Artificial intelligence server device

Abstract

The embodiment of the application discloses a kind of artificial intelligence server equipment, including cabinet and the central processing unit module, big model inference module and heat dissipation system being set in cabinet, heat dissipation system includes liquid cooling heat dissipation component being set in cabinet interval, each liquid cooling heat dissipation component includes water cooling head, water pump, cold row, fan and water pipe;In each liquid cooling heat dissipation component, water cooling head, water pump and cold row are communicated by water pipe and form closed cooling circuit;First water cooling head is in thermal contact with the heating device in central processing unit module by thermal interface material, and second water cooling head is in thermal contact with the heating device in big model inference module by thermal interface material;Cabinet is provided with front air inlet and corresponding air guide channel with each liquid cooling heat dissipation component, and air guide channel guides the airflow introduced by front air inlet to corresponding cold row;In the direction that air current is discharged from inside cabinet to outside, cold row and fan are sequentially arranged, to make fan suck the airflow passing through cold row from the leeward side of cold row.

Description

Technical Field

[0001] This application relates to the field of server heat dissipation technology, and in particular to an artificial intelligence server device. Background Technology

[0002] With the application of artificial intelligence models in local environments, individual users have developed a corresponding demand for AI server devices capable of performing model inference and data processing locally. AI server devices are typically configured with a central processing unit (CPU) module and a large model inference module. The CPU module can handle device scheduling and general computing tasks, while the large model inference module can handle model inference, tensor computation, and neural network computation tasks. Both modules generate a significant amount of heat during operation.

[0003] Compared to server equipment deployed in dedicated server rooms, AI server equipment for personal use typically needs to be placed directly on a desktop or in a location near where people are active. Such deployment environments make it difficult to configure dedicated server room cooling and soundproofing facilities, thus placing higher demands on the equipment's size, heat dissipation capacity, operating noise, dustproof performance, and ease of maintenance.

[0004] Server equipment can employ either air cooling or liquid cooling. Liquid cooling typically uses a water cooling head to absorb heat generated by heat-generating components, and a water pump drives coolant to circulate between the water cooling head and the radiator. A fan then dissipates the heat released by the radiator. However, heat-generating components in AI server equipment are relatively dispersed. While liquid cooling is used for the main heat-generating components, other heat-generating components still require corresponding cooling structures, which can increase the space occupied by the cooling system and the complexity of the internal structure of the equipment.

[0005] Furthermore, within the limited chassis space suitable for personal desktop deployments, the airflow from different radiators can easily interfere with each other, and issues such as unclear airflow paths or localized air circulation may arise inside the chassis. When fans directly blow air onto the radiators and their heatsink fins, the high-speed airflow can also generate significant aerodynamic noise. Therefore, how to effectively dissipate heat from the central processing unit module, large-scale inference module, and related heat-generating components within the limited chassis space, rationally organize the airflow through different radiators, and balance the operating noise and dust prevention performance of the equipment are technical problems that need to be solved. Summary of the Invention

[0006] One objective of this application is to provide an artificial intelligence server device, which at least addresses the problem that in artificial intelligence server devices suitable for personal desktop deployment, the heat generated by the central processing unit module and the large model inference module needs to be dissipated within a limited chassis space, while the airflows flowing through different radiators are prone to mutual interference and generate significant operating noise.

[0007] To achieve the above objectives, some embodiments of this application provide an artificial intelligence server device, including a chassis and a central processing unit module, a large model inference module, and a heat dissipation system disposed within the chassis; The central processing unit module and the large model inference module are located on the same mounting side within the chassis; The heat dissipation system includes a first liquid cooling heat dissipation component and a second liquid cooling heat dissipation component arranged at intervals within the chassis. Each liquid cooling heat dissipation component includes a water cooling head, a water pump, a radiator, a fan, and water pipes. In each liquid cooling heat dissipation component, the water cooling head, the water pump, and the radiator are connected through the water pipes to form a closed cooling loop. The water cooling head of the first liquid cooling heat dissipation component is the first water cooling head, and the water cooling head of the second liquid cooling heat dissipation component is the second water cooling head; the first water cooling head is in thermal contact with the heat-generating device in the central processing unit module through a thermal interface material, and the second water cooling head is in thermal contact with the heat-generating device in the large model inference module through a thermal interface material. The chassis is provided with a front air inlet and air guide channels corresponding to each of the liquid cooling heat dissipation components. The air guide channels guide the airflow introduced by the front air inlet to the corresponding radiator. Along the direction of airflow from the inside of the chassis to the outside, the radiator and the fan are arranged in sequence so that the fan draws the airflow passing through the radiator from the leeward side of the radiator.

[0008] Compared with related technologies, the solution provided in this application embodiment, by setting a first liquid cooling heat dissipation component and a second liquid cooling heat dissipation component corresponding to the central processing unit module and the large model inference module respectively in the chassis, can absorb and discharge the heat generated by the heat-generating devices in the two modules by using the corresponding water cooling heads and closed cooling circuits respectively; by guiding the airflow introduced by the front air inlet to the corresponding radiator through the air guide channel, the flow path of the cooling airflow in the chassis can be limited, reducing the mutual interference of airflows flowing to different radiators; by setting the fan on the leeward side of the radiator and drawing the airflow through the radiator, the aerodynamic noise generated by the high-speed airflow directly impacting the radiator and its heat dissipation fins can be reduced, thereby taking into account the heat dissipation capacity and operating noise of the artificial intelligence server equipment, making it more suitable for personal desktop deployment.

[0009] Furthermore, by forming a heat conduction network connected to the corresponding water-cooling head through the heat dissipation module and heat pipes, heat generated by heat-generating devices that are not in direct thermal contact with the water-cooling head can be transferred to the corresponding closed cooling circuit. By using gravity-type heat pipes, reverse heat transfer from the water-cooling head to heat-generating devices in thermal contact with the heat dissipation module can be suppressed. By isolating the closed cooling circuits from each other in terms of fluid flow and setting up mutually isolated airflow channels, the heat and airflow effects between different liquid cooling components can be reduced. Combined with a detachable dustproof screen, leakage monitoring components, and flexible vibration damping and noise reduction pins, the dustproof, leakage monitoring, and vibration isolation performance of the equipment can also be improved. Attached Figure Description

[0010] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0011] Figure 1 A schematic diagram of the external structure of an artificial intelligence server device from a first-view perspective is provided for some embodiments of this application; Figure 2 A schematic diagram of the external structure of an artificial intelligence server device from a second perspective, provided for some embodiments of this application; Figure 3 This application provides a schematic diagram of the internal structure of an artificial intelligence server device according to some embodiments; Figure 4 This application provides a schematic diagram of the structure of a liquid cooling heat dissipation assembly and a leakage monitoring assembly according to some embodiments; Figure 5 A schematic diagram of a heat conduction network consisting of a heat dissipation module, heat pipes, and a water cooling head, provided for some embodiments of this application; Figure 6 A schematic diagram illustrating the arrangement of the evaporation and condensation sections of a gravity-type heat pipe, provided for some embodiments of this application; Figure 7 This application provides a schematic diagram of the structure of an air guide channel according to some embodiments; Figure 8 This is a schematic diagram of an assembly structure of a metal front panel, slot, and dust filter provided for some embodiments of this application.

[0012] Explanation of reference numerals in the attached figures: Artificial intelligence server equipment 1; chassis 10; front air inlet 11; air duct 12; air outlet 13; metal front panel 14; slot 15; dust filter 16; recess 17; central processing unit module 20; large model inference module 30; first liquid cooling heat dissipation component 41; second liquid cooling heat dissipation component 42; water cooling head 43; water pump 44; radiator 45; fan 46; water pipe 47; heat dissipation module 50; heat pipe 51; evaporation section 52; condensation section 53; linear water immersion sensor 61. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] like Figures 1 to 7 As shown, this embodiment provides an artificial intelligence server device 1, including a chassis 10 and a central processing unit module 20, a large model inference module 30, and a heat dissipation system disposed within the chassis 10. The artificial intelligence server device 1 can be placed vertically on a desktop, workbench, or other supporting surface, or can be placed in other orientations depending on the usage environment.

[0015] The chassis 10 has internal installation space for mounting the central processing unit module 20, the large model inference module 30, and the cooling system. The central processing unit module 20 and the large model inference module 30 are located on the same mounting side within the chassis 10, which facilitates the connection of the cooling system to the central processing unit module 20 and the large model inference module 30 respectively. It also helps to reduce the overall thickness of the chassis 10, making the artificial intelligence server device 1 suitable for desktop deployment.

[0016] In the illustrated embodiment, the central processing unit (CPU) module 20 and the large model inference module 30 are arranged at intervals along the height of the chassis 10. The CPU module 20 includes a CPU and circuitry that works in conjunction with it, while the large model inference module 30 includes computing devices for performing artificial intelligence model inference tasks and corresponding circuitry. The heat-generating device in the CPU module 20 can be a CPU chip, and the heat-generating device in the large model inference module 30 can be a graphics processing unit (GPU), a neural network processor, or other artificial intelligence acceleration devices.

[0017] In one specific implementation, the central processing unit module 20 uses a Huawei Kunpeng processor; the large model inference module 30 uses a Huawei Ascend AI processor.

[0018] The heat dissipation system includes a first liquid cooling component 41 and a second liquid cooling component 42, spaced apart within the chassis 10. The first liquid cooling component 41 is correspondingly configured with the central processing unit (CPU) module 20 and is used to dissipate heat from the heat-generating components in the CPU module 20. The second liquid cooling component 42 is correspondingly configured with the large model inference module 30 and is used to dissipate heat from the heat-generating components in the large model inference module 30. By configuring liquid cooling components for the CPU module 20 and the large model inference module 30 respectively, heat transfer and dissipation can be carried out separately according to the heat generation of each component, reducing the thermal impact between them.

[0019] Both the first liquid cooling heat dissipation assembly 41 and the second liquid cooling heat dissipation assembly 42 include a water cooling head 43, a water pump 44, a radiator 45, a fan 46, and water pipes 47. In each liquid cooling heat dissipation assembly, the water cooling head 43, the water pump 44, and the radiator 45 are connected sequentially via water pipes 47, thereby forming a closed cooling loop for the circulation of coolant. The water pump 44 drives the coolant to circulate within the corresponding closed cooling loop, the water cooling head 43 absorbs the heat generated by the corresponding heat-generating device, and the radiator 45 transfers the heat carried by the coolant to the air flowing through it.

[0020] For ease of distinction, the water-cooling head 43 in the first liquid cooling assembly 41 is referred to as the first water-cooling head, and the water-cooling head 43 in the second liquid cooling assembly 42 is referred to as the second water-cooling head. The first water-cooling head makes thermal contact with the heat-generating devices in the central processing unit module 20 through a thermal interface material, and the second water-cooling head makes thermal contact with the heat-generating devices in the large model inference module 30 through a thermal interface material. The thermal interface material fills the tiny gap between the water-cooling head 43 and the corresponding heat-generating device to reduce the contact thermal resistance between them. The thermal interface material can be thermally conductive silicone grease, thermally conductive pads, phase change thermally conductive materials, or other materials that can improve heat transfer performance.

[0021] When the AI ​​server device 1 is working, the heat generated by the heat-generating devices in the central processing unit module 20 is transferred to the first water-cooling head via the thermal interface material. The coolant in the first liquid cooling assembly 41 absorbs this heat and, driven by the water pump 44, flows along the water pipe 47 to the corresponding radiator 45, where it transfers the heat to the air. Similarly, the heat generated by the heat-generating devices in the large model inference module 30 is transferred to the second water-cooling head via the thermal interface material. The coolant in the second liquid cooling assembly 42 absorbs this heat and, driven by the corresponding water pump 44, flows to the corresponding radiator 45 for heat dissipation.

[0022] The chassis 10 is provided with a front air inlet 11 and air guide channels 12 corresponding to the first liquid cooling heat dissipation component 41 and the second liquid cooling heat dissipation component 42, respectively. The air guide channels 12 are used to receive the external air entering the chassis 10 through the front air inlet 11 and guide the incoming air to the air intake side of the corresponding radiator 45, thereby reducing the disorderly diffusion of airflow inside the chassis 10 and enabling the cooling airflow to flow through the radiator 45 in a more concentrated manner.

[0023] Along the direction of airflow from inside the chassis 10 to the outside, the corresponding radiator 45 and fan 46 are arranged sequentially, with the radiator 45 located on the intake side of the fan 46. When the fan 46 is working, it draws air from the leeward side of the radiator 45, causing the air entering from the front air inlet 11 to flow along the air guide channel 12 to the radiator 45, and then through the radiator 45 before being exhausted to the outside of the chassis 10 by the fan 46. This suction arrangement allows the airflow to pass more evenly through the heat exchange area of ​​the radiator 45, improving the heat exchange efficiency between the radiator 45 and the air.

[0024] Thus, the first liquid cooling heat dissipation component 41 and the second liquid cooling heat dissipation component 42 respectively undertake the heat dissipation tasks of the central processing unit module 20 and the large model inference module 30, and together with the front air inlet 11, the air guide channel 12, the radiator 45 and the fan 46, they form an organized heat dissipation airflow path, which not only meets the heat dissipation capacity required for artificial intelligence computing, but also makes the device have a structural form suitable for deployment in a personal desktop environment.

[0025] In this embodiment, by setting a first liquid cooling component and a second liquid cooling component corresponding to the central processing unit module and the large model inference module respectively in the chassis, the heat-generating components in the two modules can be liquid cooled separately. By guiding the airflow introduced by the front air inlet to the corresponding radiator through the air guide channel, the flow path of the cooling airflow in the chassis can be limited. By placing the fan on the leeward side of the radiator and drawing airflow through the radiator, the aerodynamic noise generated by the high-speed airflow directly impacting the radiator and its heat dissipation fins can be reduced. Thus, the artificial intelligence server device balances heat dissipation capacity and operating noise, making it more suitable for personal desktop deployment.

[0026] In one embodiment, such as Figure 5 As shown, the heat dissipation system also includes multiple heat dissipation modules 50. The heat dissipation modules 50 are in thermal contact with the heat-generating devices in the central processing unit module 20 and the large model inference module 30 that are not in direct thermal contact with the corresponding water cooling head 43. Each heat dissipation module 50 is thermally connected to the corresponding water cooling head 43 through at least one heat pipe 51 to form a heat conduction network composed of heat dissipation modules 50, heat pipes 51 and corresponding water cooling heads 43.

[0027] Specifically, the heat dissipation module 50 includes a thermally conductive base that is in thermal contact with the heat-generating device. The thermally conductive base can be made of copper, aluminum, or a copper-aluminum composite material. Thermal grease, thermal pads, or phase-change thermally conductive materials can be placed between the thermally conductive base and the heat-generating device to fill the tiny gaps between them and reduce contact thermal resistance. The heat dissipation module 50 may also include heat dissipation fins disposed on the thermally conductive base to increase the heat exchange area of ​​the heat dissipation module 50.

[0028] One end of the heat pipe 51 is connected to the heat dissipation module 50, and the other end is connected to the corresponding water cooling head 43. The heat pipe 51 can be connected to the heat dissipation module 50 and the water cooling head 43 by embedding, pressing, welding, or bonding with thermally conductive material. Depending on the heating power of the heat-generating device, a heat dissipation module 50 can be connected to one or more heat pipes 51; when a heat dissipation module 50 is connected to multiple heat pipes 51, the multiple heat pipes 51 can be spaced apart to transfer heat from different locations of the heat dissipation module 50 to the water cooling head 43.

[0029] The heat dissipation module 50 corresponding to the heat-generating devices in the central processing unit module 20 is connected to the first water cooling head of the first liquid cooling assembly 41 via heat pipes 51. The heat dissipation module 50 corresponding to the heat-generating devices in the large model inference module 30 is connected to the second water cooling head of the second liquid cooling assembly 42 via heat pipes 51. The water cooling head 43 can extend outward relative to the heat-generating devices it directly contacts to provide an installation area for connecting multiple heat pipes 51. Multiple heat pipes 51 can be connected at different positions on the water cooling head 43, so that multiple heat-generating devices on the corresponding modules form a branched heat conduction structure.

[0030] During operation, the heat generated by the heat-generating devices that are not in direct thermal contact with the water-cooling head 43 is first transferred to the corresponding heat dissipation module 50, and then transferred to the corresponding water-cooling head 43 through the heat pipe 51, and carried away by the coolant flowing through the water-cooling head 43. Therefore, in addition to absorbing the heat generated by the heat-generating devices that are in direct thermal contact with it, the water-cooling head 43 can also receive the heat transferred through the heat dissipation module 50 and the heat pipe 51.

[0031] In the above embodiments, multiple heat-generating devices are connected to the corresponding water cooling head 43 through the heat dissipation module 50 and heat pipe 51, which can form a centralized heat conduction network, reducing the need to set up fans or independent liquid cooling circuits at different heat-generating devices, so that the heat on the corresponding modules can be discharged in a concentrated manner through the same liquid cooling heat dissipation component, and it is beneficial to reduce the chassis space occupied by the heat dissipation structure.

[0032] In one embodiment, such as Figure 6As shown, heat pipe 51 is a gravity heat pipe. When placed vertically, the evaporation section 52 of the gravity heat pipe 51 is located below the condensation section 53 and is thermally connected to the corresponding heat dissipation module 50. The condensation section 53 is thermally connected to the corresponding water cooling head 43 to suppress reverse heat transfer from the water cooling head 43 to the heat-generating device that is in thermal contact with the heat dissipation module 50.

[0033] Specifically, the gravity heat pipe 51 forms a sealed cavity inside, and a phase change working fluid is placed inside the sealed cavity. The phase change working fluid can be water, alcohol, or a medium suitable for gas-liquid phase change within a predetermined temperature range, depending on the operating temperature of the gravity heat pipe 51. The interior of the gravity heat pipe 51 can also be evacuated to lower the boiling point of the phase change working fluid and improve the phase change heat transfer efficiency.

[0034] When the AI ​​server device 1 is placed vertically in the posture shown in the figure, the evaporation section 52 is located below the condensation section 53. The evaporation section 52 is connected to the corresponding heat dissipation module 50, and the condensation section 53 is connected to the corresponding water cooling head 43, so that a height difference along the direction of gravity is formed between the evaporation section 52 and the condensation section 53.

[0035] When the heat-generating device is working, the heat it generates is transferred to the evaporation section 52 via the heat dissipation module 50. The phase change working fluid absorbs heat in the evaporation section 52 and vaporizes. The vaporized phase change working fluid flows along the internal cavity of the gravity-driven heat pipe 51 to the condensation section 53, where it releases heat and liquefies. The heat released in the condensation section 53 is transferred to the water-cooling head 43 and carried away by the circulating coolant within the water-cooling head 43. The liquefied phase change working fluid flows back to the evaporation section 52 under gravity to begin the next phase change cycle.

[0036] When the temperature of the water cooling head 43 and the condensing section 53 is higher than the temperature of the evaporating section 52, the condensation process of the phase change working fluid in the condensing section 53 and the reflux process after liquefaction are suppressed, thereby reducing the heat in the water cooling head 43 from being transferred in the reverse direction through the gravity heat pipe 51 to the heat dissipation module 50 and the heat-generating device that is in thermal contact with the heat dissipation module 50.

[0037] In the above embodiments, by utilizing the height difference between the evaporation section 52 and the condensation section 53 and the gravity reflux characteristics of the phase change working fluid, directional heat transfer can be achieved without setting mechanical valves or active control components, suppressing the reverse heat transfer from the high-temperature water cooling head 43 to the low-temperature heat-generating device, and at the same time, no additional driving force is required for the gravity heat pipe 51.

[0038] In one embodiment, such as Figure 3 As shown, the closed cooling circuit of the first liquid cooling heat dissipation component 41 and the closed cooling circuit of the second liquid cooling heat dissipation component 42 are fluidly isolated from each other, and each water pump 44 drives the coolant circulation in the corresponding closed cooling circuit.

[0039] Specifically, the first water cooling head, water pump 44, and radiator 45 in the first liquid cooling heat dissipation assembly 41 are connected by corresponding water pipes 47 to form a first closed cooling circuit; the second water cooling head, water pump 44, and radiator 45 in the second liquid cooling heat dissipation assembly 42 are connected by corresponding water pipes 47 to form a second closed cooling circuit. No connecting pipes for coolant to circulate between the first and second closed cooling circuits are provided.

[0040] The coolant in each closed cooling loop may include deionized water and a corrosion inhibitor. The deionized water is used to absorb and transport the heat received by the water block 43, and the corrosion inhibitor is used to mitigate corrosion of the water block 43, radiator 45, and pipe joints. The water pipes 47 may be flexible hoses or, in some locations, rigid pipes, and are sealed to the water block 43, water pump 44, and radiator 45 via pipe joints.

[0041] The two water pumps 44 can be selected with rated flow rate and head according to the heat generation power of the central processing unit module 20 and the large model inference module 30, respectively. The two water pumps 44 can also adjust their operating speeds to ensure that the first closed cooling circuit and the second closed cooling circuit have coolant circulation flow rates that match the corresponding heat dissipation loads.

[0042] In the above embodiment, by setting up two isolated closed cooling circuits, the heat dissipation loads of the central processing unit module 20 and the large model inference module 30 can be respectively handled, avoiding direct mixing of heat from the two modules in the coolant. When the workload of one module changes, the operating state of the corresponding water pump 44 can be adjusted without changing the fluid state of the other closed cooling circuit, which helps to improve the flexibility of heat dissipation adjustment and operational reliability.

[0043] In one embodiment, the air ducts 12 are spaced apart along the height direction of the chassis 10, and each air duct 12 is formed by a channel wall extending from the front air inlet 11 to the air inlet side of the corresponding radiator 45; the inner cavities of each air duct 12 are isolated from each other so that the airflow introduced by the front air inlet 11 is diverted and flows to the corresponding radiator 45 along the corresponding air duct 12.

[0044] Specifically, the two air ducts 12 can be arranged vertically within the chassis 10. The upper air duct 12 corresponds to one of the radiators 45, and the lower air duct 12 corresponds to the other radiator 45. Each air duct 12 has an air inlet end that communicates with the front air inlet 11 and an air outlet end that corresponds to the air inlet side of the corresponding radiator 45.

[0045] The channel wall of the air guide duct 12 can be made of metal plate, engineering plastic plate or composite plate, or it can be formed by the structural plate set in the chassis 10 and the chassis wall together. A sealing strip can be installed at the position where the channel wall meets the chassis 10 or the radiator 45 to reduce airflow leakage from the gap between the air guide duct 12 and the adjacent structure.

[0046] The air outlet of the air guide channel 12 can be configured with a corresponding cross-sectional size according to the windward area of ​​the radiator 45, so that the airflow output by the air guide channel 12 covers the main heat exchange area of ​​the radiator 45. The air guide channel 12 can also gradually change its cross-sectional shape from the air inlet to the air outlet, so that the air inlet area of ​​the front air inlet 11 and the windward area of ​​the radiator 45 can transition smoothly, reducing the local vortices formed by sudden changes in airflow direction.

[0047] In the above embodiment, the mutually isolated air guide channels 12 can limit the flow direction of each cooling airflow, prevent airflows flowing to different radiators 45 from crossing each other, and reduce eddies and hot air recirculation inside the chassis 10. The air entering from the front air inlet 11 can pass through the corresponding radiator 45 in a more concentrated manner, thereby improving the utilization rate of the air pressure and air volume generated by the fan 46, and helping to reduce noise caused by airflow turbulence.

[0048] In one embodiment, such as Figure 8 As shown, the metal front panel 14 of the chassis 10 is provided with a slot 15 corresponding to the front air inlet 11. A removable dust filter 16 covering the front air inlet 11 is installed in the slot 15. The dust filter 16 is magnetically attached to the metal front panel 14.

[0049] Specifically, the slot 15 extends along the height of the metal front panel 14 and forms an opening on the outside of the chassis 10 for inserting and removing the dust filter 16. Limiting edges are provided on both sides of the slot 15 so that the dust filter 16 can move in a predetermined direction after being inserted, and restrict the dust filter 16 from disengaging in a direction perpendicular to the metal front panel 14.

[0050] The dust filter 16 includes a mesh filter section and a support frame disposed around the periphery of the mesh filter section. The mesh filter section can be made of metal mesh, nylon mesh, or porous filter material, and the support frame is used to maintain the shape of the dust filter 16. Magnetic strips or multiple magnetic elements can be provided on the support frame so that the dust filter 16 can be magnetically attached to the metal front panel 14 after being inserted into the slot 15.

[0051] The dust filter 16 can also be provided with a pull part for easy gripping. When cleaning or replacing, the user can pull the dust filter 16 from the outside of the chassis 10, so that the dust filter 16 overcomes the magnetic force and moves out along the slot 15; after cleaning, the dust filter 16 is inserted into the slot 15 and attached to the metal front panel 14.

[0052] In the above embodiment, the dust filter 16 is guided and positioned by the slot 15, and magnetic force is used to hold the dust filter 16 on the metal front panel 14. The dust filter 16 can be removed and installed without disassembling the chassis 10 and without using special tools. The dust filter 16 covers the concentrated front air intake 11, which can reduce the amount of dust entering the chassis 10 and adhering to the central processing unit module 20, the large model inference module 30 and the radiator 45, while also facilitating daily maintenance by individual users.

[0053] In one embodiment, the chassis 10 is provided with an air outlet 13 corresponding to each radiator 45, and each radiator 45 is positioned close to the corresponding air outlet 13. The output air pressure of the fan 46 at the working air volume is not less than the airflow pressure drop generated by the corresponding radiator 45 at the working air volume.

[0054] Specifically, the radiator 45 is positioned on the intake side of the fan 46, and the fan 46 is positioned between the radiator 45 and the corresponding exhaust port 13. When the fan 46 is running, it draws air from the leeward side of the radiator 45, causing the air to pass sequentially through the radiator 45 and the fan 46, and then exit the chassis 10 through the corresponding exhaust port 13. The distance between the radiator 45 and the exhaust port 13 can be determined based on the installation thickness of the fan 46 to shorten the exhaust path from the radiator 45 to the exhaust port 13.

[0055] The radiator 45 can be a 240mm or 280mm radiator. A 240mm radiator can be equipped with two 120mm diameter fans 46, arranged along the length of the radiator 45. A 280mm radiator can be equipped with two 140mm diameter fans 46, also arranged along the length of the radiator 45. When other radiator sizes are used, the dimensions, number, and mounting hole positions of the fans 46 correspond to the dimensions and frontal area of ​​the radiator 45.

[0056] The first liquid cooling component 41 and the second liquid cooling component 42 can use the same radiator 45, or they can use different radiators 45 depending on the heat output of the central processing unit module 20 and the large model inference module 30. For example, the module with higher heat output can be configured with a 280mm radiator and two 140mm fans 46, while the other module can be configured with a 240mm radiator and two 120mm fans 46, in order to balance heat exchange capacity and space utilization within the chassis 10.

[0057] When selecting fan 46, the airflow pressure drop generated by radiator 45 can be determined based on the density of its heat dissipation fins, the thickness of radiator 45, and the predetermined operating airflow. The appropriate fan 46 can then be selected based on its pressure-volume characteristic curve, ensuring that the output air pressure of fan 46 at the operating airflow can overcome the airflow resistance formed by radiator 45 and its heat dissipation fins. Fan 46 can be selected with higher static pressure and lower operating speed to meet the ventilation requirements of radiator 45 and control aerodynamic noise.

[0058] In the above embodiment, the radiator 45 and fan 46 are combined in a corresponding configuration, which allows the airflow generated by fan 46 to cover the heat exchange area of ​​radiator 45, reducing the problem of insufficient local airflow in radiator 45. The radiator 45 is positioned close to the air outlet 13, and the fan 46 is located on the leeward side of the radiator 45, which shortens the exhaust path, reduces the recirculation of hot air within the chassis 10, and reduces the noise generated by high-speed airflow directly impacting the radiator 45 and its heat dissipation fins.

[0059] In one embodiment, such as Figure 4 As shown, the heat dissipation system also includes a leakage detection component, which includes a linear water immersion sensor 61 and a controller electrically connected to the linear water immersion sensor 61. The linear water immersion sensor 61 is wrapped around the outside of the corresponding water pipe 47 along the extension direction of each water pipe 47. The controller is not shown in the figure.

[0060] Specifically, the linear water immersion sensor 61 has a flexible linear structure and can be spirally wound along the water pipe 47. When the linear water immersion sensor 61 passes through the connection position of the water pipe 47 with the water cooling head 43, the water pump 44 and the radiator 45, the distance between adjacent winding sections can be reduced, so that there is a denser detection area near the pipe joints that are prone to leakage.

[0061] The linear water immersion sensor 61 has two normally insulated conductive detection wires inside. When there is no leakage in the water pipe 47, the resistance between the two conductive detection wires is within the resistance range corresponding to the dry state, and the controller determines that the closed cooling circuit is in a normal state based on this.

[0062] When water pipe 47 or its joint leaks, the leaking coolant comes into contact with the linear water immersion sensor 61, creating a conductive path between the two conductive detection wires. Once the controller detects a decrease in resistance between the two conductive detection wires, it can output a leak alarm message in the form of an audible, visual, or electrical signal. The controller can also connect to the management system of the artificial intelligence server device 1 to record the leak status, reduce the operating power of the corresponding heat-generating device, or stop the water pump 44 in the closed cooling circuit where the leak occurred.

[0063] In the above embodiment, the linear water immersion sensor 61 extends along and surrounds the water pipe 47, enabling continuous detection of the water pipe 47 and its connection points. Compared to point sensors located at fixed positions, it can detect localized leaks earlier. Upon detecting a leak, the controller outputs an alarm or controls the corresponding components to stop operating, reducing the risk of further coolant leakage damaging electronic components inside the chassis 10.

[0064] In one embodiment, one of the air outlets 13 is a lower air outlet; when the chassis 10 is placed vertically on the support surface, the lower air outlet is located on the lower side of the chassis 10 facing the support surface. The lower edge of the chassis 10 is provided with an upwardly recessed portion 17 corresponding to the lower air outlet. The recessed portion 17 separates the lower air outlet from the support surface and forms an exhaust space between the lower air outlet and the support surface.

[0065] Specifically, the recess 17 can extend along the thickness direction of the chassis 10, and the cross-section of the recess 17 can be arc-shaped, trapezoidal, rectangular, or formed by connecting multiple inclined surfaces. The lower edges of the chassis 10 on both sides of the recess 17 form support parts. When the artificial intelligence server device 1 is placed vertically, the support parts contact the support surface, so that the inner wall of the recess 17 and the support surface are kept apart.

[0066] The recess 17 is positioned opposite the lower air outlet, and the exhaust space formed by the recess 17 is connected to both the lower air outlet and the outside of the chassis 10. When the fan 46 corresponding to the lower air outlet is running, it exhausts the air passing through the corresponding radiator 45 into the exhaust space through the lower air outlet, and the airflow is then exhausted to the external environment from one or both sides of the recess 17.

[0067] The width and depth of the recess 17 can be determined according to the size of the lower air outlet, the working air volume of the fan 46, and the support stability of the chassis 10, so that the exhaust space formed by the recess 17 can meet the exhaust requirements of the lower air outlet, and the chassis structure located on both sides of the recess 17 can provide stable support for the artificial intelligence server device 1.

[0068] In the above embodiment, the recess 17 can reserve exhaust space for the lower air outlet when the chassis 10 is placed vertically, avoiding obstruction of the lower air outlet by the desktop or other supporting surfaces. This structure can maintain the stability of the chassis 10 when placed vertically while ensuring unobstructed lower exhaust path, which is beneficial to maintaining the ventilation and heat dissipation capacity of the corresponding radiator 45.

[0069] In one embodiment, the fan 46 and the water pump 44 are respectively mounted on their respective mounting bases by flexible shock-absorbing and noise-reducing nails made of elastic material, so as to flexibly isolate the fan 46 and the water pump 44 from their respective mounting bases. The flexible shock-absorbing and noise-reducing nails are not shown in the figure.

[0070] Specifically, the mounting base for fan 46 can be a fan bracket, a radiator bracket, or a mounting plate of chassis 10, and the mounting base for water pump 44 can be a water pump bracket or a mounting plate installed inside chassis 10. Flexible vibration damping and noise reduction nails can be made of silicone, rubber, or thermoplastic elastic materials.

[0071] The flexible vibration damping and noise reduction nail includes a connecting section and limiting parts at both ends of the connecting section. During installation, the connecting section of the flexible vibration damping and noise reduction nail passes through the mounting hole of the fan 46 or water pump 44 and the corresponding connecting hole on the mounting base. The limiting parts at both ends are located outside the mounting hole and the connecting hole, respectively, thereby preventing the fan 46 or water pump 44 from detaching from the mounting base.

[0072] The connecting section of the flexible vibration damping and noise reduction pin can undergo tensile, compressive, or bending deformation during the operation of the fan 46 or water pump 44, maintaining a flexible connection between the fan 46 and water pump 44 and their corresponding mounting base. The material hardness, connecting section diameter, and connecting section length of the flexible vibration damping and noise reduction pin can be determined based on the weight, vibration amplitude, and vibration frequency of the fan 46 or water pump 44. In one specific embodiment, the flexible vibration damping and noise reduction pin can be made of silicone or rubber material with a Shore A hardness of 20 to 60, balancing connection strength and elastic deformation capability.

[0073] In the above embodiments, flexible vibration damping and noise reduction nails can replace the rigid connection formed by screws, so that the vibration generated by the operation of fan 46 and water pump 44 is preferentially converted into the elastic deformation of flexible vibration damping and noise reduction nails, and the viscoelastic properties of elastic materials are used to consume part of the vibration energy, thereby reducing the transmission of vibration to the mounting base and chassis 10, suppressing chassis structural resonance and friction and collision noise caused by hard connection.

[0074] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. An artificial intelligence server device, comprising a chassis and a central processing unit module, a large model inference module, and a heat dissipation system disposed within the chassis, characterized in that, The central processing unit module and the large model inference module are located on the same mounting side within the chassis; The heat dissipation system includes a first liquid cooling heat dissipation component and a second liquid cooling heat dissipation component arranged at intervals within the chassis. Each liquid cooling heat dissipation component includes a water cooling head, a water pump, a radiator, a fan, and water pipes. In each liquid cooling heat dissipation component, the water cooling head, the water pump, and the radiator are connected through the water pipes to form a closed cooling loop. The water cooling head of the first liquid cooling heat dissipation component is the first water cooling head, and the water cooling head of the second liquid cooling heat dissipation component is the second water cooling head; the first water cooling head is in thermal contact with the heat-generating device in the central processing unit module through a thermal interface material, and the second water cooling head is in thermal contact with the heat-generating device in the large model inference module through a thermal interface material. The chassis is provided with a front air inlet and air guide channels corresponding to each of the liquid cooling heat dissipation components. The air guide channels guide the airflow introduced by the front air inlet to the corresponding radiator. Along the direction of airflow from the inside of the chassis to the outside, the radiator and the fan are arranged in sequence so that the fan draws the airflow passing through the radiator from the leeward side of the radiator.

2. The artificial intelligence server device according to claim 1, characterized in that, The heat dissipation system also includes multiple heat dissipation modules. Each heat dissipation module is in thermal contact with the heat-generating devices in the central processing unit module and the large model inference module that are not in direct thermal contact with the corresponding water cooling head. Each heat dissipation module is thermally connected to the corresponding water cooling head through at least one heat pipe to form a heat conduction network consisting of the heat dissipation module, the heat pipe and the corresponding water cooling head.

3. The artificial intelligence server device according to claim 2, characterized in that, The heat pipe is a gravity heat pipe, with the evaporation section located below the condensation section and thermally connected to the corresponding heat dissipation module. The condensation section is thermally connected to the corresponding water cooling head to suppress reverse heat transfer from the water cooling head to the heat-generating device that is in thermal contact with the heat dissipation module.

4. The artificial intelligence server device according to claim 1, characterized in that, The closed cooling circuit of the first liquid cooling heat dissipation component and the closed cooling circuit of the second liquid cooling heat dissipation component are fluidly isolated from each other, and each water pump drives the coolant circulation in the corresponding closed cooling circuit.

5. The artificial intelligence server device according to claim 1, characterized in that, The air guide channels are spaced apart along the height of the chassis, and each air guide channel is formed by a channel wall extending from the front air inlet to the air inlet side of the corresponding radiator; the inner cavities of each air guide channel are isolated from each other so that the airflow introduced by the front air inlet is diverted and flows to the corresponding radiator along the corresponding air guide channel.

6. The artificial intelligence server device according to claim 1, characterized in that, The metal front panel of the chassis is provided with a slot corresponding to the front air inlet. A removable dust filter covering the front air inlet is installed in the slot. The dust filter is magnetically attached to the metal front panel.

7. The artificial intelligence server device according to claim 5, characterized in that, The chassis is provided with an air outlet corresponding to each of the radiators. Each radiator is positioned close to its corresponding air outlet. The output air pressure of the fan at the working air volume is not less than the airflow pressure drop generated by the corresponding radiator at the working air volume.

8. The artificial intelligence server device according to claim 7, characterized in that, One of the air outlets is a lower air outlet; when the chassis is placed vertically on the support surface, the lower air outlet is located on the lower side of the chassis facing the support surface. The lower edge of the chassis has an upwardly recessed portion corresponding to the lower air outlet. The recessed portion separates the lower air outlet from the support surface and forms an exhaust space between the lower air outlet and the support surface.

9. The artificial intelligence server device according to claim 1, characterized in that, The heat dissipation system also includes a leakage detection component, which includes a linear water immersion sensor and a controller electrically connected to the linear water immersion sensor. The linear water immersion sensor is wrapped around the outside of the corresponding water pipe along the extension direction of each water pipe.

10. The artificial intelligence server device according to claim 1, characterized in that, The fan and the water pump are respectively mounted on their respective mounting bases using flexible shock-absorbing and noise-reducing nails made of elastic material, so as to flexibly isolate the fan and the water pump from their respective mounting bases.

Images