtual Test Benches for Extreme Scenarios

Full‑scale physical testing of a 2 MW CDU is expensive and often impractical, especially for failure analysis. The Deschutes simulation model, applied as a virtual test bench, enables engineers to study transient events such as rapid IT load spikes, extreme ambient temperatures, or chiller power failures, which are difficult to obtain experimentally but essential for designing resilient cooling systems.

Figure 5: A thermal ride through analysis demonstrates how buffer tank volume affects supply temperatures when the chiller loses power, but pumps and IT load remain online.

Digital Twin and Fault Detection Capabilities

GT‑SUITE models can power machine‑learning metamodels to augment real‑time system monitoring. By comparing expected versus measured behavior, the digital twin can detect and classify anomalies such as valve failures (e.g. stuck or leaky valves), heat exchanger fouling, or pump cavitation to name a few. See this GT webinar for a deeper dive on fault detection: GT-SUITE For Increased Robustness of Fault Detection.

Figure 6: Demonstration of a valve fault that is properly identified by an anomaly detection ML model. When the valve cannot fully open, the flow rate is reduced which may not be obvious to the naked eye during dynamic operation. However, the anomaly detection model identifies deviation from expected conditions which is secretly costing multiple degrees on the coolant supply temperature.

The ability to generate large synthetic datasets makes GT-SUITE an efficient platform for training AI‑based condition‑monitoring tools. Coupled with the ability to interface with actual operational data via SCADA systems, etc., GT-SUITE lends itself well to the training of operators (e.g. what action to take when encountering a given fault). This ensures operators have confidence in the corrective action taken to ensure predictive maintenance and lowest possible downtime.

Figure 7: Illustration of real-time communication between GT model (either physics-based or from machine learning) and the system controls platform (e.g. SCADA) for live use during operation.

From Reference Design to Scalable Cooling Innovation

The GT‑SUITE model of the OCP Google Project Deschutes CDU provides a powerful foundation for engineers working on high‑density data center cooling. By combining a modular environment, rapid design exploration, and advanced digital‑twin capabilities, it accelerates innovation while reducing physical testing needs. The model helps teams design safer, more efficient, and more resilient cooling systems as data center demands continue to grow.

Ready to advance your data center thermal management strategy?

Visit our Data Center Solutions page to explore more data center simulation topics, learn more about our Digital Twin Solutions and Machine Learning capabilities, and browse our blog page for additional insights on digital twins and data center innovation or reach out to our team for a demo of the Deschutes CDU model and associated toolchain. Follow our LinkedIn channel to stay updated on the latest advancements in our solutions.