AI computing server convenient for wind-passing heat dissipation

Abstract

The invention relates to the technical field of AI computing, in particular to an AI computing server convenient for wind-passing heat dissipation. The server comprises a case, and an installation area used for installing an accelerator card and a heat dissipation area used for installing a heat dissipation device are sequentially arranged in the case in the length direction of the case. A main air duct is formed between the heat dissipation area and the mounting area; the acceleration cards are arranged at intervals in the width direction of the case, and a heat dissipation air channel is formed between every two adjacent acceleration cards; the top of the case is provided with a power distribution board, the power distribution board is provided with a plurality of power transmission interfaces used for being connected with an accelerator card, the two ends of the power distribution board in the length direction are each provided with a lateral connector, and the lateral connectors are used for electrically connecting the power distribution board and a power supply board installed at the bottom of the case. The technical problem that in the prior art, a PCIE switching flat cable used for connecting an AI computing board card and a mainboard occupies the space of a heat dissipation channel, airflow resistance is increased, and consequently heat dissipation of an AI server is poor is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of AI computing technology, and in particular to an AI computing server that facilitates heat dissipation through airflow. Background Technology

[0002] AI computing servers are dedicated servers that provide high-efficiency computing power support for artificial intelligence (AI) model training, inference, and data processing. At their core, they achieve efficient support for AI model training and inference through a heterogeneous computing architecture of CPU and dedicated accelerator cards, coupled with high-bandwidth interconnects, large-capacity high-speed storage, efficient heat dissipation systems, and AI framework adaptation software. They are the core hardware carriers for supporting AI applications such as large language models, computer vision, and speech recognition.

[0003] Currently, patent document CN110134205B discloses an AI computing server. This server includes a chassis, which is divided into a first area, a second area, and a third area by a bracket. The first area houses a server motherboard and an image processing board, with the image processing board connected to the graphics card interface of the server motherboard. The third area houses multiple stacked AI computing boards and a hard disk array. The AI ​​computing boards are electrically connected to the server motherboard via PCIe adapter cables, and the side of each AI computing board is fixed to the side wall of the server. Further... The system includes an adapter board and a computing power board. The adapter board has an M.2 socket, a bridging chip, and a PCIe interface. The computing power board has an M.2 connector and an AI chip. The two are detachably connected to the M.2 socket via the M.2 connector. The bridging chip is responsible for data transmission and distribution. The hard disk array is electrically connected to the server motherboard via a data cable. A first heat dissipation device is set in the second area. This heat dissipation device includes a heat sink bracket that separates the first area and the third area. The heat sink bracket has a heat dissipation channel connecting the two areas. A cooling fan is fixed in the heat dissipation channel to realize air circulation and heat dissipation inside the chassis.

[0004] The aforementioned existing technologies can achieve basic heat dissipation functions by setting up dedicated heat dissipation channels and cooling fans, but they still have significant drawbacks in practical applications: because the PCIe adapter cable used to connect the AI ​​computing board to the motherboard is located directly within the heat dissipation airflow path (i.e., the heat dissipation channel), its physical structure occupies part of the flow channel space, resulting in a reduction in the effective flow cross-sectional area of ​​the heat dissipation channel. This not only directly increases airflow resistance and disrupts the uniform flow of cooling air, but also seriously affects the air delivery efficiency of the cooling fan and the heat exchange effect of the heat sink surface, thus leading to poor heat dissipation performance of the entire AI computing server. When the high-power AI computing board in the server operates at full load for a long time, if the accumulated heat cannot be dissipated in a timely and effective manner, it will easily cause problems such as chip overheating and frequency reduction, decreased system stability, and even hardware damage. Summary of the Invention

[0005] This invention provides an AI computing server that facilitates airflow heat dissipation, thereby solving the technical problem in the prior art where the PCIe adapter cable used to connect the AI ​​computing board and the motherboard occupies the space of the heat dissipation channel, increases airflow resistance, and leads to poor heat dissipation in the AI ​​server.

[0006] To address the aforementioned issues, the AI ​​computing server provided by this invention, which facilitates heat dissipation through airflow, employs the following technical solution: An AI computing server with convenient airflow and heat dissipation includes a chassis. Inside the chassis, along its length, are sequentially arranged an installation area for mounting accelerator cards and a heat dissipation area for mounting heat dissipation devices. A main airflow duct is formed between the heat dissipation area and the installation area. Multiple accelerator cards are arranged at intervals along the width of the chassis, with heat dissipation ducts forming between adjacent accelerator cards. A power distribution board is located on the top of the chassis, and the power distribution board has multiple power input interfaces for connecting to the accelerator cards. The power input interfaces and the power cables connecting them to the accelerator cards are spatially isolated from the main airflow duct and the heat dissipation duct. A lateral connector is located at each end of the power distribution board along its length, for electrically connecting the power distribution board to a power supply board mounted at the bottom of the chassis.

[0007] The beneficial effects of the AI ​​computing server with convenient airflow heat dissipation provided by this invention are: In existing high-density AI computing servers, multiple accelerator cards (such as GPUs) not only communicate with the motherboard via gold fingers, but also receive power from the power supply through independent power cables (usually multiple thick, stiff copper cables). These power cables typically extend from the motherboard or power supply backplate, meandering along the internal space of the chassis, and finally connect to the top or side power supply interface of each accelerator card.

[0008] Because the accelerator cards are densely packed, the cables need to pass through the heat dissipation airflow, which will cause the following negative effects: 1. Encroachment on the effective flow cross-section: The physical cable blocks part of the cross-sectional area of ​​the heat dissipation airflow, reducing the volume of air that can pass through per unit time; 2. Increased airflow resistance and turbulence: The rough surface and irregular shape of the cable will generate significant frictional resistance and eddies when the airflow passes through, disturbing the laminar flow state and reducing cooling efficiency; 3. Interference with uniform heat dissipation: The obstruction of the cable may lead to uneven airflow distribution, local heat accumulation, and the formation of hot spots.

[0009] Furthermore, as the power consumption of accelerator cards such as GPUs continues to rise (e.g., a single card's power consumption exceeds 1000W), the problem of significantly increased cable diameter and stiffness in traditional power supply solutions will be further exacerbated. Specifically, to meet the demands of high current transmission, power cables need to use thicker conductors (e.g., upgrading from 12AWG to 8AWG), resulting in increased cable bending radius and decreased flexibility. This, in turn, leads to a chain reaction of problems such as difficulties in internal wiring within the chassis, reduced heat dissipation space, and concentrated mechanical stress.

[0010] Therefore, this application sets up a power distribution board on the top of the chassis, which serves as a centralized power supply node for each accelerator card. The power distribution board obtains power directly from the power supply board at the bottom of the chassis through the side connectors at both ends of its length.

[0011] On the one hand, the accelerator card directly connects to the power distribution board located at the top of the chassis, and the power connection point is concentrated at the top of the chassis, which completely avoids the accelerator card's power cable from entering the core area of ​​the heat dissipation airflow. The airflow space that was originally occupied by the cable is completely released, eliminating the direct obstruction to airflow.

[0012] On the other hand, multiple accelerator cards are evenly spaced along the width of the chassis, forming regular and straight independent heat dissipation channels between adjacent accelerator cards. Furthermore, the power cables of the accelerator cards do not occupy the heat dissipation channels. The unobstructed and regular heat dissipation channels allow the airflow generated by the heat dissipation device to pass smoothly through the heat dissipation channels in a straight line without having to bypass the cables, and directly act on the surface of the accelerator cards. This significantly reduces turbulence phenomena such as airflow diversion and vortices, reduces pressure loss when airflow passes through, and increases the ventilation volume per unit time.

[0013] In addition, the two side connectors are located at both ends of the power distribution board along its length, rather than in the middle area corresponding to the main air duct. Their connection cables to the power supply board at the bottom of the chassis can be arranged along the inner edge of the chassis. This edge routing design ensures that the effective flow cross-sectional area of ​​the main air duct is maximized, avoiding the occupation of the main air duct space by power supply cables or connectors, further reducing airflow resistance, and ensuring that the cooling effect of the heat dissipation device is fully utilized.

[0014] Through the above-mentioned design, the present invention effectively solves the technical problem in the prior art where the PCIe adapter cable used to connect the AI ​​computing board and the motherboard occupies the space of the heat dissipation channel, increases airflow resistance, and leads to poor heat dissipation of the AI ​​server.

[0015] Furthermore, the bottom of the chassis is provided with a bottom slot for installing the accelerator card, and a baffle is installed on the side wall of the chassis. The baffle is located on the side of the accelerator card away from the heat dissipation device and is used to fix the accelerator card.

[0016] Beneficial effects: On the one hand, the bottom slot provides bottom support and positioning for the accelerator card, while the baffle provides lateral fixation from the side of the accelerator card away from the heat dissipation device. This forms a double-reinforced structure of bottom slot limiting and side baffle fixing, effectively resisting impacts on the server during transportation and installation, as well as vibration interference generated during long-term operation in the data center. This prevents the accelerator card from loosening or shifting, reducing the risk of faults such as poor power supply contact and abnormal signal transmission caused by mechanical stress, and ensuring long-term stable operation of the server. On the other hand, the baffle is installed on the side of the accelerator card away from the heat dissipation device, rather than inside the heat dissipation duct. It does not extend into the heat dissipation channel between adjacent accelerator cards, avoiding obstruction or blockage of airflow and ensuring that the unobstructed heat dissipation duct structure formed by the spacing between adjacent accelerator cards remains intact. At the same time, the lateral fixing method of the baffle eliminates the need for additional fixing components inside the duct, does not increase airflow resistance or generate turbulence, and ensures that the airflow generated by the heat dissipation device can smoothly pass through the duct and act on the surface of the accelerator card, maintaining the original heat dissipation advantages.

[0017] Furthermore, the accelerator card has a connector at one end facing the power transmission interface, and the connector is detachably connected to the power distribution board.

[0018] Beneficial effects: On the one hand, the accelerator card connects to the motherboard via a bottom slot, and its top is detachably connected to the power distributor via a connector. This creates a triple support structure: bottom slot positioning, side wall support, and top power distributor fixation. This effectively distributes the stress points of the accelerator card, avoiding the stress concentration problem caused by the cantilever design in the middle of the card when only the bottom slot is used for fixation. This significantly enhances the overall structural rigidity of the accelerator card and greatly improves its bending resistance, effectively preventing deformation and bottom slot damage caused by external forces during transportation, installation, or long-term use. On the other hand, in data center environments, the operation of server cooling devices and the start-up and shutdown of equipment generate continuous vibrations. The detachable connection between the connector and the power distributor firmly limits the top of the accelerator card, preventing it from shaking or shifting due to vibration when the top is in a free state. This reduces the impact of vibration on the electrical connection between the accelerator card and the motherboard and power distributor, reducing the risk of poor contact and signal transmission interruption, and ensuring stable operation of the accelerator card even under long-term vibration. In addition, the detachable connection design of the connector and the power distribution board ensures the server's maintainability while securing the top of the accelerator card. It allows for quick separation of the connector and the power distribution board without damaging the power supply link or the power distribution board or accelerator card, facilitating the disassembly, replacement or repair of the accelerator card. This balances structural stability and ease of maintenance, improving the user experience throughout the entire lifecycle of the equipment.

[0019] Furthermore, the connector includes a vertical connecting part and a horizontal connecting part. The vertical connecting part is connected to the accelerator card, and the horizontal connecting part is disposed on the top of the vertical connecting part and is detachably connected to the power distribution board.

[0020] Beneficial effects: The connector adopts an L-shaped integrated structure design including a vertical connection part and a horizontal connection part. The vertical connection part is firmly connected to the accelerator card, and the horizontal connection part extends laterally to the power distribution board and can be detached. This forms a rigid support link of accelerator card - vertical connection part - horizontal connection part - power distribution board, which enables the top of the accelerator card to obtain a dual force of lateral restraint and longitudinal support. It can evenly distribute the bending stress of the accelerator card, effectively suppress the mid-section deflection caused by the cantilever installation of the accelerator card, significantly enhance the overall structural rigidity, greatly improve the bending resistance and deformation resistance of the accelerator card, and avoid problems such as damage to the accelerator card circuit board caused by long-term use or external impact.

[0021] In addition, the segmented design of the vertical and horizontal connecting parts allows for precise size adjustment based on the internal space of the chassis, the height of the accelerator card, and the installation position of the power distribution board. This ensures that the vertical connecting part fits snugly against the side of the accelerator card without occupying the heat dissipation airflow space between adjacent accelerator cards. The horizontal connecting part connects laterally along the bottom / top surface of the power distribution board, resulting in a compact size that does not extend into the core area of ​​the heat dissipation airflow, thus ensuring heat dissipation performance.

[0022] Furthermore, the bottom of the accelerator card has gold fingers that are inserted into the bottom slot.

[0023] Beneficial effects: The gold fingers on the bottom of the accelerator card adopt a pluggable adapter design with the pin arrangement of the gold fingers precisely matching the contacts of the bottom slot. This provides a precise installation positioning reference for the accelerator card, ensuring that the accelerator card is installed smoothly in the preset direction and avoiding interface misalignment caused by offset. It also enables stable conduction of signal transmission and basic power supply between the accelerator card and the motherboard through the tight fit between the gold fingers and the slot contacts.

[0024] Furthermore, the power distribution board includes a conductive layer and a support layer arranged sequentially from top to bottom. The power transmission interface is disposed on the conductive layer, and the support layer is provided with a through hole for the lateral connector to pass through. The lateral connector passes through the through hole and connects to the conductive layer.

[0025] Beneficial effects: The power distribution board adopts a layered structure including a conductive layer and a support layer, achieving a precise division of labor between power supply function and structural support: the conductive layer specifically carries the power transmission interface and power distribution circuit, ensuring the stability of power supply to the accelerator card; the support layer, as the basic load-bearing structure, provides rigid support for the conductive layer, while integrating lateral connectors to avoid mutual interference between the power supply interface and the support structure. This layered design ensures the power transmission efficiency of the conductive layer, while strengthening the overall structural strength of the power distribution board through the support layer, preventing deformation of the power distribution board due to high current heating or mechanical vibration, and improving long-term operational reliability.

[0026] Furthermore, connecting plates are provided at both ends of the support layer along its length to connect the support layer and the chassis.

[0027] Beneficial effects: The connecting plates at both ends of the support layer provide symmetrical and stable mounting points for the power distributor board, enabling it to be rigidly fixed to the chassis via the connecting plates. Compared to single-point or side fixing, this symmetrical fixing method can more evenly distribute the weight of the power distributor board itself and the pressure transmitted by the accelerator card, preventing the power distributor board from warping or deforming due to long-term load or vibration. This ensures the docking accuracy between the power distributor board and the accelerator card and side connectors, while also strengthening the structural reliability of the power distributor board as the top support of the accelerator card, indirectly improving the accelerator card's vibration and bending resistance.

[0028] Furthermore, the heat dissipation device includes a plurality of cooling fans spaced apart along the width direction of the chassis, and the plurality of cooling fans form a cooling fan wall.

[0029] Beneficial effects: The cooling fans spaced along the width of the chassis can completely cover all the parallel-arranged accelerator cards with their exhaust area. This fan wall layout creates uniform surface airflow, ensuring that each accelerator card receives a stable and sufficient airflow, avoiding overheating hotspots caused by insufficient local airflow. Furthermore, the fan wall, composed of multiple independent cooling fans, essentially constitutes an N+1 redundancy system. When a single cooling fan fails, the remaining fans can still maintain some cooling capacity, preventing the server from immediately crashing due to localized overheating and providing a buffer time for maintenance response.

[0030] Furthermore, the power transmission interface is connected to at least one power supply female connector, and the accelerator card is equipped with a power supply male connector, with the power supply female connector and the power supply male connector on the same horizontal plane.

[0031] Beneficial effects: The power supply female connector and male connector are positioned on the same horizontal plane, achieving precise horizontal alignment and avoiding misalignment and poor contact caused by height discrepancies. This coaxial, same-plane design allows the male connector to quickly and accurately insert into the power supply female connector when the accelerator card connects to the power distribution board via connectors, eliminating the need for repeated angle or height calibrations. This improves assembly efficiency and ensures full contact between the two, reducing contact resistance. Simultaneously, horizontal alignment reduces the risk of wear and loosening of contacts during long-term use or under vibration, ensuring the stability of high-power power transmission and preventing accelerator card performance fluctuations and downtime due to power interruptions or fluctuations. Furthermore, the compact layout of the power supply female connector and male connector on the same horizontal plane ensures that the overall height of the connected interface is flush with the mounting surface of the power distribution board and the top of the accelerator card, preventing any protrusions into the airflow duct.

[0032] Furthermore, each of the power transmission interfaces is connected to two power supply female connectors via the power supply cable.

[0033] Beneficial effects: Each power input interface connects to two power supply female connectors via power cables, creating a one-in-two-out power distribution mode, allowing the power from a single power input interface to simultaneously power two accelerator cards. This design eliminates the need for a separate power input interface for each accelerator card, doubling the number of power output paths without increasing the total number of power input interfaces on the power distribution board, significantly improving the utilization rate of the power distribution board's power resources; at the same time, it reduces the number of power input interfaces on the power distribution board, making the power distribution board structure more compact, reserving more installation space for other components, and optimizing the internal layout of the chassis. Attached Figure Description

[0034] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 Schematic diagram of the structure of the AI ​​computing server for easy heat dissipation provided by the present invention Figure 1 ; Figure 2 Schematic diagram of the structure of the AI ​​computing server for easy heat dissipation provided by the present invention Figure 2 ; Figure 3 for Figure 2 A magnified view of a portion of point A in the middle; Figure 4 for Figure 2 A magnified view of a portion of point B in the middle; Figure 5 A schematic diagram of the power distribution board in the AI ​​computing server with easy airflow heat dissipation provided by the present invention; Figure 6 for Figure 5 Exploded view.

[0035] Explanation of reference numerals in the attached figures: 1. Chassis; 2. Accelerator Card; 3. Power Distribution Board; 31. Conductive Layer; 32. Support Layer; 321. Connecting Plate; 4. Power Transmission Interface; 5. Power Supply Board; 6. Baffle; 7. Connector; 71. Vertical Connector; 72. Horizontal Connector; 8. Cooling Fan; 9. Power Supply Female Connector; 10. Power Supply Male Connector; 11. Power Supply Cable; 12. Connecting Terminal. Detailed Implementation

[0036] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0037] An embodiment of the AI ​​computing server with convenient airflow heat dissipation provided by the present invention: like Figures 1 to 6As shown, the AI ​​server that facilitates airflow and heat dissipation includes a chassis 1. Inside the chassis 1, along its length, there are an installation area for installing the accelerator card 2 and a heat dissipation area for installing the heat dissipation device. A power distribution board 3 is provided on the top of the chassis 1. The power distribution board 3 is used to connect the accelerator card 2 and the power supply board 5 installed at the bottom of the chassis 1.

[0038] The heat dissipation area and the installation area form a main air duct; there are multiple accelerator cards 2, which are spaced apart along the width of the chassis 1, and a heat dissipation air duct is formed between two adjacent accelerator cards 2, and a heat dissipation air duct is also formed between the accelerator card 2 and the side wall of the chassis 1.

[0039] Regarding the power distribution board 3 and its connection with the accelerator card 2 and the power supply board 5: The power distribution board 3 is equipped with multiple power transmission interfaces 4 for connecting to the accelerator card 2. The power transmission interfaces 4 and the power supply cables 11 used to connect the power transmission interfaces 4 to the accelerator card 2 are spatially isolated from the main air duct and the heat dissipation air duct. A lateral connector is provided at each end of the length direction of the power distribution board 3, which is used for electrical connection between the power distribution board 3 and the power supply board 5. Each power transmission interface 4 is connected to at least one power supply female connector 9, and the accelerator card 2 is equipped with a power supply male connector 10. The power supply female connector 9 and the power supply male connector 10 are on the same horizontal plane.

[0040] like Figure 5 and Figure 6 As shown, in this embodiment, the power distribution board 3 includes a conductive layer 31 and a support layer 32 arranged sequentially from top to bottom. The power transmission interface 4 is disposed on the conductive layer 31. The support layer 32 is provided with a through hole for a lateral connector to pass through, and the lateral connector passes through the through hole to connect to the conductive layer 31. Connecting plates 321 are provided at both ends of the support layer 32 along its length, and the connecting plates 321 are used to connect the support layer 32 and the chassis 1.

[0041] like Figure 5 and Figure 6 As shown, in this embodiment, the lateral connector includes a power supply cable 11 and two mating terminals 12, which are respectively mounted on the power supply board 5 and the support layer 32.

[0042] like Figure 2 and Figure 3 As shown, in this embodiment, the bottom of the chassis 1 is provided with a bottom slot for installing the accelerator card 2, and a baffle 6 is installed on the side wall of the chassis 1. The baffle 6 is located on the side of the accelerator card 2 away from the heat dissipation device, and the baffle 6 is used to fix the accelerator card 2 laterally. In this embodiment, the bottom of the accelerator card 2 has gold fingers, which are inserted into the bottom slot.

[0043] like Figure 1 As shown, in this embodiment, the end of the accelerator card 2 facing the power transmission interface 4 is provided with a connector 7, which is detachably connected to the power distribution board 3. Figure 2and Figure 4 As shown, the connector 7 includes a vertical connecting part 71 and a horizontal connecting part 72. The vertical connecting part 71 is connected to the accelerator card 2, and the horizontal connecting part 72 is disposed on the top of the vertical connecting part 71 and is detachably connected to the power distribution board 3.

[0044] In this embodiment, the heat dissipation device includes a plurality of cooling fans 8 spaced apart along the width direction of the chassis 1, and the plurality of cooling fans 8 form a cooling fan wall.

[0045] like Figure 1 , Figure 5 and Figure 6 As shown, in this embodiment, each power transmission interface 4 is connected to two power supply female connectors 9 via a power supply cable 11.

[0046] It's important to note that an Accelerator Card (ACC) is a hardware device specifically designed to enhance the performance of specific computing tasks. It's typically integrated into a computer system as a plug-in card (such as a PCIe slot) and significantly accelerates specific types of data processing through parallel computing, dedicated architecture, or optimized algorithms. An ACC is a CPU-independent hardware module that achieves higher processing efficiency than a general-purpose CPU by offloading computational loads or optimizing specific algorithms. Its core objective is to reduce the execution time of specific tasks, with particularly significant effects when handling large-scale, highly parallel, or computationally intensive tasks.

[0047] Accelerator cards 2 include GPU accelerator cards (graphics processor accelerator cards), FPGA accelerator cards (field programmable gate array accelerator cards), ASIC accelerator cards (application-specific integrated circuit accelerator cards), and DPU accelerator cards (data processor accelerator cards).

[0048] The working principle of the AI ​​computing server with easy airflow heat dissipation provided by this invention is as follows: The electrical energy output from the power supply board 5 is connected to the support layer 32 through two lateral connectors, and then transmitted to the power transmission interface 4 on the conductive layer 31 through conductive posts. The power transmission interface 4 is connected to two power supply female connectors 9 through power supply cables 11. The power supply female connectors 9 and the power supply male connectors 10 are horizontally connected to achieve the power supply to the accelerator card 2. When each cooling fan 8 is running, it generates directional airflow. The directional airflow enters the installation area through the main air duct between the heat dissipation area and the installation area. Afterward, the airflow passes through each heat dissipation air duct, carrying away the heat generated by the operation of the accelerator card 2.

[0049] Based on the above description in this specification, those skilled in the art will also understand that the following terms, such as "upper," "lower," "front," "rear," "left," "right," "width," "horizontal," "top," "bottom," "inner," and "outer," which indicate orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings of this specification. They are only for the purpose of facilitating the explanation of the present invention and simplifying the description, and do not explicitly or implicitly suggest that the device or element involved must have the specific orientation, or be constructed and operated in a specific orientation. Therefore, the above-mentioned orientation or positional relationship terms should not be understood or interpreted as limitations on the present invention.

[0050] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

Claims

1. An AI computing server with convenient airflow and heat dissipation, comprising a chassis, wherein an installation area for mounting accelerator cards and a heat dissipation area for mounting heat dissipation devices are sequentially arranged along the length of the chassis; a main airflow duct is formed between the heat dissipation area and the installation area; the number of accelerator cards is multiple, and the multiple accelerator cards are spaced apart along the width of the chassis, with a heat dissipation airflow duct formed between adjacent accelerator cards; characterized in that, The top of the chassis is equipped with a power distribution board, which has multiple power transmission interfaces for connecting to the accelerator card. The power transmission interfaces and the power supply cables for connecting the power transmission interfaces to the accelerator card are spatially isolated from the main air duct and the heat dissipation air duct. Each end of the power distribution board is equipped with a lateral connector, which is used to electrically connect the power distribution board to the power supply board installed at the bottom of the chassis.

2. The AI ​​computing server with convenient airflow heat dissipation according to claim 1, characterized in that, The bottom of the chassis is provided with a bottom slot for installing the accelerator card, and a baffle is installed on the side wall of the chassis. The baffle is located on the side of the accelerator card away from the heat dissipation device and is used to fix the accelerator card.

3. The AI ​​computing server with convenient airflow heat dissipation according to claim 2, characterized in that, The accelerator card has a connector at one end facing the power transmission interface, and the connector is detachably connected to the power distribution board.

4. The AI ​​computing server with convenient heat dissipation via airflow according to claim 3, characterized in that, The connector includes a vertical connecting part and a horizontal connecting part. The vertical connecting part is connected to the accelerator card, and the horizontal connecting part is located on top of the vertical connecting part and is detachably connected to the power distribution board.

5. The AI ​​computing server with convenient heat dissipation according to any one of claims 2 to 4, characterized in that, The accelerator card has gold fingers at the bottom, which are inserted into the bottom slot.

6. The AI ​​computing server with convenient heat dissipation according to any one of claims 1 to 4, characterized in that, The power distribution board includes a conductive layer and a support layer arranged sequentially from top to bottom. The power transmission interface is disposed on the conductive layer, and the support layer is provided with a through hole for the lateral connector to pass through. The lateral connector passes through the through hole and connects to the conductive layer.

7. The AI ​​computing server with convenient heat dissipation via airflow according to claim 6, characterized in that, Connecting plates are provided at both ends of the support layer along its length to connect the support layer and the chassis.

8. The AI ​​computing server with convenient heat dissipation according to any one of claims 1 to 4, characterized in that, The heat dissipation device includes multiple cooling fans spaced apart along the width of the chassis, forming a cooling fan wall.

9. The AI ​​computing server with convenient heat dissipation according to any one of claims 1 to 4, characterized in that, The power transmission interface is connected to at least one power supply female connector, and the accelerator card is equipped with a power supply male connector. The power supply female connector and the power supply male connector are on the same horizontal plane.

10. The AI ​​computing server with convenient airflow heat dissipation according to claim 9, characterized in that, Each of the power transmission interfaces is connected to two power supply female connectors via the power supply cable.

Images