Modular liquid-cooled backplane for gpu cluster servers

Abstract

The utility model discloses a modularization liquid cooling backplate of GPU cluster server, include: support wall, locking component and heat dissipation component, support wall side fixedly connected with fixed guide rail, the upper slide of fixed guide rail is connected with slide seat, and the slide seat end fixedly connected with the positioning pin, locking component is used for locking the positioning pin limit the slide of slide seat, heat dissipation component is used for improving the heat dissipation efficiency of server GPU, through the cooperation of positioning pin and automatic locking component, in the process of positioning pin insertion through the guide slope guide, utilize the automatic locking of folding frame mechanism of spring drive's limit bump and limit groove, ensure the accuracy and consistency of module installation, also provide the firm mechanical lock, effectively prevent the module loosening caused by vibration or accidental external force.

Description

Technical Field

[0001] This utility model relates to the field of server heat dissipation technology, specifically a modular liquid-cooled backplate for a GPU cluster server. Background Technology

[0002] GPU cluster servers are the core infrastructure of modern high-performance computing and data centers. Their core characteristic lies in integrating a large number of graphics processors (GPUs) into a unified server chassis or rack via a high-speed interconnect network, forming powerful parallel computing capabilities. These servers primarily serve computationally intensive fields such as artificial intelligence training and inference, scientific computing, big data analysis, and graphics rendering. Their computing performance is directly related to the number and density of GPUs. As the power consumption of a single GPU chip continues to climb to hundreds of watts or even higher, the heat generated when multiple high-power GPUs are densely deployed in a limited space is considerable, and traditional air-cooling methods are gradually reaching their limits. Regarding mechanical stability, most existing GPU cluster servers lack reliable automatic locking mechanisms. The connection between modules and the chassis relies on simple pins or screws, which is not only inconvenient to operate but also prone to loosening under continuous vibration during operation, posing a safety hazard of connection failure. Utility Model Content

[0003] The purpose of this invention is to provide a modular liquid-cooled backplate for a GPU cluster server in order to solve the problems mentioned above.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a modular liquid-cooled backplate for a GPU cluster server, comprising: a support wall, a locking component and a heat dissipation component, wherein a fixed guide rail is fixedly connected to the side of the support wall, a sliding seat is slidably connected above the fixed guide rail, and a positioning pin is fixedly connected to the end of the sliding seat. The locking component is used to lock the positioning pin and restrict the sliding of the slide seat; Thermal components are used to improve the heat dissipation efficiency of server GPUs.

[0005] The locking component automatically locks the positioning pin when the slide moves to the predetermined position, thereby reliably restricting the movement of the slide and ensuring the stability of the module in operation. The heat dissipation component is specifically responsible for efficiently dissipating heat from the GPU module. Through optimized heat conduction paths and cooling medium circulation, it significantly improves the server's heat dissipation efficiency.

[0006] As a further embodiment of this utility model: a heat dissipation frame is fixedly connected to the side of the sliding seat, a limit groove is opened on the side of the positioning pin, and a heat dissipation component is provided above the heat dissipation frame; The limiting groove on the side of the locating pin works with the locking component to achieve mechanical interlocking, ensuring the reliability of the connection.

[0007] As a further embodiment of this utility model: the locking assembly includes a locking compartment fixedly connected to the side of the support wall, a pin groove is provided on the inner side of the locking compartment, a placement groove is provided on the inner side of the pin groove, a folding frame is hinged to the side wall of the placement groove, a limiting protrusion that matches the size of the limiting groove is hinged to the side of the folding frame, and a spring is fixedly connected between the two sets of folding plates of the folding frame. The spring connecting the two sets of folding plates provides a continuous clamping force to the locking mechanism, ensuring the stability of the locked state.

[0008] As a further embodiment of this utility model: the heat dissipation assembly includes a fixed bracket fixedly connected to the top of the heat dissipation frame, a water-cooled heat pipe fixedly connected in the middle of the fixed bracket, a docking end fixedly provided on the top of the fixed bracket, a card plate movably connected above the docking end, a heat dissipation copper plate fixedly connected above the card plate, a silicone ring gasket fixedly connected to the upper edge of the heat dissipation copper plate, a silicone grease groove opened on the back of the top of the heat dissipation copper plate, a GPU module placed in the middle area above the heat dissipation copper plate, and the silicone ring gasket attached to the bottom edge area of ​​the GPU module; The silicone ring pad at the upper edge provides both sealing and cushioning and prevents thermal conductive material from overflowing. The thermal grease groove on the back of the heat sink is specifically designed to hold thermal grease, ensuring sufficient thermal contact with the bottom of the GPU module.

[0009] As a further improvement of this utility model: the connection between the heat dissipation frame and the sliding seat is provided with reinforcing ribs, the two ends of the reinforcing ribs are fixedly connected to the heat dissipation frame and the sliding seat respectively, and the reinforcing ribs are evenly distributed along the sliding direction of the sliding seat. The reinforcing ribs at the connection between the heat dissipation frame and the sliding seat significantly improve the structural rigidity and load-bearing capacity.

[0010] As a further improvement of this utility model: a guide slope is provided at the front edge of the limiting protrusion, the guide slope is facing the insertion direction of the positioning pin, and when the positioning pin is fully inserted into the pin groove, the limiting protrusion is fully inserted into the limiting groove under the elastic force of the spring. The guide slope set at the front edge of the limiting protrusion allows the positioning pin to smoothly push the limiting protrusion during insertion, achieving automatic locking without impact.

[0011] As a further improvement of this utility model: the inner wall of the grease tank is set as an inclined surface, the inclined surface gradually shrinks inward from the opening of the grease tank to the bottom of the tank, and the grease tank is filled with thermal conductive grease, which is in contact with the bottom of the GPU module. The thermal grease filling the slot forms a tight thermal contact with the bottom of the GPU module, reducing contact thermal resistance and improving heat dissipation efficiency.

[0012] Compared with the prior art, the beneficial effects of this utility model are: In this invention, the positioning pin and the automatic locking component work together. During the insertion of the positioning pin, the guide slope guides the pin, and the spring-driven folding frame mechanism achieves automatic locking between the limiting protrusion and the limiting groove. This ensures the accuracy and consistency of module installation and provides a firm mechanical lock, effectively preventing the module from loosening due to vibration or accidental external force.

[0013] The copper heat sink uses a thermal grease groove structure to ensure a sufficient and uniform thermal interface with the bottom of the GPU chip. Combined with the silicone ring pads on the edges, it provides a buffer seal and prevents the thermal material from overflowing. Heat is efficiently conducted to the water cooling pipes. The modular heat sink architecture forms a complete heat dissipation system, ensuring the efficient operation of the heat dissipation path. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall structure of the modular liquid-cooled backplate of the GPU cluster server described in this utility model; Figure 2 This is a schematic diagram of the structure of a single modular liquid-cooled backplane of a GPU cluster server according to the present invention; Figure 3 This is a schematic diagram of the positioning pin structure in the modular liquid-cooled backplate of a GPU cluster server according to the present invention; Figure 4 This is a schematic diagram of the locking component in the modular liquid-cooled backplane of a GPU cluster server according to the present invention; Figure 5 This is a schematic diagram of the heat dissipation rack in the modular liquid-cooled backplate of a GPU cluster server according to the present invention; Figure 6 This is a schematic diagram of the heat dissipation copper plate in the modular liquid cooling backplate of a GPU cluster server according to the present invention.

[0015] In the diagram: 1. Support wall; 2. Fixed guide rail; 3. Sliding seat; 4. Positioning pin; 41. Limiting groove; 5. Locking assembly; 6. Heat dissipation frame; 7. Heat dissipation assembly; 8. GPU module; 51. Locking compartment; 52. Pin groove; 53. Folding frame; 54. Limiting protrusion; 55. Spring; 72. Fixed bracket; 73. Water cooling heat pipe; 74. Connecting end; 75. Card plate; 76. Heat dissipation copper plate; 77. Silicone ring gasket; 78. Silicone grease groove. Detailed Implementation

[0016] The technical solutions of the present utility model will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present utility model, and not all embodiments. Based on the embodiments of the present utility model, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present utility model.

[0017] Reference Figures 1 to 6 In this embodiment of the present invention, a modular liquid-cooled backplate for a GPU cluster server includes: a support wall 1, a locking component 5, and a heat dissipation component 7. The support wall 1 serves as the basic support structure for the entire backplate, and a fixed guide rail 2 is fixedly connected to its side by welding or integral molding. The fixed guide rail 2 is preferably made of high-strength aluminum alloy profile. A sliding seat 3 is slidably connected above the fixed guide rail 2 by clearance fit. A positioning pin 4 is fixedly connected to the end of the sliding seat 3 by threaded fasteners. A limit groove 41 is opened on the side of the positioning pin 4.

[0018] The locking assembly 5 includes a locking compartment 51 fixedly connected to the side of the support wall 1. The inner side of the locking compartment 51 has a pin groove 52 that cooperates with the positioning pin 4. The inner side of the pin groove 52 has a placement groove. The inner side wall of the placement groove is hinged to a folding frame 53. The side of the folding frame 53 is hinged to a limiting protrusion 54 that matches the size of the limiting groove 41. A guide slope is provided at the front edge of the limiting protrusion 54. A spring 55 is fixedly connected between the two sets of folding plates of the folding frame 53. When the positioning pin 4 is fully inserted into the pin groove 52, the limiting protrusion 54 is fully engaged in the limiting groove 41 under the elastic force of the spring 55.

[0019] A heat dissipation frame 6 is fixedly connected to the side of the sliding seat 3. A reinforcing rib is provided at the connection between the heat dissipation frame 6 and the sliding seat 3. The two ends of the reinforcing rib are fixedly connected to the heat dissipation frame 6 and the sliding seat 3 respectively. The reinforcing rib is evenly distributed along the sliding direction of the sliding seat 3. A heat dissipation component 7 is provided above the heat dissipation frame 6. The heat dissipation component 7 includes a fixed bracket 72 fixedly connected above the heat dissipation frame 6. A water-cooled heat dissipation pipe 73 is fixedly clamped in the middle of the fixed bracket 72. A docking end 74 is fixedly provided above the fixed bracket 72. A clamping plate 75 is movably clamped above the docking end 74. A heat dissipation copper plate 76 is fixedly connected above the clamping plate 75. A silicone ring gasket 77 is fixedly connected to the upper edge of the heat dissipation copper plate 76. A silicone grease groove 78 is opened on the back of the upper part of the heat dissipation copper plate 76. The inner side wall of the silicone grease groove 78 is set as an inclined surface. The inclined surface gradually shrinks inward from the groove opening to the bottom of the groove. The silicone grease groove 78 is filled with thermally conductive silicone.

[0020] The working principle of this utility model is as follows: When a GPU module 8 needs to be installed, the operator pushes the entire heat sink 6 unit along the fixed guide rail 2 on the side of the support wall 1 through the sliding seat 3 at its bottom. As the sliding seat 3 moves, the positioning pin 4 fixed at its end is also inserted into the pin groove 52 on the locking compartment 51. During the insertion process, the side of the positioning pin 4 will press the guide slope at the front end of the limiting protrusion 54 in the locking assembly 5. This force causes the hinged folding frame 53 to overcome the elasticity of its internal spring 55 and fold, so that the limiting protrusion 54 is temporarily retracted into the placement groove, making way for the smooth insertion of the positioning pin 4. When the positioning pin 4 is fully inserted, the limiting groove 41 on its side just moves to the position aligned with the limiting protrusion 54. The folding frame 53 quickly unfolds under the action of the spring 55's restoring elasticity, driving the limiting protrusion 54 to accurately engage in the limiting groove 41, thereby automatically completing the mechanical locking of the sliding seat 3 and the entire heat sink 6, preventing them from accidentally sliding out during operation and ensuring the stability of the connection. After the heat dissipation rack 6 is reliably locked, its heat dissipation function begins to work. The large amount of heat generated when the GPU module 8 is working is first conducted to the heat dissipation copper plate 76, which is in direct and close contact with its bottom. The thermal grease groove 78 on the top of the heat dissipation copper plate 76 is filled with thermally conductive thermal grease, which ensures efficient heat transfer from the GPU module 8 to the heat dissipation copper plate 76. The silicone ring pad 77 is attached to the bottom edge of the GPU module 8, which plays a role in sealing and buffering. After the huge amount of heat is quickly conducted into the heat dissipation copper plate 76, it is then carried away by the coolant flowing inside the water-cooling heat pipe 73 fixed and snapped below it, which continuously transfers the heat to the external heat dissipation system, thereby achieving efficient cooling of the GPU module 8.

[0021] The above description is only a preferred embodiment of the present utility model, but the protection scope of the present utility model is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present utility model, based on the technical solution and inventive concept of the present utility model, should be included within the protection scope of the present utility model.

Claims

1. A modular liquid-cooled backplane for a GPU cluster server, characterized in that, include: Support wall (1), locking component (5) and heat dissipation component (7), a fixed guide rail (2) is fixedly connected to the side of the support wall (1), a sliding seat (3) is slidably connected above the fixed guide rail (2), and a positioning pin (4) is fixedly connected to the end of the sliding seat (3). The locking component (5) is used to lock the positioning pin (4) and restrict the sliding of the sliding seat (3); The heat dissipation component (7) is used to improve the heat dissipation efficiency of the server GPU.

2. The modular liquid-cooled backplane for a GPU cluster server according to claim 1, characterized in that, The sliding seat (3) is fixedly connected to a heat dissipation frame (6) on its side, and a limit groove (41) is opened on the side of the positioning pin (4). A heat dissipation component (7) is provided above the heat dissipation frame (6).

3. The modular liquid-cooled backplane for a GPU cluster server according to claim 2, characterized in that, The locking assembly (5) includes a locking compartment (51) fixedly connected to the side of the support wall (1). A pin groove (52) is provided inside the locking compartment (51). A placement groove is provided inside the pin groove (52). A folding frame (53) is hinged to the side wall of the placement groove. A limiting protrusion (54) matching the size of the limiting groove (41) is hinged to the side of the folding frame (53). A spring (55) is fixedly connected between the two sets of folding plates of the folding frame (53).

4. The modular liquid-cooled backplane of a GPU cluster server according to claim 2, characterized in that, The heat dissipation assembly (7) includes a fixed bracket (72) fixedly connected above the heat dissipation frame (6). A water-cooled heat pipe (73) is fixedly connected in the middle of the fixed bracket (72). A docking end (74) is fixedly provided above the fixed bracket (72). A card plate (75) is movably connected above the docking end (74). A heat dissipation copper plate (76) is fixedly connected above the card plate (75). A silicone ring pad (77) is fixedly connected at the upper edge of the heat dissipation copper plate (76). A silicone grease groove (78) is opened on the back of the heat dissipation copper plate (76). A GPU module (8) is placed in the middle area above the heat dissipation copper plate (76). The silicone ring pad (77) is attached to the bottom edge area of ​​the GPU module (8).

5. The modular liquid-cooled backplane of a GPU cluster server according to claim 2, characterized in that, The connection between the heat dissipation frame (6) and the sliding seat (3) is provided with reinforcing ribs. The two ends of the reinforcing ribs are fixedly connected to the heat dissipation frame (6) and the sliding seat (3) respectively, and the reinforcing ribs are evenly distributed along the sliding direction of the sliding seat (3).

6. The modular liquid-cooled backplane of a GPU cluster server according to claim 3, characterized in that, The front edge of the limiting protrusion (54) is provided with a guide slope, which faces the insertion direction of the positioning pin (4). When the positioning pin (4) is fully inserted into the pin groove (52), the limiting protrusion (54) is fully inserted into the limiting groove (41) under the elastic force of the spring (55).

7. The modular liquid-cooled backplane of a GPU cluster server according to claim 4, characterized in that, The inner wall of the grease tank (78) is set as an inclined surface. The inclined surface gradually shrinks inward from the opening of the grease tank (78) to the bottom of the tank. The grease tank (78) is filled with thermal grease, and the thermal grease is in contact with the bottom of the GPU module (8).

Images