With the explosive growth of AI large model training and reasoning demand, the thermal management of high power density servers is facing unprecedented challenges. When GPU cluster power exceeds 20kW/cabinet, the cooling efficiency of traditional air-cooled systems is approaching the physical limit, and liquid-cooling technology is becoming the core support for AI computing power infrastructure by virtue of its advantage of improving the cooling capacity per unit volume by more than 100 times.
In the actual application of liquid cooling system, the flow control accuracy directly affects the cooling performance and system safety. The industry is currently facing three major technical pain points.
Dynamic traffic matching imbalance
In the scenario of drastic load fluctuation of the server, the fixed traffic pattern will easily lead to the formation of local hotspots, and the sudden rise of GPU core temperature may trigger computing downclocking or even hardware damage.
Energy efficiency bottleneck
Over-configuring the flow rate to cope with peak heat dissipation demand not only results in a waste of more than 30% of the pump's power consumption, but may also lead to the risk of pipeline leakage due to a surge in system pressure.
Multi-physical field coupling interference
Coolant viscosity with temperature changes generated by the flow measurement drift, as well as high-frequency vibration triggered by the sensor signal distortion, seriously affecting the long-term stability of flow monitoring.
AUDIOWELL has launched a series of flow sensors for AI server liquid cooling. Developed specifically for AI liquid cooling systems, the series of flow sensors adopt ultrasonic and vortex dual-technology architecture to build a full-link flow monitoring system from chip-level heat dissipation to cabinet-level thermal management. With +3% measurement accuracy and millisecond response speed, it achieves a dynamic balance between energy consumption and performance of the heat dissipation system.
Vortex flow sensor for large flow, high flow rate and high viscosity cold coolant scene of the vortex sensor, based on the principle of Kamen vortex effect, through accurate detection of fluid through the vortex generator generated by the vortex frequency, combined with fluid dynamics modelling and signal processing technology, to achieve stable, real-time flow measurement. The product can dynamically adjust the coolant flow rate to ensure continuous and efficient heat dissipation of hardware under high load computing.
Ultrasonic flow sensor is based on the principle of ultrasonic time difference method, through accurate measurement of the time difference between the ultrasonic wave propagation in the coolant downstream and countercurrent, combined with digital signal processing and flow algorithms, to achieve non-interference, high-precision monitoring of the liquid flow, which can be seamlessly integrated in the liquid cooling system of AI servers, real-time control of the coolant flow, to ensure that the GPU/CPU and other core components of the overclocking computing under the high efficiency of the heat dissipation.
Bimodal Synergy Application Scenarios
Transient Thermal Shock Protection
At the moment of GPU cluster startup, the ultrasonic sensor instantly captures the flow pulse signal and links with the inverter pump to complete the flow compensation to control the chip temperature fluctuation in the safe zone.
Closed Loop Energy Efficiency Optimisation
Vortex sensors continuously monitor the main pipe flow baseline, combined with server load profiles to generate dynamic flow profiles to achieve comprehensive energy efficiency improvements in the cooling system.
Redundant Safety and Security
Abnormal flow events can be identified through a dual-sensor cross-validation mechanism that automatically switches the backup circulation loop and triggers an early warning when a micro-leak or bubble build-up is detected.
With the liquid cooling penetration rate moving towards 80%, flow sensing technology is becoming the core pivot of intelligent thermal management, and AUDIOWELL's intelligent flow monitoring solution will continue to drive the evolution of the thermal system towards the integration of ‘sensing-decision-making-execution’ to build a reliable cornerstone for the sustainable growth of the A-calculation power.
Focus on us