As the density of AI computing power increases exponentially, the heat dissipation system of data centers is accelerating the migration from air cooling to liquid cooling.The new generation of computing power platforms represented by NVIDIA high-power GPU servers is forcing the cooling system to break through traditional boundaries. Its stability requirements are no longer an additional attribute of "energy saving and efficiency improvement", but have become a core infrastructure standard that supports the continuous output of computing power.
How to maintain the lifeline of "zero leakage" in long-term high-load operation?It has become one of the most concerned issues for manufacturers of computing power centers, energy storage systems, and high-power electronic equipment.
Liquid-cooled secondary side: an underestimated risk dark area
The liquid cooling system is divided into a primary side and a secondary side. The secondary side is a complex network composed of cold plates, stainless steel bellows, water distributors, UQD quick connectors and other components. It has a large number of interfaces, dense welding points, and high assembly accuracy requirements. It is a concentrated area of leakage risks.
In the harsh operating environment of high-power servers, even an undetectable initial leak may continue to amplify over time: at least it will cause coolant loss and high operation and maintenance costs, or at worst it will directly cause a shutdown accident, which can easily cause millions or even tens of millions of yuan in losses.
For this reason, the industry has quietly raised the leakage control standard to the 10⁻⁸ Pa·m³/s level - this iconic value declares that liquid cooling has officially entered the "micro-leakage control era", and the requirements for technical accuracy have reached unprecedented heights.
Highly sensitive helium mass spectrometry leak detection (helium detection) is the key to ensuring this bottom line.This article will focus on the three core components of the secondary side cooling system: the cold plate, stainless steel pipes, Manifold, and UQD quick-change joints, analyze the importance of helium inspection, and introduce Hiraysi’s solutions.
1Cold plate: Helium inspection is the last line of safety defense
Cold plate liquid cooling is currently the most mature and widely used solution. Its principle is similar to a "heat-reducing patch". A copper water-cooling plate is attached to heating chips such as CPU/GPU to take away heat.According to the heat exchange method with the cabinet, it can be divided into:
liquid to gas (L2A)：The cabinet is liquid-cooled and heat is dissipated through the heat dissipation back door (RDHx), which is suitable for the renovation of old computer rooms.
liquid to liquid (L2L)：The liquid heat exchange between the secondary side and the primary side is carried out through the CDU, which has the highest efficiency and is the mainstream solution for new generation AI servers such as GB200 NVL72.
Microchannel Cold Plate (MCCP) is the future trend.It shrinks the internal flow channel to 80-90μm, greatly increasing the heat exchange area and significantly improving the heat dissipation capacity.In NVIDIA's planned full liquid cooling system VR NVL 144, the computing layer will use MCCP water cooling plates directly attached to the Vera Rubin chip, making it a high-density heat dissipation core spread throughout "precision liquid blood vessels".
Why does the cold plate need to be tested for helium?
1Microleak detection capabilities
The internal flow channels of the cold plate are complex, and micro-channel vacuum brazing is often used.Helium molecules are extremely small, and helium detection can detect as low as The extremely tiny leak point of 10⁻¹² mbar·L/s far exceeds traditional methods such as water detection and bubble detection, and can completely eliminate hidden dangers before leaving the factory.
2Ensure equipment and business security
Data center liquid coolants are mostly conductive or non-conductive coolants.Once a leak occurs, it can easily cause the CPU/GPU and motherboard to short-circuit, causing a fire or the entire cabinet to burn down, causing tens of millions of dollars in losses.Helium inspection prevents "small leaks from turning into big problems" by intercepting micro leaks in advance.
3Ensure high reliability and operational efficiency
Helium inspection can accurately locate welds or material defects to ensure that the cold plate leakage rate is lower than 0.001mL/min.This can reduce performance degradation and maintenance downtime caused by long-term operation, which is crucial for AI computing clusters that require 7×24 hours of high-reliability operation.
4Support high computing density thermal management
In AI cabinets where the power of a single cabinet can easily reach tens of kW, local overheating is enough to cause the chip to reduce frequency or even shut down.Reliable sealing can maintain effective heat transfer of coolant and prevent local hot spots and system freezes caused by micro-leakage.
⬆ Cold plate helium detection system
2Pipelines and Manifold: Helium inspection is a “lifeline”
The new generation of NVIDIA Rubin solutions uses stainless steel corrugated pipes to replace traditional rubber hoses on a large scale, and widely uses stainless steel Manifolds (manifold pipes).Manifold is a key component connecting the CDU and the cold plate, and its sealing reliability is the "lifeline" of the liquid cooling system.
⬆ Liquid cooling pipeline
Manifold’s main leakage risk points
Welding defects:There may be microscopic defects invisible to the naked eye in dozens to hundreds of solder joints.
Connection parts:Threads, flanges, quick connectors, etc. rely on sealing rings and are easily affected by installation processes and environmental aging.
Material defects:Microscopic pores or stress cracks that may occur during metal processing.
Traditional water inspection or pressure drop test can only detect large leaks and micro leaks that are sensitive to liquid cooling systems.(The leakage rate is lower than 10⁻⁵ mbar·L/s）Then there is nothing you can do.The helium detection technology can detect as low as 10⁻¹² mbar·L/s The ability of extremely small leak points has become an irreplaceable ultimate detection method in the manifold manufacturing and integration process, and can achieve "zero leakage" prevention in the budding stage of hidden dangers.
⬆ Bellows Helium Inspection System
3 UQD quick coupling: 100% helium inspection as production standard
UQD (quick change connector) is the key "joint" of the liquid cooling circulation system. It is installed between the Manifold and the server to achieve instantaneous plugging and unplugging under liquid and pressure without leakage.It eliminates the need to drain the entire pipeline of coolant for server maintenance, greatly improving operation and maintenance efficiency and reducing the risk of leakage.
With the explosive growth in the demand for AI computing power, UQD, as a core component for large-scale deployment of liquid cooling, has also seen a significant increase in its market value.
Helium inspection is an important part of reducing coolant leakage in the use of UQD. Many industry leaders such as Danfoss and Parker Hannifin have adopted 100% helium inspection as the production standard for UQD. The UQD Interoperability Alliance promoted by Intel and industry partners emphasizes the establishment of a unified and reliable quality control system through high-precision helium inspection.
Through helium inspection, microscopic defects in joint sealing surfaces, welds or materials can be effectively identified, ensuring that UQD still maintains excellent sealing performance under the harsh working conditions of high-pressure and high-frequency plugging and unplugging, thereby significantly reducing the risk of coolant leakage and improving the safety, stability and operation and maintenance efficiency of the data center liquid cooling system.
⬆ Liquid-cooled quick connector automated testing equipment
4Hirays HM-400: escorting "zero leakage" of liquid cooling
In response to the high-standard testing requirements for secondary-side liquid cooling components, the HM-400 helium mass spectrometer leak detector independently developed by Hirays has multiple advantages and can meet the requirements for high-precision and efficient sealing testing of liquid-cooling system components.
At present, the leak detection solution with HM-400 as the core has been recognized by many leading customers in the liquid cooling industry, and has formed multiple sets of standardized detection solutions that can be directly delivered.
⬆ Helium detection test platform
Core advantages
Ultra-high precision:The minimum detectable leak rate can reach 5×10⁻¹³ Pa·m³/s, easily covering the strict leakage rate requirements of cold plates, Manifolds, UQD, etc.
Flexible leak detection:Supports positive and negative pressure leak detection, automatically tracks background, and adapts to the testing needs of different components and working conditions.
Efficient and stable:The optimized mass spectrometry system and intelligent algorithm ensure rapid response across the full range; the high-reliability vacuum system achieves rapid helium removal and improves production line tempo.
Strong anti-interference:Supports nitrogen vaporization and is not interfered by ambient helium, ensuring the accuracy of detection results.
Industrial grade reliable:It supports 7×24 hours long-term online testing and high-frequency detection every few seconds; the dual iridium wire ion source design can effectively prevent atmospheric impact and ensure long-term stable operation of the equipment.
Customized services:We can provide customized leak detection systems and full-process training services according to customer needs to ensure customers have peace of mind.
Reduce costs and increase efficiency:The core components are completely self-developed, with short delivery time, easy replacement, and fast after-sales response, helping companies reduce quality risks and comprehensive costs.
Maintain "zero leakage" and win-win the era of AI computing power
Today, computing power has become the “new electricity” in the digital era, and data centers are the “new factories.”On the liquid cooling track, whether the company can strictly maintain the "zero leakage" standard for key components such as cold plates, UQD, and Manifolds directly determines the company's advantageous position in technology iteration and mass production competition.
Hirays will continue to use professional sealing testing solutions and work hand in hand with liquid cooling industry partners to provide reliable technical and service support for the safe and stable operation of AI computing power, and jointly welcome the full arrival of the era of high-density and high-reliability AI computing power.
