Singapore National Supercomputing Centre Liquid Cooling Retrofit Monitoring Project

The National Supercomputing Centre operates as a national HPC facility where computing equipment must support demanding workloads while maintaining reliable thermal conditions. For this project, 512 computing nodes were retrofitted with direct-to-chip cold-plate cooling.

In a direct-to-chip system, coolant circulates through cold plates positioned on heat-generating components. Heat is transferred into the circulating liquid and carried away through the cooling loop. This approach makes the condition of the coolant an important part of overall cooling-system management.

The facility used a coolant consisting of deionised water and glycol. Monitoring such a system requires more than simply checking whether coolant is flowing. Parameters related to water quality can help operators understand changes occurring inside the cooling loop.

For example, pH is used to indicate the acidity or alkalinity of a liquid. Electrical conductivity can provide information about ionic content. Depending on the system design and monitoring objectives, these and other parameters can contribute to a broader liquid coolant quality monitoring strategy.

The environmental conditions added another consideration. Singapore has a hot and humid tropical climate. For electronic monitoring equipment, high humidity can increase the risk of condensation, particularly where equipment is positioned close to cooling infrastructure.

For this reason, the project required a monitoring solution that could fit the physical environment while also supporting reliable communication with the centre’s existing control infrastructure.

One of the first challenges was the restricted space around the existing liquid cooling system.

Because this was a retrofit rather than a completely new installation, monitoring equipment had to be added around existing pipes and infrastructure. Large or complicated devices could have made installation more difficult and increased the amount of modification required.

Compact and easy-to-install monitoring devices were therefore important for the project.

The centre already used a Profibus-based SCADA system. New monitoring devices could not simply operate as isolated instruments; their data needed to become part of the existing monitoring environment.

This introduced a communication challenge.

The monitoring equipment used Modbus RTU communication, while the existing SCADA infrastructure was based on Profibus. A Modbus RTU-to-Profibus-DP gateway was therefore required to connect the two systems.

During integration, a baud-rate mismatch occurred. This prevented communication from operating as intended until the configuration was corrected.

The issue highlights an important consideration for any data center coolant water quality sensor project: selecting a suitable sensor is only one part of the deployment. Communication protocols, gateways and configuration settings also need to be considered during system design and commissioning.

A supercomputing centre cannot simply shut down for an extended period whenever new monitoring equipment needs to be installed.

For this project, installation work had to be carefully coordinated around maintenance windows. Eighteen monitoring points needed to be added to DN40 pipes, and the work required water shutoff.

The installation was therefore completed over three nights during scheduled maintenance windows.

This approach reduced the need for prolonged interruption while allowing the monitoring devices to be installed correctly.

Singapore’s tropical environment presented another practical challenge: condensation.

Junction boxes containing electrical connections and communication components needed additional protection against moisture associated with the humid environment. Anti-condensation heaters were therefore installed inside the junction boxes as part of the final solution.

The main requirement was to establish a practical monitoring system that could provide regular coolant information without relying solely on infrequent manual laboratory testing.

Previously, coolant assessment depended on monthly laboratory tests. Although laboratory analysis can provide valuable information, the long interval between tests means operators have less visibility into what happens between sampling dates.

The new system therefore needed to support much more frequent monitoring.

Eighteen monitoring points were planned across the cooling system. These points had to be installed on DN40 pipes while meeting the space restrictions of the existing infrastructure.

The monitoring arrangement also needed to communicate with the centre’s Profibus-based SCADA system. Rather than requiring operators to use an entirely separate platform, data from the monitoring points needed to become accessible through the facility’s existing monitoring environment.

Once commissioned, SCADA would poll the monitoring data every five minutes. This created a much more detailed operating record and supported daily trend monitoring.

The project therefore required a combination of:

• Compact monitoring equipment suitable for retrofit installation
• Compatibility with the deionised water and glycol coolant environment
• Installation on existing DN40 piping
• Integration between Modbus RTU devices and Profibus-DP
• Reliable operation in hot and humid conditions
• Frequent data collection for trend monitoring
• A design that could be deployed during limited maintenance windows

Together, these requirements shaped the final coolant water quality sensor monitoring solution.

The project installed 18 monitoring points across the liquid cooling system.

G3/4 threaded tee fittings were used to install the monitoring points on the DN40 pipes. This provided a practical connection method while working within the space limitations of the retrofit environment.

Communication presented a separate challenge. Because the monitoring devices and existing SCADA infrastructure used different communication standards, a Modbus RTU-to-Profibus-DP gateway was introduced.

The gateway allowed data from the Modbus RTU monitoring equipment to be transferred to the centre’s Profibus-based SCADA environment.

An integration issue was identified during commissioning when a baud-rate mismatch affected communication. The configuration was checked and corrected, after which communication could proceed as required.

Environmental protection was also considered. Anti-condensation heaters were added inside the junction boxes to address the high-humidity operating conditions.

Following integration, SCADA was configured to poll monitoring data every five minutes.

This combination of physical installation, communication integration and environmental protection created the foundation for regular coolant monitoring without requiring the centre to abandon its existing SCADA infrastructure.

Deployment had to be planned around the operating requirements of the HPC facility.

Because installation on the DN40 pipelines required water shutoff, the work was divided across three nights of scheduled maintenance windows. This allowed the team to install all 18 monitoring points while working within the facility’s operational constraints.

The deployment process involved installing the G3/4 threaded tee fittings, positioning the monitoring equipment and completing the necessary electrical and communication connections.

Once the physical installation was complete, attention shifted to SCADA integration.

The Modbus RTU devices were connected through the Profibus-DP gateway so that their data could be incorporated into the existing monitoring system. During commissioning, the team discovered the baud-rate mismatch. After the communication parameters were reconfigured, the integration issue was corrected.

Anti-condensation heaters were also installed inside the junction boxes to provide additional protection in the tropical environment.

Finally, the SCADA polling interval was configured to five minutes.

This commissioning process demonstrates why a successful CDU water quality sensor or wider coolant-monitoring deployment should consider the complete data path. The sensing device, installation interface, communication protocol, gateway configuration and central monitoring platform all need to work together.

For liquid cooling applications, Rika Sensor offers specialised sensors that can form part of a broader coolant-monitoring solution.

The RK500-12 Liquid Cooling pH Sensor is designed for continuous pH monitoring in liquid cooling environments. According to Rika Sensor’s current product specifications, it is compatible with deionised water, PG25 and EG25 media and supports RS485 and 4–20 mA outputs. The liquid-cooling model has a stated pH accuracy of ±0.1 at 25°C, 0.01 pH resolution, a 0–14 pH range and IP68 protection for the probe. It also offers G3/4 and other process-connection options, depending on configuration.

A coolant pH sensor can provide continuous information about acidity and alkalinity rather than relying entirely on occasional manual measurements. This makes pH monitoring a useful component of a wider coolant-condition monitoring strategy.

Rika Sensor also provides liquid-cooling EC, turbidity and ORP sensors. These can be selected according to the parameters that need to be observed within a particular cooling loop.

For example, a conductivity sensor for coolant monitoring can be used when electrical conductivity is an important water-quality parameter. Combining appropriate sensors can provide operators with a broader view of coolant conditions rather than depending on a single measurement.

The correct sensor combination should always be selected according to coolant chemistry, operating conditions, installation requirements and the monitoring objectives of the cooling system.

The most important operational change from the project was the move towards much more frequent coolant-condition visibility.

Before the retrofit monitoring project, the customer relied on monthly laboratory testing. After deployment and commissioning, monitoring data could be polled by the SCADA system every five minutes, and the customer accepted daily trend monitoring as the new approach.

This represented a major change in how coolant conditions could be observed.

Instead of waiting for the next monthly laboratory test to review coolant conditions, operators gained access to regularly collected data that could be used to follow daily trends.

The project also successfully addressed several practical implementation challenges. Eighteen monitoring points were installed on existing DN40 piping during three nights of maintenance windows. The Modbus RTU monitoring devices were integrated with the existing Profibus-based SCADA system through a gateway, while the baud-rate issue identified during commissioning was resolved through reconfiguration.

Anti-condensation heaters were incorporated into the junction boxes to accommodate the site’s hot and humid environment.

No quantified figures for cost savings, failure reduction or equipment-life improvement were supplied for this project. Therefore, the confirmed project outcome is centred on the successful deployment and integration of the monitoring system and the transition from monthly laboratory checks to daily trend monitoring.

This Singapore supercomputing centre project demonstrates that effective liquid cooling water quality sensor deployment involves more than installing measurement devices.

Physical space, pipe connections, coolant compatibility, communication protocols, environmental conditions and maintenance windows all affect how successfully a monitoring system can be introduced into an existing HPC facility.

In this case, 18 monitoring points were integrated into a 512-node direct-to-chip liquid cooling retrofit. G3/4 tee fittings provided a practical installation approach on DN40 pipes, while a Modbus RTU-to-Profibus-DP gateway connected the monitoring equipment to the existing SCADA infrastructure.

Most importantly, the facility moved from monthly laboratory testing to daily trend monitoring supported by data collected every five minutes.

For data centres, AI computing facilities and supercomputing centres adopting higher-density liquid cooling, a carefully designed liquid coolant quality monitoring system can provide operators with more regular visibility into coolant conditions and make water-quality information part of routine cooling-system management.