Mainboard assembly and liquid cooling server

By designing an OCP interface in the liquid-cooled server to connect the CPU, CPLD and auxiliary power supply, the hot plug and unplug of the GPU chip is solved, and the problem of GPU card failure in the existing technology needs to be shut down and replaced, improving operation and maintenance efficiency and reducing costs.

CN221977373UActive Publication Date: 2024-11-08SHENZHEN QIANHAI EVOC ASIA-PACIFIC ELECTRONIC EQUIP TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202420191557.7
Authority / Receiving Office
CN · China
Patent Type
Utility models(China)
Current Assignee / Owner
Filing Date
2024-01-25
Publication Date
2024-11-08
Estimated Expiration
2034-01-25

Smart Images

  • Figure CN221977373U_ABST
    Figure CN221977373U_ABST
Patent Text Reader

Abstract

The utility model relates to the technical field of heat dissipation, and discloses a mainboard assembly and a liquid cooling server, the mainboard assembly comprises a server mainboard, the server mainboard comprises a CPU, a CPLD, an auxiliary power supply and a first OCP connector, the CPU, the CPLD and the auxiliary power supply are electrically connected with the first OCP connector, and the CPLD is electrically connected with the auxiliary power supply. The GPU card comprises a GPU chip and a second OCP connector which are electrically connected; the server mainboard and the GPU card are connected in an inserted mode through the first OCP connector and the second OCP connector so that the GPU chip can be electrically connected with the CPU to receive PCIE signals of the CPU. And the CPLD is electrically connected with the GPU chip so as to respond to the hot plug instruction to send an interrupt PCIE signal to the GPU chip and pull down a power supply enable signal input to the auxiliary power supply after a preset number of clock cycles. And the auxiliary power supply is electrically connected with the GPU chip and is used for stopping supplying power to the GPU chip when the power supply enable signal is pulled down. Through the method, the influence on the working efficiency of the data center when the GPU card is operated and maintained is avoided, so that the operation and maintenance cost of the data center is saved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of server technology, and specifically to a motherboard assembly and a liquid-cooled server. Background Art

[0002] In order to meet the demand for high computing power, the GPU (Graphics Processing Unit) card installed in the existing liquid-cooled server is usually connected to the server motherboard using the PCIE (Peripheral Component Interconnect Express, a high-speed serial computer expansion bus standard) interface. When a GPU card device fails, the data center operation and maintenance personnel must first shut down the server node and power it off before replacing the faulty GPU card. The process of replacing the GPU card includes the power-on and power-off time of the server, which is time-consuming, reduces the work efficiency of the data center, and increases the operation and maintenance cost of the data center. Utility Model Content

[0003] In view of the above problems, the embodiments of the present application provide a motherboard assembly and a liquid cooling server, which avoids affecting the work efficiency of the data center when operating and maintaining the GPU card, thereby saving the operation and maintenance costs of the data center.

[0004] According to one aspect of an embodiment of the present application, a motherboard assembly is provided, which includes: a server motherboard, including a CPU, a CPLD, an auxiliary power supply and a first OCP connector, wherein the CPU, the CPLD and the auxiliary power supply are all electrically connected to the first OCP connector, and the CPLD is electrically connected to the auxiliary power supply; and a GPU card, including an electrically connected GPU chip and a second OCP connector; wherein the server motherboard and the GPU card are plugged in through the first OCP connector and the second OCP connector, so that: the GPU chip is electrically connected to the CPU to receive a PCIE signal of the CPU; and the CPLD is electrically connected to the GPU chip to send an interrupt PCIE signal to the GPU chip in response to a hot plug instruction, so that the GPU chip caches data, and pulls down a power enable signal input to the auxiliary power supply after a preset number of clock cycles; and the auxiliary power supply is electrically connected to the GPU chip, and is used to stop supplying power to the GPU chip when the power enable signal is pulled low, so that the GPU card can be hot plugged.

[0005] In an optional manner, the CPLD includes a first pin, and the first pin is electrically connected to the GPU chip through the first OCP connector and the second OCP connector; the GPU chip is used to input a high level to the first pin when the first OCP connector and the second OCP connector are plugged in, so that the CPLD pulls up a power enable signal input to the auxiliary power supply, so that the auxiliary power supply supplies power to the GPU chip.

[0006] In an optional manner, the auxiliary power supply is electrically connected to the CPU and is configured to be enabled after receiving an enable signal sent by the CPU, so that the GPU chip can be powered when the CPLD pulls up the power enable signal.

[0007] In an optional manner, the CPLD is electrically connected to the CPU, and after the second OCP connector is unplugged from the first OCP connector, the first pin is at a low level, and the CPLD is used to send a signal to the CPU indicating that the GPU card is not in place.

[0008] In an optional manner, the GPU chip includes a main chip and a memory unit, both of which are electrically connected to the second OCP connector; the auxiliary power supply includes a first auxiliary power supply and a second auxiliary power supply, both of which are electrically connected to the first OCP connector; the CPLD includes a second pin and a third pin; the first auxiliary power supply and the second auxiliary power supply are electrically connected to the second pin and the third pin respectively, and the CPLD inputs the power enable signal to the first auxiliary power supply and the second auxiliary power supply respectively through the second pin and the third pin; the first auxiliary power supply is electrically connected to the main chip through the first OCP connector and the second OCP connector, and is used to supply power to the main chip when the CPLD pulls up the power enable signal, and stops supplying power to the main chip when the CPLD pulls down the power enable signal; the second auxiliary power supply is electrically connected to the memory unit through the first OCP connector and the second OCP connector, and is used to supply power to the memory unit when the CPLD pulls up the power enable signal, and stops supplying power to the memory unit when the CPLD pulls down the power enable signal.

[0009] In an optional manner, the mainboard component also includes a hot-swap indicator light, the CPLD includes a fourth pin, and the hot-swap indicator light is electrically connected to the fourth pin; when the GPU card is in place, the CPLD inputs a high level to the hot-swap indicator light through the fourth pin to make the hot-swap indicator light always on; when the power enable signal is pulled low, a pulse signal is input to the hot-swap indicator light through the fourth pin to make the hot-swap indicator light flash; when the GPU card is not in place, a low level is input to the hot-swap indicator light through the fourth pin to turn off the hot-swap indicator light.

[0010] According to another aspect of an embodiment of the present application, a liquid cooling server is provided, comprising a chassis and a mainboard assembly provided by any one of the above embodiments, wherein the mainboard assembly is placed in the chassis.

[0011] In an optional manner, a accommodating space is provided in the chassis, and the accommodating space is used to accommodate the GPU card; the slot of the first OCP connector faces the front side of the accommodating space, so that the first OCP connector and the second OCP connector are plugged into each other when the GPU card is inserted from the front side of the accommodating space.

[0012] In an optional manner, the GPU card includes an OCP board, the GPU chip is arranged on the OCP board, the second OCP connector is arranged at one end of the OCP board, and a handle is arranged at the other end of the OCP board.

[0013] In an optional manner, a threaded fastener is further provided at the other end of the OCP board; a threaded hole corresponding to the threaded fastener is provided on the chassis; and the threaded fastener is used to be screwed into the threaded hole when the first OCP connector and the second OCP connector are plugged in, so as to fix the GPU card in the accommodating space.

[0014] The motherboard assembly provided by the embodiment of the present application, the server motherboard and the GPU card are plugged in through the second OCP connector and the first OCP connector, so that the CPU, CPLD and the auxiliary power supply can be electrically connected to the GPU chip, so that the CPU can send a PCIE signal to the GPU chip, and the CPLD can send an interrupt PCIE signal to the GPU chip in response to the hot plug instruction, so that the GPU chip caches data in response to the hot plug, and by connecting the CPLD to the auxiliary power supply, the CPLD pulls down the power enable signal after a preset number of clocks after sending the interrupt PCIE signal, so that the auxiliary power supply stops supplying power to the GPU chip, thereby realizing the hot plug of the GPU card. Through the above method, in the process of replacing the GPU card during the operation and maintenance of the entire liquid-cooled server, it is possible to support the GPU card to be powered off separately before maintenance, without shutting down the entire server, providing a single component fault operation and maintenance, and a solution that does not affect the normal operation of the entire server, avoiding affecting the work efficiency of the data center, thereby saving the operation and maintenance cost of the data center.

[0015] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0017] Figure 1 A schematic diagram of the structure of the mainboard assembly provided in an embodiment of the present application;

[0018] Figure 2 A schematic diagram of the structure of the mainboard assembly provided in an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the structure of a liquid cooling server provided in an embodiment of the present application;

[0020] Figure 4 A three-dimensional diagram of a liquid cooling server provided in an embodiment of the present application;

[0021] Figure 5 A front view structural schematic diagram of a liquid cooling server provided in an embodiment of the present application;

[0022] Figure 6 A schematic diagram of the top view of the GPU card provided in an embodiment of the present application.

[0023] The reference numerals in the specific implementation manner are as follows:

[0024] 1000, liquid cooling server;

[0025] 10. Motherboard components;

[0026] 100, server motherboard; 110, CPU; 120, CPLD; 130, auxiliary power supply; 131, first auxiliary power supply; 132, second auxiliary power supply; 140, first OCP connector; 141, slot; 200, GPU card; 210, GPU chip; 211, main chip; 212, memory unit; 220, second OCP connector; 230, OCP board; 231, handle; 232, threaded fastener; 300, hot-swap indicator light; 20, chassis; 21, accommodation space; 22, threaded hole. DETAILED DESCRIPTION

[0027] The following embodiments of the technical solution of the present application are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0029] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0030] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0031] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists, A and B exist at the same time, and B exists. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0032] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0033] In the description of the embodiments of the present application, the technical terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the embodiments of the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the embodiments of the present application.

[0034] In the description of the embodiments of the present application, unless otherwise clearly specified and limited, technical terms such as "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0035] Liquid-cooled servers are one of the key development trends of new data center server technologies in the future. Currently, manufacturers in the industry generally adopt the technical path of transforming air-cooled servers into liquid-cooled servers, that is, the original air cooling path is changed to a liquid cooling path, but the interface form and position layout of the main components in the server have not changed according to the advantages of liquid cooling, resulting in the server not reaching the optimal design solution suitable for liquid cooling.

[0036] In order to meet the demand for high computing power, in data centers, GPU cards installed in existing liquid-cooled servers are usually connected to the server motherboard using a PCIE interface. When a GPU card device fails, the data center operation and maintenance personnel must first shut down and power off the server, and then replace the faulty GPU card. In a scenario where a 2U server is typically configured with 4 GPU cards, when one of the failed GPU cards needs to be replaced, even if the other 3 GPU cards can still work normally to provide computing power services, the entire server still needs to be shut down and powered off. Therefore, the operation and maintenance process of replacing the GPU includes powering on and off the server, which is time-consuming, reduces the work efficiency of the data center, and increases the operation and maintenance costs of the data center.

[0037] For example, some technical solutions fix the GPU card in a metal shell, and then immerse the metal shell directly in the coolant. Therefore, the coolant can take away the heat transferred from the GPU card to the metal shell when flowing outside the shell. At this time, the heat dissipation path of the GPU is: GPU card → metal shell → coolant. In addition, since the GPU card is connected to the server motherboard using the PCIE interface and is assembled inside the chassis, the installation and removal of the GPU card requires the entire server to be shut down and powered off.

[0038] In other solutions, a GPU module consisting of 8 GPU chips is used to form a GPU function board, and a separate GPU liquid cooling module is used to dissipate heat for the 8 GPU chips. However, the heat dissipation method is still indirect, using specially designed pipes to load liquid. The heat generated by the 8 GPU chips is first transferred to the liquid through metal, and then the liquid takes away the heat. There are many media in the heat dissipation path, and the heat dissipation efficiency needs to be improved. In addition, when a GPU chip fails, the server of the entire GPU module has to be powered off before maintenance can be performed. Hot-swap operation and maintenance is not supported, resulting in a long operation and maintenance time.

[0039] However, in actual applications, the applicant has found that most servers in the data center (for example, at least 60% of the servers) do not need high-computing GPU cards of more than 300 watts. Based on this, in response to the above problems, the present application provides a motherboard component, including a GPU card using an OCP (Open Compute Project, Open Compute Project, International Open Source Hardware Organization) interface, which is plugged into the server motherboard through the OCP interface, so that the CPLD (Complex Programmable Logic Device) of the server motherboard can control the GPU chip cache data during the hot plug process, and control the auxiliary power supply to stop supplying power to the GPU chip, thereby realizing the hot plug of the GPU card. In this way, in the process of replacing the GPU card during the operation and maintenance of the entire liquid-cooled server, it is possible to support the GPU card to be powered off separately before maintenance, without shutting down the entire server, providing a single component fault operation and maintenance solution that does not affect the normal operation of the entire server, avoiding the impact on the work efficiency of the data center, thereby saving the operation and maintenance costs of the data center.

[0040] See also Figure 1 , Figure 1 The structural schematic diagram of the motherboard assembly provided by the embodiment of the present application is shown. As shown in the figure, the motherboard assembly 10 includes a server motherboard 100 and a GPU card 200. The server motherboard 100 includes a CPU 110, a CPLD 120, an auxiliary power supply 130 and a first OCP connector 140. The CPU 110, the CPLD 120 and the auxiliary power supply 130 are all electrically connected to the first OCP connector 140, and the CPLD 120 is electrically connected to the auxiliary power supply 130. The GPU card 200 includes an electrically connected GPU chip 210 and a second OCP connector 220. Among them, the server motherboard 100 and the GPU card 200 are plugged in through the first OCP connector 140 and the second OCP connector 220, so that the GPU chip 210 is electrically connected to the CPU 110 to receive the PCIE signal of the CPU 110. The CPLD 120 is electrically connected to the GPU chip 210 to send an interrupt PCIE signal to the GPU chip 210 in response to a hot-plug instruction, so that the GPU chip 210 caches data, and pulls down a power enable signal input to the auxiliary power supply 130 after a preset number of clock cycles. The auxiliary power supply 130 is electrically connected to the GPU chip 210, and is used to stop supplying power to the GPU chip 210 when the power enable signal is pulled low, so that the GPU card 200 can be hot-plugged.

[0041] The first OCP connector 140 and the second OCP connector 220 both use an OCP 4C+ interface, which can support the transmission of power supply circuits and data signals at the same time.

[0042] Optionally, multiple first OCP connectors 140 may be provided on the server motherboard 100, for example, 4 or 6 first OCP connectors 140 may be provided, so that multiple GPU cards 200 may be plugged into the server motherboard 100 in one server, thereby enabling each server to provide higher computing power support.

[0043] Specifically, the GPU card 200 of the present application adopts an OCP3.0 LFF (Large Form Factor) version, and a second OCP connector 220 (such as a gold finger connector) is provided at one end of the LFF board. The GPU chip 210 is mounted on the LFF board through surface mount technology (SMT), so that the GPU chip 210 is electrically connected to the second OCP connector 220. By plugging the second OCP connector 220 of the GPU card 200 into the slot of the first OCP connector 140 of the server motherboard 100, the first OCP connector 140 and the second OCP connector 220 are electrically connected, so that the GPU chip 210 can transmit signals with the server motherboard 100 through the first OCP connector 140 and the second OCP connector 220.

[0044] Figure 2 The structure diagram of the mainboard assembly provided in the embodiment of the present application is shown as follows: Figure 2 As shown, optionally, the GPU chip 210 includes a main chip 211 and a memory unit 212, wherein the main chip 211 is the core of the GPU chip 210, and its main task is to process the data input by the CPU 110, and the number of the memory unit 212 can be multiple, and its main task is to store the graphic information to be displayed and save the intermediate data of the graphic operation.

[0045] Since the CPU 110, the CPLD 120 and the auxiliary power supply 130 are all electrically connected to the first OCP connector 140, and the GPU chip 210 is electrically connected to the second OCP connector 220, after the second OCP connector 220 is plugged into the slot of the first OCP connector 140, the GPU chip 210 is electrically connected to the CPU 110, the CPLD 120 and the auxiliary power supply 130.

[0046] The maximum power consumption supported by the GPU chip 210 of the present application is 150W, and it can support 2 PCIE x16 high-speed signal channels to meet the needs of high-graphics computing services. The CPU 110 can also provide 2 groups of PCIE x16 high-speed signal channels PCIE[0:31] to the GPU chip 210. Specifically, PCIE[0:15] is a group, including PCIE[0:7] and PCIE[8:15], and PCIE[0:7] is correspondingly configured with a REFCLK clock signal and a PERST reset signal, and PCIE[8:15] is correspondingly configured with a REFCLK clock signal and a PERST reset signal; PCIE[16:31] is another group, including PCIE[16:23] and PCIE[24:31], and PCIE[16:23] is correspondingly configured with a REFCLK clock signal and a PERST reset signal, and PCIE[24:31] is correspondingly configured with a REFCLK clock signal and a PERST reset signal. It can be seen that the CPU 110 can send a PCIE signal to the GPU chip 210 through any one of the above-mentioned PCIE x16 high-speed signal channels, and the GPU chip 210 processes the data in the PCIE signal.

[0047] The auxiliary power supply 130 is an auxiliary power supply for the server mainboard 100, and is not powered off when the server mainboard 100 is restarted. Figure 2 As shown, when the server is powered on and the GPU card 200 is in place, the CPU 110 first sends an enable signal PWR_EN to the auxiliary power supply 130 to enable the auxiliary power supply 130; then, when the auxiliary power supply 130 receives the power enable signal POWER_GOOD pulled high by the CPLD 120, the auxiliary power supply 130 can supply power to the GPU chip 210.

[0048] Specifically, when the GPU card 200 is in place, that is, when the first OCP connector 140 is plugged into the second OCP connector 220, the CPLD 120 can receive the signal PCIE_PRSNTA#, and the signal PCIE_PRSNTA# is low. Then, the CPLD 120 can determine that the GPU card 200 is in place, and then pull up the power enable signal POWER_GOOD to enable the auxiliary power supply 130 to power the GPU card 200.

[0049] Optionally, the first pin 1 of the CPLD 120 is connected to the GPU chip 210 through the first OCP connector 140 and the second OCP connector 220 in sequence. When the first OCP connector 140 and the second OCP connector 220 are plugged in, the GPU chip 210 inputs a low level to the first pin 1 (that is, the input signal PCIE_PRSNTA# is low). Therefore, when the CPLD 120 detects that the first pin 1 is at a low level, it determines that the GPU card 200 is in place, and then pulls up the power enable signal input to the auxiliary power supply 130, so that the auxiliary power supply 130 supplies power to the GPU chip 210.

[0050] Alternatively, if Figure 2 As shown, the auxiliary power supply 130 may include a first auxiliary power supply 131 (VAUX12_Switch, 12V) and a second auxiliary power supply 132 (VAUX33_Switch, 3.3V), and the first auxiliary power supply 131 and the second auxiliary power supply 132 are both connected to the first OCP connector 140, and then connected to the main chip 211 and the memory unit 212 through the second OCP connector 220. Among them, the first auxiliary power supply 131 is connected to the second pin 2 of the CPLD 120, and is used to receive the power enable signal input by the CPLD 120 through the second pin 2; the second auxiliary power supply 132 is connected to the third pin 3 of the CPLD 120, and is used to receive the power enable signal input by the CPLD 120 through the third pin 3. The first auxiliary power supply 131 and the second auxiliary power supply 132 are also connected to the CPU 110, respectively. When the CPU 110 sends an enable signal PWR_EN to the first auxiliary power supply 131 and the second auxiliary power supply 132, the first auxiliary power supply 131 and the second auxiliary power supply 132 are enabled. Furthermore, the first auxiliary power supply 130 is connected to the main chip 211 through the first OCP connector 140 and the second OCP connector 220 in sequence, and the second auxiliary power supply 132 is connected to the memory unit 212 through the first OCP connector 140 and the second OCP connector 220 in sequence. Therefore, when the CPLD 120 pulls up the power enable signal input to the first auxiliary power supply 131 and the second auxiliary power supply 132, the first auxiliary power supply 131 supplies power to the main chip 211, and the second auxiliary power supply 132 supplies power to the memory unit 212; when the CPLD 120 pulls down the power enable signal input to the first auxiliary power supply 131 and the second auxiliary power supply 132, the first auxiliary power supply 131 stops supplying power to the main chip 211, and the second auxiliary power supply 132 stops supplying power to the memory unit 212.

[0051] When a user or an operation and maintenance personnel needs to hot-plug a GPU card 200 of a certain liquid-cooled server, first, a hot-plug instruction of the GPU card 200 needs to be issued to the CPU 110 of the liquid-cooled server in the control center, and the CPU 110 sends the hot-plug instruction to the CPLD 120. Then, in response to the hot-plug instruction, the CPLD 120 first sends an interruption PCIE signal PCIE_RST# to the GPU chip 210 to notify the GPU chip 210 of the message that the PCIE signal transmitted by the CPU 110 is about to be interrupted, so that the GPU chip 210 can back up and cache data and work site parameters in time. Afterwards, after a preset number of clock cycles, that is, after the GPU chip 210 completes the data cache, the CPLD 120 pulls down the power enable signal POWER_GOOD input to the auxiliary power supply 130, so that the auxiliary power supply 130 stops supplying power to the GPU chip 210. Finally, the GPU card 200 is pulled out from the slot of the first OCP connector 140 in the server motherboard 100 to achieve hot-swap of the GPU card 200. For example, if a server is configured with four GPU cards 200, if one of the GPU cards 200 fails, the operation and maintenance personnel only need to hot-swap and replace the failed GPU card 200 without shutting down the entire server, thereby not affecting the normal operation of the other three GPU cards 200, thereby ensuring the computing power support of the server.

[0052] Optionally, the clock signal of CPLD 120 may be generated by an internal timer thereof, or may use the clock signal of CPU 110 ; for example, the 13th pin CPLD_CLK of CPLD 120 may be electrically connected to CPU 110 to receive the clock signal of CPU 110 .

[0053] In the motherboard assembly 10 provided in the embodiment of the present application, the server motherboard 100 and the GPU card 200 are plugged in through the second OCP connector 220 and the first OCP connector 140, so that the CPU 110, the CPLD 120 and the auxiliary power supply 130 can all be electrically connected to the GPU chip 210, so that the CPU 110 can send a PCIE signal to the GPU chip 210, and the CPLD 120 can send an interrupt PCIE signal to the GPU chip 210 in response to a hot plug instruction, so that the GPU chip 210 caches data in response to hot plugging, and by connecting the CPLD 120 to the auxiliary power supply 130, the CPLD 120 pulls down the power enable signal after a preset number of clocks after sending the interrupt PCIE signal, so that the auxiliary power supply 130 stops supplying power to the GPU chip 210, thereby realizing hot plugging of the GPU card 200. Through the above method, during the entire liquid-cooled server operation and maintenance process of replacing the GPU card 200, it is possible to support the GPU card 200 to be powered off separately before maintenance, without the need to shut down the entire server. This provides a single component failure operation and maintenance solution that does not affect the normal operation of the entire server, avoids affecting the work efficiency of the data center, and thus saves the operation and maintenance costs of the data center.

[0054] In order to feedback the hot-swap status of the GPU card 200 to the CPU 110, the present application further proposes an implementation method, please continue to refer to Figure 2 As shown in the figure, the CPLD 120 is electrically connected to the CPU 110, and after the second OCP connector 220 is unplugged from the first OCP connector 140, the first pin 1 is at a high level, and the CPLD 120 is used to send a signal to the CPU 110 indicating that the GPU card 200 is not in place.

[0055] The pin CPLD_DATA of the CPLD 120 is connected to the CPU 110 and is used to feed back the status information of the hot-swappable GPU card 200 to the CPU 110 .

[0056] Specifically, after the GPU card 200 is hot-plugged, that is, after the second OCP connector 220 is pulled out from the slot of the first OCP connector 140, the first pin 1 is suspended, that is, the first pin 1 of the CPLD 120 is at a high level. Therefore, when the CPLD 120 detects that the first pin 1 is at a high level, it can be determined that the GPU card 200 has been pulled out, and the CPLD 120 can feed back a signal that the GPU card 200 is not in place (that is, a signal that has been hot-plugged) to the CPU 110 through the pin CPLD_DATA, so that the CPU 110 does not need to continue to wait to send a PCIE signal to the GPU chip 210.

[0057] In order to indicate the in-place status of the GPU card 200 to the operation and maintenance personnel, the present application further proposes an implementation method, please continue to refer to Figure 2 As shown in the figure, the motherboard assembly 10 also includes a hot-swap indicator light 300, and the CPLD 120 includes a fourth pin 4, and the hot-swap indicator light is electrically connected to the fourth pin 4. When the GPU card 200 is in place, the CPLD 120 inputs a high level to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 is always on. When the power enable signal is pulled low, a pulse signal is input to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 flashes. When the GPU card 200 is not in place, a low level is input to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 is turned off.

[0058] The hot-swap indicator light 300 is a light-emitting diode (LED), which can realize three states of on, off and flashing according to the control signal of the CPLD 120 .

[0059] Specifically, when the GPU card 200 is in place, the CPLD 120 inputs a high level to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 is always on, so that the operation and maintenance personnel can know that the GPU card 200 is in place. When the GPU card 200 needs to be hot-swapped, and after the CPLD 120 receives the hot-swap instruction, a pulse signal is sent to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 flashes, so that the operation and maintenance personnel can know that the hot-swap of the GPU card 200 is being performed. When the GPU card 200 is not in place, for example, after the GPU card 200 is hot-swapped, the CPLD 120 inputs a low level to the hot-swap indicator light 300 through the fourth pin 4, so that the hot-swap indicator light 300 is off, so that the operation and maintenance personnel can know that the GPU card 200 is not in place.

[0060] According to another aspect of the embodiment of the present application, a liquid cooling server is provided. Figure 3 , Figure 3 A structural schematic diagram of a liquid cooling server provided in an embodiment of the present application is shown. As shown in the figure, the liquid cooling server 1000 includes a chassis 20 and a motherboard assembly 10 provided in any one of the above embodiments, and the motherboard assembly 10 is placed in the chassis 20.

[0061] In the embodiment of the present application, the GPU card 200 adopts an immersion liquid cooling design, which is directly immersed in the coolant and directly exchanges heat with the coolant. Compared with the method of using a metal plate as an intermediate heat dissipation bridge and air cooling, the heat dissipation efficiency is higher. Therefore, the GPU chip 210 can operate in a lower temperature environment, with a longer life cycle and a lower failure rate.

[0062] In the prior art, since the GPU card is connected to the server motherboard using a PCIE interface and is installed inside the server chassis, when a GPU card device fails, the data center operation and maintenance personnel must first shut down the server node and power it off, then remove the server from the coolant, then open the server chassis cover, and finally replace the faulty GPU card. Therefore, the process of operation and maintenance to replace the GPU card not only includes the time to power on and off the server, but also the time to wait for the coolant to drain, which is time-consuming and labor-intensive, reducing the efficiency of GPU card operation and maintenance.

[0063] Therefore, under the premise of realizing hot-swappable GPU card 200, in order to improve the efficiency of operating and maintaining GPU card 200, the present application further proposes an implementation method, please refer to Figure 4 and Figure 5 , Figure 4 A three-dimensional diagram of a liquid cooling server provided in an embodiment of the present application is shown. Figure 5 The front view of the liquid cooling server provided by the embodiment of the present application is shown. As shown in the figure, a receiving space 21 is provided in the chassis 20, and the receiving space 21 is used to receive the GPU card 200. The slot 141 of the first OCP connector 140 faces the front side of the receiving space 21, so that when the GPU card 200 is inserted from the front side of the receiving space 21, the first OCP connector 140 is plugged with the second OCP connector 220.

[0064] Specifically, the size and number of the accommodation spaces of the chassis 20 can be set according to the size and number of the GPU cards 200 to be accommodated. For example, when the number of GPU cards 200 is 4, two rows of accommodation spaces aligned with each other can be set in the chassis 20, or the accommodation spaces 21 aligned with each other can be set in sequence along the vertical direction in the chassis 20.

[0065] Optionally, the accommodation space 21 may be provided at the front end of the liquid-cooled server 1000 so that the front side of the accommodation space 21 is located at the front cover of the chassis 20, or the accommodation space 21 may be provided at the rear end of the liquid-cooled server 1000 so that the front side of the accommodation space 21 is located at the back plate of the chassis 20. Usually, in order to facilitate the installation and replacement of components of the liquid-cooled server 1000, the liquid-cooled server 1000 is placed vertically in the liquid-cooled cabinet with the front end facing upwards. Therefore, it is preferred to provide the accommodation space 21 at the front end of the liquid-cooled server 1000 so that the GPU card 200 can be hot-swapped by directly opening the upper cover of the liquid-cooled cabinet, without having to lift the liquid-cooled server 1000 out of the coolant and then hot-swapped the GPU card 200.

[0066] In the present application, a slot 141 of the first OCP connector 140 of the server motherboard 100 is provided on the rear side of the accommodation space, and the slot 141 of the first OCP connector 140 faces the front side of the accommodation space 21, so that when the GPU card 200 is inserted into the accommodation space 21 from the front side of the accommodation space 21 with one end provided with the second OCP connector 220 facing the accommodation space 21, the second OCP connector 220 can be plugged into the slot 141 of the first OCP connector 140 on the rear side of the accommodation space 21, so that the GPU card 200 can be directly installed and hot-swapped on the front side of the accommodation space 21.

[0067] In this way, during the operation and maintenance of the GPU card 200, there is no need to shut down and power off the liquid-cooled server 1000, there is no need to lift the coolant out of the liquid-cooled server 1000, and there is no need to disassemble the chassis of the liquid-cooled server 1000, and there is even no need to open the front cover of the liquid-cooled server 1000. The GPU card 200 can be directly hot-swapped at the front or rear end of the liquid-cooled server 1000. Therefore, a faulty GPU card 200 can be replaced in seconds, which not only greatly saves the operation and maintenance time of the GPU card 200, but also helps to reduce the operation and maintenance costs.

[0068] In order to make it easier to plug and unplug the GPU card 200, the present application further proposes an implementation method, please continue to refer to Figure 5 , and combined with Figure 6 , Figure 6 A schematic diagram of the top structure of a GPU card provided in an embodiment of the present application is shown. As shown in the figure, the GPU card 200 includes an OCP board 230, on which a GPU chip 210 is arranged, a second OCP connector 220 is arranged at one end of the OCP board 230, and a handle 231 is arranged at the other end of the OCP board 230.

[0069] The OCP board 230 is also the LFF board mentioned above, which will not be described in detail here.

[0070] A handle 231 is provided at the other end of the OCP board 230, so that during the process of hot-plugging the GPU card 200, the operation and maintenance personnel can pull the GPU card 200 outward by holding the handle 231, thereby pulling the GPU card 200 out of the slot 141 of the first OCP connector 140. Therefore, the GPU card 200 can be pulled out of the accommodation space 21 more conveniently through the handle 231, and the GPU card 200 can be quickly hot-plugged, saving time for operating and maintaining the GPU card 200.

[0071] In order to improve the stability of the GPU card 200 plugged into the server motherboard 100, the present application further proposes an implementation method, please continue to refer to Figure 5 and Figure 6As shown in the figure, the other end of the OCP board 230 is also provided with a threaded fastener 232. The chassis 20 is provided with a threaded hole 22 corresponding to the threaded fastener 232. The threaded fastener 232 is used to be screwed into the threaded hole 22 when the first OCP connector 140 is plugged into the second OCP connector 220, so as to fix the GPU card 200 in the accommodation space 21.

[0072] The threaded fastener 232 is a hand-tightened nut.

[0073] After the GPU card 200 is inserted into the accommodating space 21, the threaded fastener 232 is screwed into the threaded hole 22 on the chassis 20, so that the GPU card 200 can be fixedly connected to the chassis 20 when the first OCP connector 140 and the second OCP connector 220 are plugged in, so that the GPU card 200 can be fixed in the accommodating space 21, so that when the GPU card 200 is immersed in the coolant, the GPU card 200 can be prevented from being separated from the slot 141 of the second OCP connector 220 during the flow of the coolant, thereby ensuring the normal operation of the GPU card 200.

Claims

1. A motherboard assembly, characterized in that: The mainboard assembly comprises: A server motherboard comprises a CPU, a CPLD, an auxiliary power supply and a first OCP connector, wherein the CPU, the CPLD and the auxiliary power supply are all electrically connected to the first OCP connector, and the CPLD is electrically connected to the auxiliary power supply; and The GPU card comprises a GPU chip and a second OCP connector which are electrically connected; wherein, The server motherboard and the GPU card are plugged in through the first OCP connector and the second OCP connector so that: The GPU chip is electrically connected to the CPU to receive a PCIE signal from the CPU; and The CPLD is electrically connected to the GPU chip to send an interrupt PCIE signal to the GPU chip in response to a hot plug instruction, so that the GPU chip caches data, and pulls down a power enable signal input to the auxiliary power supply after a preset number of clock cycles; and The auxiliary power supply is electrically connected to the GPU chip and is used to stop supplying power to the GPU chip when the power enable signal is pulled low, so as to enable hot plugging of the GPU card.

2. The motherboard assembly according to claim 1, characterized in that: The CPLD comprises a first pin, and the first pin is electrically connected to the GPU chip through the first OCP connector and the second OCP connector; The GPU chip is used to input a low level to the first pin when the first OCP connector is plugged into the second OCP connector, so that the CPLD pulls up the power enable signal input to the auxiliary power supply, so that the auxiliary power supply supplies power to the GPU chip.

3. The motherboard assembly according to claim 1, characterized in that: The auxiliary power supply is electrically connected to the CPU, and is used to be enabled after receiving an enable signal sent by the CPU, so that the GPU chip can be powered when the CPLD pulls up the power enable signal.

4. The motherboard assembly according to claim 2, characterized in that: The CPLD is electrically connected to the CPU, and after the second OCP connector is pulled out from the first OCP connector, the first pin is at a high level, and the CPLD is used to send a signal indicating that the GPU card is not in place to the CPU.

5. The motherboard assembly according to claim 1, characterized in that: The GPU chip includes a main chip and a memory unit, both of which are electrically connected to the second OCP connector; the auxiliary power supply includes a first auxiliary power supply and a second auxiliary power supply, both of which are electrically connected to the first OCP connector; the CPLD includes a second pin and a third pin; The first auxiliary power supply and the second auxiliary power supply are electrically connected to the second pin and the third pin respectively, and the CPLD inputs the power enable signal to the first auxiliary power supply and the second auxiliary power supply respectively through the second pin and the third pin; The first auxiliary power supply is electrically connected to the main chip through the first OCP connector and the second OCP connector, and is used to supply power to the main chip when the CPLD pulls up the power enable signal, and stop supplying power to the main chip when the CPLD pulls down the power enable signal; The second auxiliary power supply is electrically connected to the memory unit through the first OCP connector and the second OCP connector, and is used to supply power to the memory unit when the CPLD pulls up the power enable signal, and stop supplying power to the memory unit when the CPLD pulls down the power enable signal.

6. The motherboard assembly according to claim 1, characterized in that: The mainboard assembly further includes a hot-swap indicator light, the CPLD includes a fourth pin, and the hot-swap indicator light is electrically connected to the fourth pin; When the GPU is in place, the CPLD inputs a high level to the hot-swap indicator light through the fourth pin, so that the hot-swap indicator light is always on; After responding to a hot-plug instruction, inputting a pulse signal to the hot-plug indicator light through the fourth pin, so that the hot-plug indicator light flashes; When the GPU card is not in place, a low level is input to the hot-plug indicator light through the fourth pin to turn off the hot-plug indicator light.

7. A liquid cooling server, characterized in that: The liquid-cooled server comprises: a chassis and a motherboard assembly as described in any one of claims 1 to 6, wherein the motherboard assembly is placed in the chassis.

8. The liquid cooling server according to claim 7, characterized in that: A storage space is provided in the chassis, and the storage space is used to store the GPU card; The slot of the first OCP connector faces the front side of the accommodation space, so that the first OCP connector is plugged into the second OCP connector when the GPU card is inserted from the front side of the accommodation space.

9. The liquid cooling server according to claim 8, characterized in that: The GPU card comprises an OCP board, on which the GPU chip is arranged, one end of the OCP board is provided with the second OCP connector, and the other end of the OCP board is provided with a handle.

10. The liquid cooling server according to claim 9, characterized in that: The other end of the OCP board is also provided with a threaded fastener; The chassis is provided with threaded holes corresponding to the threaded fasteners; The threaded fastener is used to be screwed into the threaded hole when the first OCP connector is plugged into the second OCP connector, so as to fix the GPU card in the accommodation space.