Control method, device and equipment of server liquid cooling heat dissipation system and storage medium

Abstract

The invention relates to a control method, device and equipment for a liquid cooling heat dissipation system of a server and a storage medium, and belongs to the field of radiators, and the control method for the liquid cooling heat dissipation system of the server comprises the steps that CPU heat load data, liquid cooling heat dissipation system state data and liquid cooling heat dissipation system surrounding environment data are obtained, and the CPU heat load trend and grade are generated; and in response to the thermal load trend and grade of the CPU, adjusting one or more parameters of the flow of the cooling liquid in the branches, the energy consumption of the circulating pump and the energy consumption of the heat dissipation terminal, so that the temperature of the CPU is maintained in a reasonable range. According to the control method of the server liquid cooling heat dissipation system, heat dissipation resources are finely managed and allocated, energy consumption of the circulating pump and the heat dissipation terminal is adjusted in real time, the temperature of each CPU is independently monitored and controlled, power consumption waste of the circulating pump and the heat dissipation terminal is avoided, the heat dissipation efficiency of the liquid cooling heat dissipation system is effectively improved, and energy consumption of the liquid cooling heat dissipation system is reduced.

Description

Technical Field

[0001] This application relates to the field of heat sinks, and in particular to a control method, apparatus, equipment and storage medium for a server liquid cooling system. Background Technology

[0002] With the rapid development of artificial intelligence, big data, and cloud computing technologies, AI servers, as core computing power carriers, are iteratively upgrading towards multi-chip integration, high-density deployment, and high-performance computing. To meet the high-intensity computing power demands of large-scale deep learning training and massive data inference, AI servers typically integrate multiple high-performance CPUs and dedicated computing chips (such as GPUs and NPUs). The chip's operating frequency and core density are continuously increasing, leading to an exponential increase in heat load per unit volume. Related data shows that the power consumption of a single high-end AI computing chip has exceeded 300W, and the overall heat dissipation of a multi-chip integrated AI server can reach several kilowatts, with a heat density far exceeding that of traditional general-purpose servers. This places stringent requirements on the heat dissipation efficiency, temperature control accuracy, and stability of the cooling system.

[0003] Currently, AI server cooling solutions are mainly divided into two categories: air cooling and liquid cooling. Air cooling, with its advantages of simple structure and low cost, was once widely used in low-to-medium load server scenarios. However, due to the physical characteristics of low air convection heat transfer coefficient, it is prone to heat dissipation bottlenecks in high-density, high-heat-load AI servers. It cannot effectively solve the problem of local hot spots caused by multi-chip parallel computing, leading to chips triggering frequency reduction protection due to excessive temperature, or even hardware aging due to high temperature, which seriously restricts the release of computing power.

[0004] Compared to air cooling, liquid cooling, based on the advantages of high heat transfer efficiency and good temperature uniformity of liquid convection, can significantly improve heat dissipation capacity and effectively control the overall temperature and local hot spots of multi-chip clusters, and has become the mainstream heat dissipation solution for high-end AI servers.

[0005] However, existing liquid cooling systems for AI servers mostly adopt "constant speed control" or "simple threshold-triggered passive control" logic. Their core design philosophy prioritizes "meeting the heat dissipation requirements of extreme heat loads," lacking dynamic balance optimization between heat dissipation efficiency and system power consumption. In actual operation, this exposes many technical defects, making it difficult to adapt to the operating characteristics of AI servers, which have large load fluctuations and uneven heat load distribution. Specific problems are as follows: Firstly, the allocation of heat dissipation resources is inefficient, resulting in both localized overheating and energy waste. Existing liquid cooling systems mostly employ a fixed allocation strategy of "evenly distributing coolant flow," failing to consider the varying thermal loads of multiple CPUs and computing chips under different task scenarios. This leads to localized overheating in chips with higher thermal loads, while coolant flow is redundant in branches containing chips with lower thermal loads, resulting in wasted power consumption by the circulating pump. Simultaneously, the circulating pumps and radiator fans often operate at constant speeds, or their speeds are triggered by a single temperature threshold. Regardless of whether the system is under low, medium, or high load, they maintain a high output of cooling power, failing to dynamically adjust according to actual heat load requirements. This results in high overall power consumption and low heat dissipation efficiency (heat dissipation power / system power consumption) for the liquid cooling system.

[0006] Secondly, the passive control response is lagging, and temperature overshoot affects the stability of computing power. The load scenarios of AI servers (such as model training startup, multi-task switching, and computing cluster scheduling) are highly variable, and the thermal load can surge from a low load range to a high load range in a short period of time. However, the existing passive control logic needs to wait for the temperature sensor to detect that the temperature exceeds the limit before it starts adjusting the heat dissipation parameters. Due to the physical response delay (50-200ms) of the actuators such as circulation pumps, radiator fans, and flow valves, temperature overshoot is prone to occur, causing frequent oscillations in the heat dissipation system and poor stability. Summary of the Invention

[0007] To address the problems of existing server liquid cooling systems, such as crude control, high energy consumption, easy temperature overshoot, frequent system oscillations, and poor stability, this application provides a control method, device, equipment, and storage medium for a server liquid cooling system.

[0008] In one aspect of this disclosure, a control method for a server liquid cooling system is provided. The liquid cooling system includes a coolant circulation pipeline, a circulation pump, a flow valve, and a heat dissipation terminal. The coolant circulation pipeline includes a main line and multiple branch lines, the branch lines being used to cool multiple CPUs of the server. The circulation pump is used to drive the coolant circulation in the pipeline. The flow valve is used to regulate the coolant flow rate in the branch lines. The heat dissipation terminal is used to cool the coolant in the main line. The control method includes: Acquire CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels; In response to CPU thermal load trends and levels, one or more parameters among several branches, such as coolant flow rate, circulation pump energy consumption, and heat dissipation terminal energy consumption, are adjusted to keep the CPU temperature within a reasonable range.

[0009] By adopting the above technical solutions, through refined management and allocation of heat dissipation resources, and real-time adjustment of the energy consumption of the circulation pump and heat dissipation terminal, the temperature of each CPU can be independently monitored and controlled, avoiding waste of power consumption of the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system and reducing the energy consumption of the liquid cooling system.

[0010] Preferably, the CPU thermal load data includes: average CPU core temperature, CPU hot spot temperature, CPU real-time power consumption, and CPU load rate; The liquid cooling system status data includes: coolant flow rate in the main circuit and temperature before and after entering and exiting the branch circuits, coolant flow rate in each branch circuit and temperature before and after entering and exiting the CPU, coolant temperature before and after entering and exiting the heat dissipation terminal, and energy consumption of the circulation pump and heat dissipation terminal. The environmental data surrounding the liquid cooling system includes: ambient temperature and humidity.

[0011] By adopting the above technical solutions, through refined management and allocation of heat dissipation resources, and real-time adjustment of the energy consumption of the circulation pump and heat dissipation terminal, the temperature of each CPU can be independently monitored and controlled, avoiding waste of power consumption of the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system and reducing the energy consumption of the liquid cooling system.

[0012] Preferably, the heat dissipation terminal has a fan, and the energy consumption of the heat dissipation terminal includes the fan speed. The environmental data around the liquid cooling system also includes the fan's intake air temperature.

[0013] Preferably, the CPU thermal load data and liquid cooling system status data are acquired more frequently than the liquid cooling system surrounding environment data. The higher the CPU load, the more frequently CPU thermal load data and liquid cooling system status data are acquired.

[0014] By adopting the above technical solutions, the control precision of the liquid cooling system is enhanced, and the heat dissipation of the CPU is better guaranteed.

[0015] Preferably, the step of acquiring CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels includes: Establish a CPU thermal load level system, and set the trigger threshold and corresponding control strategy for each level; Acquire CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system; adjust the trigger threshold and corresponding control strategy for each level; and determine the current level of CPU thermal load. Based on the acquired CPU thermal load data, predict the CPU thermal load trend.

[0016] By adopting the above technical solution, the CPU thermal load trend can be predicted in advance based on the acquired CPU thermal load data. Then, the heat dissipation capacity of the liquid cooling system can be adjusted in advance to match the upcoming CPU thermal load. This eliminates the temperature overshoot caused by the physical response delay of the execution components and the problems of frequent oscillation and poor stability of the cooling system, thereby improving the stability of the liquid cooling system and ensuring the stability of CPU computing power.

[0017] Preferably, predicting the CPU thermal load trend based on the acquired CPU thermal load data includes: predicting the CPU thermal load trend within the next 3 minutes based on the CPU thermal load data of the past 5 minutes.

[0018] Preferably, the coolant flow rate in the main circuit is in the range of 0.8-1.5 m / s; The temperature difference between the coolant entering and exiting the branch circuit in the main circuit is 5-8℃. The minimum coolant flow rate in the branch circuit shall not be less than 5% of the coolant flow rate in the main circuit; The temperature of the coolant leaving the heat dissipation terminal is 35℃.

[0019] In another aspect of this disclosure, a control device for a server liquid cooling system is provided, comprising: The data acquisition module is configured to collect CPU thermal load data, liquid cooling system status data, and environmental data surrounding the liquid cooling system. The analysis and judgment module is configured to determine the CPU thermal load trend and level based on CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system. The adjustment module is configured to adjust one or more parameters, such as coolant flow rate in the branch, energy consumption of the circulation pump, and energy consumption of the heat dissipation terminal, based on the CPU thermal load trend and level, so as to keep the CPU temperature within a reasonable range.

[0020] By adopting the above technical solutions, through refined management and allocation of heat dissipation resources, and real-time adjustment of the energy consumption of the circulation pump and heat dissipation terminal, the temperature of each CPU can be independently monitored and controlled, avoiding waste of power consumption of the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system and reducing the energy consumption of the liquid cooling system.

[0021] In another aspect of this disclosure, an apparatus is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the aforementioned control method.

[0022] In another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the aforementioned control method.

[0023] Beneficial technical effects: 1. The control method of the server liquid cooling system of this application achieves independent monitoring and control of the temperature of each CPU by finely managing and allocating heat dissipation resources, adjusting the energy consumption of the circulation pump and heat dissipation terminal in real time, avoiding waste of power consumption of the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system, and reducing the energy consumption of the liquid cooling system.

[0024] 2. Based on the acquired CPU thermal load data, the CPU thermal load trend is predicted in advance, and then the heat dissipation capacity of the liquid cooling system is adjusted in advance to match the upcoming CPU thermal load. This eliminates the temperature overshoot caused by the physical response delay of the execution components and the problems of frequent oscillation and poor stability of the cooling system, thereby improving the stability of the liquid cooling system and ensuring the stability of CPU computing power. Attached Figure Description

[0025] Figure 1 This is a flowchart of the control method for a server liquid cooling system.

[0026] Figure 2 This is a flowchart showing the CPU thermal load trend and level generation.

[0027] Figure 3 This is a schematic diagram of the control device for a server's liquid cooling system.

[0028] Explanation of reference numerals in the attached drawings: 2. Control device; 21. Data acquisition module; 22. Analysis and judgment module; 23. Adjustment module. Detailed Implementation

[0029] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and therefore detailed descriptions of them will be omitted. Furthermore, the drawings are merely illustrative of this disclosure and are not necessarily drawn to scale.

[0030] The following is in conjunction with the appendix Figures 1-3 This application will be described in further detail.

[0031] In one aspect of this disclosure, a control method for a server liquid cooling system is proposed. The liquid cooling system includes a coolant circulation pipeline, a circulation pump, a flow valve, and a heat dissipation terminal. The coolant circulation pipeline includes a main pipeline and multiple branch pipelines. The multiple branch pipelines are used to cool multiple CPUs of the server. For example, one branch pipeline cools one CPU. Servers, especially AI servers, generally have multiple CPUs, and the operating temperature of each CPU is different. Therefore, setting up multiple branch pipelines, with one branch pipeline cooling one CPU, achieves refined management of heat dissipation resources.

[0032] Of course, if needed, multiple branches can be set up to cool a single CPU, or one branch can cool multiple CPUs, etc.

[0033] As mentioned above, the circulation pump is used to drive the coolant circulation flow in the coolant circulation pipeline, and the flow valve is used to regulate the coolant flow in the branch. Each branch is equipped with a flow valve to enable precise management and allocation of heat dissipation resources.

[0034] As mentioned above, the heat dissipation terminal is used to cool the coolant in the main circuit. The cooled coolant returns to the main circuit and then re-enters the branch circuit to cool the CPU.

[0035] As an example, the heat dissipation terminal has a fan, which cools the coolant in the main circuit by rotating the fan.

[0036] like Figure 1 As shown, the control method for the server liquid cooling system includes: S1. Obtain CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels; The CPU thermal load data includes: average CPU core temperature, CPU hot spot temperature, CPU real-time power consumption, and CPU load rate. The CPU thermal load data can be obtained through the CPU's built-in temperature sensor and the server's BMC system. The frequency of CPU thermal load data acquisition is positively correlated with the CPU load rate. The higher the CPU load rate, the higher the frequency of CPU thermal load data acquisition. For example, under high load (≥70%), the acquisition frequency is 100ms / time; under medium load (30%-70%), the acquisition frequency is 500ms / time; and under low load (<30%), the acquisition frequency is 1s / time. This design ensures that data acquisition meets the requirements while also taking energy-saving requirements into account, which helps to reduce the energy consumption of the liquid cooling system.

[0037] The liquid cooling system status data includes: coolant flow rate in the main circuit and temperature before and after entering and exiting the branch circuits, coolant flow rate in each branch circuit and temperature before and after entering and exiting the CPU, coolant temperature before and after entering and exiting the heat dissipation terminal, and energy consumption of the circulation pump and heat dissipation terminal. The status data of the liquid cooling system can be acquired through devices such as temperature sensors (NTC / Pt1000), electromagnetic flow meters, Hall effect velocity sensors, and power meters. The acquisition frequency is synchronized with the CPU thermal load data; that is, the higher the CPU load rate, the higher the acquisition frequency of CPU thermal load data and liquid cooling system status data. This design ensures that data acquisition meets requirements while also considering energy conservation, thus reducing the energy consumption of the liquid cooling system.

[0038] The environmental data surrounding the liquid cooling system includes ambient temperature and humidity. Ambient temperature and humidity can be acquired using temperature and humidity sensors. Because environmental fluctuations are slow, the detection frequency can be low, such as 5 seconds per detection, thus ensuring data acquisition meets requirements while also considering energy conservation, thereby reducing the energy consumption of the liquid cooling system.

[0039] As an example, when the heat dissipation terminal has a fan, and the fan rotation is used to cool the coolant in the main circuit, the energy consumption of the heat dissipation terminal also includes the fan speed, and the environmental data around the liquid cooling system also includes the fan intake temperature.

[0040] Specifically, such as Figure 2 As shown, S1, obtaining CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels includes: S11. Establish a CPU thermal load level system, and set the trigger threshold and corresponding control strategy for each level; For example, establish a five-level CPU thermal load rating system, and set the trigger threshold and corresponding control strategy for each level: Thermal load level: Level 1 (Ultra-low load). Trigger threshold: CPU hotspot temperature < 45℃, CPU load rate < 10%. Control strategy: Extreme energy saving, maintaining system operation with minimal power consumption; Thermal load level: Level 2 (low load). Trigger threshold: 45℃ ≤ CPU hotspot temperature < 55℃, 10% ≤ CPU load rate < 30%. Control strategy: Energy saving priority, slight adjustment of airflow / fan speed; Thermal load level: Level 3 (medium load). Trigger threshold: 55℃ ≤ CPU hotspot temperature < 70℃, 30% ≤ CPU load rate < 70%. Control strategy: High efficiency and balance, matching heat dissipation and power consumption as needed; Thermal load level: Level 4 (High load). Trigger threshold: 70℃ ≤ CPU hotspot temperature < 80℃, 70% ≤ CPU load rate < 90%. Control strategy: Prioritize heat dissipation, ensure no localized overheating, and consider energy saving. Thermal load level: Level 5 (Extreme load / Emergency). Trigger threshold: CPU hotspot temperature ≥80℃, CPU load rate ≥90%. Control strategy: Full-power cooling, activate redundant equipment, and prevent CPU throttling / damage.

[0041] S12. Obtain CPU thermal load data, liquid cooling system status data, and surrounding environment data of the liquid cooling system; adjust the trigger threshold and corresponding control strategy for each level; and determine the current level of CPU thermal load. The trigger thresholds and corresponding control strategies for each level are not fixed once set. Instead, they are adjusted specifically based on the acquired CPU thermal load data, liquid cooling system status data, surrounding environmental data of the liquid cooling system, and predicted CPU thermal load trends. This is done to meet the cooling needs of scenarios with sudden changes in CPU thermal load, and to adjust the cooling capacity of the liquid cooling system in advance so that it can match the upcoming CPU thermal load. This eliminates temperature overshoot caused by the physical response delay of the execution components and the problems of frequent oscillations and poor stability of the cooling system, thereby improving the stability of the liquid cooling system and ensuring stable CPU computing power.

[0042] For example, the original thresholds for thermal load level 4 are 70℃≤CPU hotspot temperature<80℃ and 70%≤CPU load rate<90%. If it is predicted that the CPU thermal load trend is that the AI ​​training task is about to start (the thermal load will rise rapidly), then the thresholds are proactively lowered to 68℃≤CPU hotspot temperature<80℃ and 65%≤CPU load rate<90%, and the high-load heat dissipation strategy is activated in advance to offset the response delay of the circulation pump, flow valve, and heat dissipation terminal.

[0043] For example, in winter, when the ambient temperature in the computer room is low and the heat dissipation efficiency of the heat dissipation terminal is high, the threshold of heat load level 4 is raised to 72℃≤CPU hot spot temperature<80℃ and 75%≤CPU load rate<90%, thereby reducing ineffective energy consumption under low load and reducing the energy consumption of the liquid cooling system.

[0044] For example, under a heat load level of 2 (low load), when the ambient temperature in the computer room is high in summer, the control strategy is to "maintain the circulation pump at a medium-low speed + turn on the fan at 50%"; when the temperature in the computer room is low in winter, the strategy is adjusted to "circulation pump at the lowest speed + turn off redundant fans" to further save energy.

[0045] Of course, the CPU thermal load level is not limited to level five and can be set flexibly according to the actual situation.

[0046] S13. Based on the acquired CPU thermal load data, predict the CPU thermal load trend.

[0047] For example, based on the CPU thermal load data of the past 5 minutes, the trend of CPU thermal load in the next 3 minutes can be predicted, and the heat dissipation capacity of the liquid cooling system can be adjusted in advance to match the upcoming CPU thermal load. This eliminates the temperature overshoot caused by the physical response delay of the execution components and the problem of frequent oscillation and poor stability of the cooling system, thereby improving the stability of the liquid cooling system and ensuring the stability of CPU computing power.

[0048] As can be seen, the control method of the server liquid cooling system in this embodiment predicts the CPU thermal load trend in advance and determines the level of thermal load for each CPU. This enables targeted adjustment of the heat dissipation capacity of the liquid cooling system for that CPU in advance, meeting the heat dissipation requirements of scenarios with sudden changes in CPU thermal load. This allows the system to match the upcoming CPU thermal load, eliminates the temperature overshoot caused by the physical response delay of the execution components, and addresses the problems of frequent oscillations and poor stability of the cooling system. This improves the stability of the liquid cooling system and ensures stable CPU computing power.

[0049] Furthermore, by precisely allocating and managing the heat dissipation resources of the liquid cooling system for the specific CPU in advance, the temperature of each CPU can be independently monitored and controlled. The energy consumption of the circulation pump and heat dissipation terminal can be adjusted in real time, avoiding waste of power consumption of the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system and reducing the energy consumption of the liquid cooling system.

[0050] It should be noted that high energy consumption of the circulating pump means high pump power, high speed, and high coolant flow rate in the main circuit; conversely, low energy consumption means low flow rate. The same applies to heat dissipation terminals: high energy consumption of the heat dissipation terminal means high heat dissipation capacity, such as high fan speed; conversely, low energy consumption means low heat dissipation capacity.

[0051] S2. In response to the CPU thermal load trend and level, adjust one or more parameters among several branches, such as coolant flow rate, circulation pump energy consumption, and heat dissipation terminal energy consumption, to keep the CPU temperature within a reasonable range.

[0052] For example, if a CPU heat load is detected to be too high in a certain branch, temperature control is achieved by increasing the coolant flow rate in that branch and / or increasing the heat dissipation capacity of the heat sink to lower the coolant temperature before it enters the CPU. If necessary, the circulation pump speed is increased to ensure that the CPU heat load is reduced to a normal range. Conversely, if a CPU heat load is detected to be too low in a certain branch, the coolant flow rate in that branch is reduced and / or the heat dissipation capacity of the heat sink is reduced to increase the coolant temperature before it enters the CPU. If necessary, the circulation pump speed is reduced to maximize energy savings and avoid wasting energy on the heat sink and circulation pump.

[0053] As can be seen, the control method of the server liquid cooling system in this embodiment manages and allocates the heat dissipation capacity of each branch in a refined manner, so that the heat dissipation capacity provided is adapted to the CPU heat load. When the CPU heat load changes, the heat dissipation capacity of the branch also changes at any time. In this way, the power consumption of the circulation pump and heat dissipation terminal can be avoided, the heat dissipation efficiency of the liquid cooling system can be effectively improved, and the energy consumption of the liquid cooling system can be reduced.

[0054] Preferably, the coolant flow rate in the main circuit is within the range of 0.8-1.5 m / s. A flow rate < 0.8 m / s results in a decrease in the convective heat transfer coefficient, a thickening of the boundary layer, and a decline in heat transfer efficiency. A flow rate > 1.5 m / s leads to a sharp increase in the power consumption of the circulating pump and is prone to pipe cavitation, noise, and accelerated pipe aging. Therefore, while ensuring the coolant flow rate in the main circuit remains within the range of 0.8-1.5 m / s, the heat dissipation capacity of each branch is adjusted by independently regulating the coolant flow rate and the temperature of the coolant before it enters the CPU, thus meeting the CPU's heat dissipation requirements.

[0055] Preferably, the temperature difference between the coolant entering and exiting the branch circuit in the main circuit is 5-8℃. If the temperature difference is too small, it means that the coolant flow in the main circuit is too large, resulting in wasted power consumption of the circulation pump. Conversely, if the temperature difference is too large, it means that the coolant flow in the main circuit is too small, which may lead to insufficient CPU heat dissipation and the risk of local overheating.

[0056] Preferably, the minimum flow rate of coolant in the branch circuit is not less than 5% of the flow rate of coolant in the main circuit; in order to avoid coolant stagnation and pipe scaling due to excessively low flow rate in the branch circuit, which would affect long-term heat exchange efficiency.

[0057] Preferably, the temperature of the coolant leaving the heat dissipation terminal is 35°C. 35°C is lower than the CPU inlet temperature threshold, which can ensure the heat exchange temperature difference and at the same time help to minimize the total power consumption of the fan.

[0058] The control logic when the heat dissipation terminal is a fan is described in detail below as an example.

[0059] When the temperature of the coolant leaving the fan is less than 32°C and the ambient temperature of the computer room is less than 25°C, reduce the fan speed to less than 50%, or even shut down some redundant fans, maintaining only the minimum airflow to minimize the energy consumption of the heat dissipation terminal. When the temperature of the coolant leaving the fan is less than 35℃ and the fan is kept in the high-efficiency speed range (60%-80%), the fan heat dissipation efficiency / power consumption ratio is the highest in this range. When 35℃≤coolant temperature leaving the fan<38℃: fan speed is increased to 80%-100%, and all redundant fans are started at the same time to maximize heat dissipation and ensure CPU cooling. When the temperature of the coolant leaving the fan is ≥38℃ (Level 5 heat load): the fan runs at full speed and sends an alarm to the server BMC system, indicating abnormal data center environment or liquid cooling system failure.

[0060] It should be noted that the fan speed control needs to be linked with the liquid cooling system to avoid energy waste caused by "the fan running at full speed but the coolant temperature being low", and to adapt to changes in the computer room temperature.

[0061] It is evident that when adjusting the heat dissipation capacity of a branch circuit, it is necessary to consider not only CPU thermal load data, but also the liquid cooling system status data and the surrounding environment data. The liquid cooling system status data and the surrounding environment data are equally important for accurately controlling the heat dissipation capacity of the liquid cooling system.

[0062] The following example illustrates the control logic for trunk flow.

[0063] 1. Control objectives: Stabilize the average hot spot temperature of the CPU at the target threshold (70℃, which can be adjusted according to the CPU manufacturer's recommendations), while controlling the temperature difference between the coolant entering and exiting the branch circuit in the main circuit within the optimal range of 5-8℃.

[0064] 2. Control Logic: The system uses "the difference between the average hot spot temperature of the CPU and the target threshold (ΔTtarget)" and "the temperature difference before and after the coolant enters and exits the branch" as inputs and "the speed of the circulating pump" as output. It adopts fuzzy PID to flexibly adjust the control strategy according to the actual working conditions. By analyzing the status of the liquid cooling system in real time, it dynamically adjusts the P, I, and D parameters to achieve the control objective.

[0065] For example, the steps of fuzzy PID control include: Data acquisition and preprocessing: Obtain the difference ΔT between the CPU hotspot temperature and the target value of 70℃, and the rate of change of the deviation (such as the temperature rise / fall per second ΔT / Δt), as input for fuzzy control; Fuzzy inference and parameter correction: Based on preset fuzzy rules, ΔT and ΔT / Δt are fuzzified and judged (e.g., "large ΔT, fast ΔT / Δt → belongs to a high load surge scenario"), and then the corresponding P, I, and D parameter correction amounts are inferred; for example, when the AI ​​server training task starts, the CPU temperature rises rapidly (large ΔT, fast ΔT / Δt), the fuzzy logic will instruct "increase the P parameter (quickly increase the circulation pump speed) and decrease the I parameter (avoid overshoot caused by integral accumulation)" to quickly match the sudden change in heat load. When the CPU temperature approaches the target value (small ΔT, slow ΔT / Δt), the fuzzy logic will instruct "decrease the P parameter (avoid oscillation) and increase the I parameter (eliminate steady-state error)" to maintain temperature stability; PID control and execution output: Using the corrected P, I, and D parameters, the final control quantity (such as the speed of the circulating pump and the speed of the fan) is calculated to drive the action of the actuator. At the same time, feedback data is collected in real time to form a closed-loop control.

[0066] In the control method of the server liquid cooling system in this embodiment, the coolant flow distribution and velocity control in the coolant circulation pipeline are based on the principle of on-demand proportioning. A two-layer strategy of "closed-loop control of total main flow + differentiated distribution of branch flow" is adopted to avoid the inefficiency of "average flow distribution" while controlling the flow velocity within the optimal range, balancing heat exchange efficiency and circulation pump power consumption. Precise distribution is achieved by using an electric proportional flow valve configured in each branch, ensuring that "the higher the heat load, the greater the branch flow."

[0067] For example, if the hot spot temperature of a CPU is >75℃ (close to the alarm threshold), even if its heat load ratio is low, the flow of that branch will be increased first (up to 1.2 times the basic allocation) until the hot spot temperature drops below 70℃ to avoid local overheating (local overheating of the CPU in AI servers can easily lead to a decrease in computing power).

[0068] Simultaneously, the adjustment of the branch flow valve is linked to the global circulation pump speed adjustment to avoid global flow / temperature imbalance caused by the adjustment of a single branch flow, with an adjustment response time of ≤50ms.

[0069] In another aspect of the embodiments of this disclosure, a control device 2 for a server liquid cooling system is provided, such as... Figure 3 As shown, it includes: The data acquisition module 21 is configured to acquire CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system. The analysis and judgment module 22 is configured to determine the CPU heat load trend and level based on CPU heat load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system. The adjustment module 23 is configured to adjust one or more parameters, such as the coolant flow rate in the branch, the energy consumption of the circulating pump, and the energy consumption of the heat dissipation terminal, based on the CPU heat load trend and level, so as to keep the CPU temperature within a reasonable range.

[0070] The control device 2 of the server liquid cooling system in this embodiment manages and allocates heat dissipation resources in a refined manner, adjusts the energy consumption of the circulation pump and heat dissipation terminal in real time, and independently monitors and controls the temperature of each CPU. This avoids wasting power on the circulation pump and heat dissipation terminal, effectively improving the heat dissipation efficiency of the liquid cooling system and reducing its energy consumption. Furthermore, by predicting CPU thermal load trends in advance and classifying the thermal load level for each CPU, the system can proactively adjust the heat dissipation capacity of the liquid cooling system for that CPU. This meets the heat dissipation needs of scenarios with sudden changes in CPU thermal load, ensuring it can match the upcoming CPU thermal load. It eliminates temperature overshoot caused by physical response delays of execution components and the problems of frequent oscillations and poor stability in the cooling system, thus improving the stability of the liquid cooling system and ensuring stable CPU computing power.

[0071] In another aspect of the present disclosure, an apparatus is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the aforementioned control method.

[0072] In another aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the aforementioned control method.

[0073] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," "third," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. "Above," "below," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0074] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A control method for a server liquid cooling system, the liquid cooling system comprising a coolant circulation pipeline, a circulation pump, a flow valve, and a heat dissipation terminal, wherein the coolant circulation pipeline includes a main line and multiple branch lines, the branch lines being used to cool multiple CPUs of the server, the circulation pump being used to drive the coolant circulation in the coolant circulation pipeline, the flow valve being used to regulate the coolant flow in the branch lines, and the heat dissipation terminal being used to cool the coolant in the main line, characterized in that... The control method includes: Acquire CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels; In response to CPU thermal load trends and levels, one or more parameters among several branches, such as coolant flow rate, circulation pump energy consumption, and heat dissipation terminal energy consumption, are adjusted to keep the CPU temperature within a reasonable range.

2. The control method according to claim 1, characterized in that, The CPU thermal load data includes: average CPU core temperature, CPU hot spot temperature, CPU real-time power consumption, and CPU load rate. The liquid cooling system status data includes: coolant flow rate in the main circuit and temperature before and after entering and exiting the branch circuits, coolant flow rate in each branch circuit and temperature before and after entering and exiting the CPU, coolant temperature before and after entering and exiting the heat dissipation terminal, and energy consumption of the circulation pump and heat dissipation terminal. The environmental data surrounding the liquid cooling system includes: ambient temperature and humidity.

3. The control method according to claim 2, characterized in that: The heat dissipation terminal has a fan, and the energy consumption of the heat dissipation terminal includes the fan speed. The environmental data around the liquid cooling system also includes the fan's intake air temperature.

4. The control method according to claim 1, characterized in that: The CPU thermal load data and liquid cooling system status data are acquired more frequently than the liquid cooling system surrounding environment data. The higher the CPU load, the more frequently CPU thermal load data and liquid cooling system status data are acquired.

5. The control method according to claim 1, characterized in that: The process of acquiring CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system to generate CPU thermal load trends and levels includes: Establish a CPU thermal load level system, and set the trigger threshold and corresponding control strategy for each level; Acquire CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system; adjust the trigger threshold and corresponding control strategy for each level; and determine the current level of CPU thermal load. Based on the acquired CPU thermal load data, predict the CPU thermal load trend.

6. The control method according to claim 5, characterized in that: The prediction of CPU thermal load trends based on the acquired CPU thermal load data includes: predicting the CPU thermal load trend within the next 3 minutes based on the CPU thermal load data of the past 5 minutes.

7. The control method according to claim 1, characterized in that: The coolant flow rate in the main circuit ranges from 0.8 to 1.5 m / s; The temperature difference between the coolant entering and exiting the branch circuit in the main circuit is 5-8℃. The minimum coolant flow rate in the branch circuit shall not be less than 5% of the coolant flow rate in the main circuit; The temperature of the coolant leaving the heat dissipation terminal is 35℃.

8. A control device for a server liquid cooling system, characterized in that, include: The data acquisition module is configured to collect CPU thermal load data, liquid cooling system status data, and environmental data surrounding the liquid cooling system. The analysis and judgment module is configured to determine the CPU thermal load trend and level based on CPU thermal load data, liquid cooling system status data, and surrounding environmental data of the liquid cooling system. The adjustment module is configured to adjust one or more parameters, such as coolant flow rate in the branch, energy consumption of the circulation pump, and energy consumption of the heat dissipation terminal, based on the CPU thermal load trend and level, so as to keep the CPU temperature within a reasonable range.

9. An apparatus comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the control method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the control method according to any one of claims 1 to 7.

Images