Efficient heat dissipation GPU server cluster connection structure
By using modular rack design and optimizing heat dissipation, the heat dissipation problem of GPU server clusters was solved, achieving efficient heat dissipation and improved system performance.
Patent Information
- Application Number
- CN202520402539.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Utility models(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-10
- Publication Date
- 2026-02-06
- Estimated Expiration
- 2035-03-10
AI Technical Summary
In the process of improving computing power, the heat dissipation problem of existing GPU server clusters has not been effectively solved, resulting in limited system performance and shortened hardware lifespan.
It adopts a modular cabinet design with an internal layered structure and vertical space. It uses honeycomb ventilation panels and high-power axial fans to accelerate air convection, combined with liquid cooling modules and bidirectional heat sinks, and optimizes the air duct design to improve heat dissipation efficiency.
It achieves efficient heat dissipation, improves heat dissipation uniformity, extends hardware lifespan, and enhances system performance.
Smart Images

Figure CN223885520U_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The utility model relates to server assembly technical field especially relates to a high -efficient heat dissipation GPU server cluster connection structure. BACKGROUND
[0002] According to the Chinese patent number CN116828757A discloses a kind of hoisting high-density cluster cloud computing server, it is related to server assembly technical field, in view of the inconvenient problem of routine operation and maintenance of the server equipment installed in high position, present and propose the following scheme, it includes floor, roof, first server and second server, the first server is located at the top of second server, lifting mechanism is arranged between the first server and floor and roof: two groups of first angle steel and second angle steel are diagonally distributed about first server, and screw rod is arranged in the first angle steel, motor is driven connected with screw rod, and the screw rod is all screw connected with nut, the nut is fixedly connected with first server shell;Translation mechanism is arranged between the second server and floor: floor is parallelly laid with two groups of line rail.
[0003] The above-mentioned comparative document and prior art have the following technical problems: the existing GPU server cluster computing capacity is improved, the heat dissipation problem is increasingly critical, the traditional connection structure often focuses on data transmission, ignores heat dissipation, which limits the system performance and shortens the hardware life. UTILITY MODEL CONTENT
[0004] The utility model aims at solving the shortcomings in the prior art and provides a high-efficiency heat-dissipation GPU server cluster connection structure.
[0005] To achieve the above object, the utility model adopts the following technical scheme: a high-efficiency heat-dissipation GPU server cluster connection structure, comprising a cabinet, the two sides of the cabinet are provided with honeycomb ventilation plates, the top end of the cabinet is provided with an upper heat dissipation piece, the bottom of the cabinet is provided with a lower heat dissipation piece, the inside of the cabinet is provided with a GPU server, and the bottom of the GPU server is provided with a support frame.
[0006] Preferably, the inside of the cabinet is provided with a layered structure, vertical spaces are arranged between the layers, and the vertical spaces are provided with GPU servers.
[0007] Preferably, the inside of the upper heat dissipation piece is provided with an upper high-power axial flow fan, and the inside of the lower heat dissipation piece is provided with a lower high-power axial flow fan.
[0008] Preferably, the outside of the upper heat dissipation piece and the outside of the lower heat dissipation piece are both provided with cabinet bodies, and the cabinet bodies are connected with the cabinet by screws.
[0009] Preferably, both sides of the GPU server are provided with heat dissipation holes, the inside of the GPU server is provided with a bidirectional radiator and a GPU chip, and the top of the GPU chip is provided with a liquid cooling module.
[0010] Preferably, the GPU chip is closely attached to the liquid cooling module, the top of the bidirectional radiator is provided with an air duct, one side of the air duct is provided with an electronic element, and the other side of the electronic element is provided with a GPU chip.
[0011] Preferably, the frame body of the support frame is in a grid shape, and the bottom of the upper heat dissipation member and the top of the lower heat dissipation member are provided with support frames.
[0012] Beneficial effects
[0013] In the utility model, modular cabinet design is adopted, GPU servers are placed in layers in each cabinet, sufficient vertical space is reserved between layers, a natural hot air rising channel is formed, a honeycomb-shaped ventilation plate is used for the cabinet side plate, air circulation area is increased, meanwhile, high-power axial flow fans are installed at the top and bottom of the cabinet, air convection is accelerated, hot air is quickly discharged from the cabinet, and cold air is supplemented from the bottom.
[0014] In the utility model, liquid cooling modules and bidirectional radiators are adopted, GPU servers are internally provided with GPU chips, the GPU chips adopt liquid cooling heat dissipation mode, are closely attached to the chips through customized liquid cooling modules, and carry away a large amount of heat, other electronic elements adopt high-efficiency air cooling heat dissipation, through bidirectional radiator and optimized air duct design, cold air first passes through these relatively low heating elements, and then flows to the GPU area, heat accumulation is avoided, and overall heat dissipation uniformity is improved. BRIEF DESCRIPTION OF DRAWINGS
[0015] Figure 1 It is a perspective view of the utility model;
[0016] Figure 2 It is a side view of the utility model;
[0017] Figure 3 It is a sectional view of the utility model;
[0018] Figure 4 It is an internal structure diagram of a single cabinet of the utility model;
[0019] Figure 5 It is an internal structure diagram of a GPU server of the utility model.
[0020] Legend:
[0021] 1, cabinet; 2, honeycomb ventilation plate; 3, upper heat dissipation piece; 301, upper high-power axial flow fan; 4, lower heat dissipation piece; 401, lower high-power axial flow fan; 5, GPU server; 501, heat dissipation hole; 502, bidirectional radiator; 503, GPU chip; 504, liquid cooling module; 6, support frame. DETAILED DESCRIPTION
[0022] In order to make the technical means, creative features, purposes and effects of the utility model easy to understand, the following further describes the utility model in combination with specific embodiments and drawings, but the following embodiments are only preferred embodiments of the utility model, not all. Based on the embodiments in the embodiments, other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the utility model.
[0023] The specific embodiments of the utility model are described below in combination with the drawings. Embodiment one
[0025] Reference Figures 1-5 The utility model provides a kind of high-efficiency heat dissipation GPU server cluster connection structure, including cabinet 1, the both sides of cabinet 1 are equipped with honeycomb ventilation plate 2, the top of cabinet 1 is equipped with upper heat dissipation piece 3, the bottom of cabinet 1 is equipped with lower heat dissipation piece 4, the inside of upper heat dissipation piece 3 is equipped with upper high-power axial flow fan 301, the inside of lower heat dissipation piece 4 is equipped with lower high-power axial flow fan 401, the outside of upper heat dissipation piece 3 and the outside of lower heat dissipation piece 4 are equipped with cabinet body, and cabinet body is connected between cabinet 1 by screw, the inside of cabinet 1 is equipped with GPU server 5, the inside of cabinet 1 is equipped with layered structure, vertical space is equipped between layer and layer, vertical space inside is equipped with GPU server 5, the both sides of GPU server 5 are equipped with heat dissipation hole 501, the inside of GPU server 5 is equipped with bidirectional radiator 502 and GPU chip 503, the top of GPU chip 503 is equipped with liquid cooling module 504, GPU chip 503 and liquid cooling module 504 are closely adhered, the top of bidirectional radiator 502 is equipped with air duct, one side of air duct is equipped with electronic element, the other side of electronic element is equipped with GPU chip 503, the bottom of GPU server 5 is equipped with support frame 6, the frame body of support frame 6 is in grid, the bottom of upper heat dissipation piece 3 and the top of lower heat dissipation piece 4 are equipped with support frame 6, modular cabinet 1 design is used, GPU server 5 is placed in each cabinet 1 layer, and sufficient vertical space is reserved between layer and layer, to form natural hot air rising channel, cabinet 1 side plate uses honeycomb ventilation plate 2, increase air flow area, while installing high-power axial flow fan at the top and bottom of cabinet 1, air convection is accelerated, to make hot air quickly discharge cabinet 1, and cold air is replenished from bottom. Embodiment two
[0027] ReferenceFigure 4 A new type of interconnection cable between servers is designed, a plurality of high-speed data transmission lines are integrated in the cable, the demand of GPU cluster large data interaction is met, a layer of high-thermal-conductivity silica gel material is wrapped outside the cable, a micro liquid cooling pipe is inserted in the cable, the heat generated by the cable itself is taken away by the warm water discharged from the server through the liquid cooling pipe, the cooperation of data transmission and heat dissipation is realized, and signal attenuation and transmission performance decline caused by overheating of the cable are avoided.
[0028] In summary:
[0029] 1. Modular cabinet 1 is adopted, GPU servers 5 are placed in the cabinet 1 in layers, and vertical space is reserved to form a hot air rising channel, the side plate is a honeycomb-shaped ventilation plate 2, and high-power axial flow fans are installed at the top and the bottom, air convection is accelerated to discharge hot air and supplement cold air;
[0030] 2. The liquid cooling module 504 and the bidirectional radiator 502 are adopted, the GPU chip 503 is cooled by the liquid cooling module 504, other electronic elements are cooled by high-efficiency air cooling, and the bidirectional radiator 502 is used to optimize the air duct design, so that cold air first passes through the low-heat elements and then flows to the GPU area, and the uniformity of heat dissipation is improved.
[0031] In the utility model, unless another definite provision and limitation, first feature is in second feature "on" or "under", can include that first and second features are in direct contact, also can include that first and second features are not in direct contact but contact through other features between them. Moreover, first feature is "on", "above" and "upper surface" of second feature includes that first feature is directly above and obliquely above second feature, or only indicates that the horizontal height of first feature is higher than that of second feature. First feature is "under", "below" and "lower surface" of second feature includes that first feature is directly below and obliquely below second feature, or only indicates that the horizontal height of first feature is less than that of second feature.
[0032] The basic principle, main features and advantages of the utility model are shown and described. It should be understood by those skilled in the industry that the utility model is not limited by the above examples, the above examples and descriptions in the specification are only preferred examples of the utility model, and are not used to limit the utility model, various changes and improvements of the utility model can be made without departing from the spirit and scope of the utility model, and these changes and improvements all fall within the scope of the claimed utility model. The scope of protection of the utility model is defined by the appended claims and their equivalents.
Claims
1. A high-efficiency heat-dissipation GPU server cluster connection structure, comprising a cabinet (1), characterized in that: Both sides of the cabinet (1) are provided with honeycomb ventilation plates (2), the top end of the cabinet (1) is provided with an upper heat dissipation piece (3), the bottom of the cabinet (1) is provided with a lower heat dissipation piece (4), the inside of the cabinet (1) is provided with a GPU server (5), and the bottom of the GPU server (5) is provided with a support frame (6).
2. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 1, characterized in that: The inside of the cabinet (1) is provided with a layered structure, the layers are provided with vertical spaces, and the inside of the vertical spaces is provided with GPU servers (5).
3. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 1, characterized in that: The inside of the upper heat dissipation piece (3) is provided with an upper high-power axial flow fan (301), and the inside of the lower heat dissipation piece (4) is provided with a lower high-power axial flow fan (401).
4. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 3, characterized in that: The outside of the upper heat dissipation piece (3) and the outside of the lower heat dissipation piece (4) are both provided with cabinet bodies, and the cabinet bodies are connected with the cabinet (1) by screws.
5. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 1, characterized in that: Both sides of the GPU server (5) are provided with heat dissipation holes (501), the inside of the GPU server (5) is provided with a bidirectional radiator (502) and a GPU chip (503), and the top of the GPU chip (503) is provided with a liquid cooling module (504).
6. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 5, characterized in that: The GPU chip (503) and the liquid cooling module (504) are closely attached, the top of the bidirectional radiator (502) is provided with an air duct, one side of the air duct is provided with an electronic element, and the other side of the electronic element is provided with a GPU chip (503).
7. The GPU server cluster connection structure with high heat dissipation efficiency according to claim 1, characterized in that: The frame body of the support frame (6) is in a grid shape, and the bottom of the upper heat dissipation piece (3) and the top of the lower heat dissipation piece (4) are both provided with support frames (6).
Citation Information
Patent Citations
Hoisting type high-density cluster cloud computing server
CN116828757A