The most significant change is the arrangement of the servers in clusters, connected in series for cooling, while the servers within each cluster remain organized in parallel, simplifying maintenance. This new design optimizes the layout of components, including bulky ones like the in-rack CDU, and helps reduce the rack's power consumption.
The rack operates with a "pull" hydraulic configuration, ensuring each server receives the appropriate water flow and pressure for its cooling needs. Hardware components, such as the CPU and GPU, are cooled by OVHcloud-designed direct-to-chip water blocks, which dissipate heat through a closed water loop that connects to a single cooling circuit throughout the entire data center.
The cooling module (also known as the Cooling Unit or CDU) is now approximately 50% more compact and is located outside the rack. Capable of cooling multiple rows of racks, it features over thirty sensors that monitor the parameters of the water coming from the racks—pressure, flow rate, and temperature—to adjust the cooling in real time. These sensors allow the rack to be aware of its immediate environment and the data center's temperatures. Thanks to this new data, the intelligent cooling module can automatically adapt to the server workload. This optimization significantly extends the lifespan of the equipment and helps optimize the infrastructure's energy consumption.
The external Smart Dry Cooler is the final component in the closed-loop water cooling system. It occupies half the space and uses half the fans of the previous generation of units. This helps reduce cooling energy consumption by up to 50%, while also lowering ambient noise levels.
“We are delighted to introduce a new generation of evaporative cooling solutions, particularly with Smart Racks. Thanks to our integrated cooling circuit design, we are paving the way for smart data centers that leverage our most advanced technologies, as well as artificial intelligence, to solve what was previously an impossible equation,” says Miroslaw Klaba, R&D Director at OVHcloud.
OVHcloud's new Smart Racks are self-regulating and collect daily operational data. Data from the racks, cooling modules, and dry coolers is fed into a data lake, where complex algorithms determine predictive behaviors that help optimize maintenance and increase equipment lifespan. The system can also connect to a local weather station to further enrich the data lakes. The models are so precise that they can accurately dose the volume of water needed to inject into the dry coolers' adiabatic cooling pad system, enabling heat exchange with the outside environment.
Thus, the new Dry Cooler design reduces water consumption by 30%, which could significantly improve OVHcloud's already effective Water Efficiency Index (WUE). Furthermore, unlike traditional adiabatic systems, the cooling pad system does not rely on a recirculation circuit: complex hydraulic units such as pumps, tanks, and level sensors are eliminated, reducing infrastructure complexity and simplifying maintenance.
The predictive AI that accompanies OVHcloud Smart Racks can now anticipate and react to operational dimensions at the infrastructure level, including pump speed (and therefore water flow), fan speed, and even valve opening to optimize the configuration.
The rack's performance can also be adjusted based on external constraints, such as noise limits, water availability, or energy costs. In this way, the algorithm can choose to consume more electricity to conserve water or adjust noise levels to suit urban environments.
The new generation of OVHcloud Smart Racks is being deployed in the Roubaix data center, in a room with approximately 60 racks and 2,000 servers, and is accompanied by its new cooling system. The entire system is planned for implementation across the group's data centers.
