A liquid-cooled server system

Abstract

The utility model discloses a liquid cooling server system relates to computer technical field, including cabinet, the cavity of forming in cabinet and the server of accommodating in the cavity, and the server includes shell, at least one GPU and interface component, still be equipped with liquid coolant in the cavity, be equipped with the driving piece for driving liquid coolant and flow in the cavity on the cabinet, and GPU is immersed in liquid coolant, and the position of interface component is higher than the liquid level of liquid coolant, and GPU includes GPU chip and the first radiator of heat conduction with GPU chip contact, and the first radiator includes the first radiating channel of the interval arrangement of many for liquid coolant to pass through, and the first radiating channel sets up along the flow direction of liquid coolant. The utility model discloses liquid cooling server system has better heat dissipation performance.

Description

Technical Field

[0001] This utility model relates to the field of computer technology, and more specifically to a liquid-cooled server system. Background Technology

[0002] A server is a computer system that needs to run 24 hours a day. It has a high hardware density. Generally, a server is configured with one CPU (Central Processing Unit) and multiple GPUs (Graphics Processing Units) in a limited space. Some servers have more GPUs and even require two or more CPUs.

[0003] Both CPUs and GPUs are energy-intensive hardware that generate a significant amount of heat during operation. If this heat is not dissipated promptly, it can lead to a decline in the performance of both components, ultimately resulting in a decrease in server performance. Typically, multiple GPUs are clustered together in a server, causing heat to accumulate and become difficult to dissipate. Therefore, servers are equipped with air-cooling systems, which include fans and air ducts. The fans guide cool outside air along the air ducts to the GPUs, carrying away the heat they generate and lowering their temperature. However, air-cooling has limited cooling capacity. When there are a small number of GPUs, it can barely maintain normal server performance, but when there are a large number of GPUs, the cooling capacity of the air-cooling system is insufficient to meet the requirements for normal server operation.

[0004] Currently, some servers use liquid cooling to dissipate heat from the CPU and GPU. Servers need to be modified to adapt to liquid cooling, while servers that rely on air cooling are difficult to adapt to liquid cooling. Utility Model Content

[0005] This invention aims to address one of the technical problems in related technologies to a certain extent. Therefore, this invention provides a liquid-cooled server system with good heat dissipation performance.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A liquid-cooled server system includes a rack, a cavity formed on the rack, and a server housed in the cavity. The server includes a casing, at least one GPU, and an interface assembly. The cavity also contains a liquid refrigerant. The rack is provided with a drive unit for driving the liquid refrigerant to flow in the cavity. The GPU is immersed in the liquid refrigerant. The interface assembly is positioned above the liquid surface of the refrigerant. The GPU includes a GPU chip and a first heat sink that contacts and conducts heat with the GPU chip. The first heat sink includes multiple spaced first heat dissipation channels for the liquid refrigerant to pass through. The first heat dissipation channels are arranged along the flow direction of the liquid refrigerant.

[0007] This utility model discloses a liquid-cooled server system, including a cabinet and a server. The server is housed in the cavity of the cabinet, and some liquid refrigerant is placed in the cavity. The liquid refrigerant can contact the server to dissipate heat from the server. The liquid refrigerant has a heat dissipation performance far superior to that of airflow, thereby enabling rapid cooling of the server.

[0008] The server comprises a casing, a GPU, and interface components. The interface components include interfaces for connecting the server to external devices, network cable ports, power interfaces, etc. The GPU, the server's primary heat-generating hardware, is housed within the casing and immersed in liquid coolant. This ensures sufficient contact between the GPU and the coolant, allowing for better heat absorption. A drive unit on the rack propels the liquid coolant to flow continuously over the GPU, further enhancing heat dissipation efficiency. The interface components generate very little heat and are positioned above the liquid coolant level, allowing for easy plugging and unplugging of cables. Therefore, the cables connecting the interface components are not submerged in the liquid coolant, preventing any obstruction of coolant flow.

[0009] A GPU comprises various hardware components: the GPU chip, capacitors, memory, etc. The majority of the GPU's heat is generated by the GPU chip. The GPU's first heatsink conducts this heat to the heatsink through contact with the GPU chip. While the GPU chip's size is limited, the first heatsink can be designed to be significantly larger, distributing the heat across its various parts and preventing heat concentration that could hinder cooling. As the first heatsink absorbs heat from the GPU chip and heats up, its heat transfer efficiency decreases. Liquid refrigerant, upon contact with the first heatsink, exchanges heat with it, cooling the heatsink and allowing it to continue rapidly absorbing heat from the GPU chip. A first cooling channel on the first heatsink allows the liquid refrigerant to pass through, increasing the contact area and improving cooling efficiency. Because the first cooling channel's orientation aligns with the refrigerant's flow direction, the sidewalls of the channel minimize obstruction, allowing the low-temperature refrigerant to continuously and rapidly pass through, ensuring efficient cooling. In this way, the server has good heat dissipation performance, ensuring that the server is in good working condition.

[0010] Optionally, the first heat dissipation channel is configured to extend through both sides in a direction perpendicular to the flow direction of the liquid refrigerant. This configuration allows the liquid refrigerant on both sides of the first heat dissipation channel to flow through it, improving heat exchange efficiency and enabling rapid cooling of the GPU.

[0011] Optionally, the first heat sink includes a first heat sink base attached to the GPU chip and multiple spaced-apart first heat sink fins connected to the first heat sink base. A first heat dissipation channel is formed between adjacent first heat sink fins. The first heat sink also includes multiple first heat pipes arranged along the length of the first heat sink fins, connecting the first heat sink base and the multiple first heat sink fins. The heat generated by the GPU chip is first directly transferred to the first heat sink base through contact, then transferred from the first heat sink base to the first heat pipes and the first heat sink fins. The heat transferred to the first heat pipes can also be transferred to the first heat sink fins, and finally dissipated through liquid refrigerant flowing through the first heat dissipation channel to fully contact the first heat sink fins. The first heat sink fins have a relatively long length, thus having a large contact area with the liquid refrigerant for rapid heat dissipation. The multiple first heat pipes can transfer temperature to different parts of the multiple first heat sink fins, preventing excessive temperature concentration on the first heat sink fins and thus improving heat dissipation efficiency.

[0012] Optionally, the rack also includes a refrigerant inlet, a refrigerant outlet, and a heat exchanger disposed outside the cavity. The driving component includes a pump that drives the liquid refrigerant from the refrigerant inlet into the cavity, and the liquid refrigerant flowing out of the cavity flows through the heat exchanger for cooling. The refrigerant outlet is located above the refrigerant inlet, and the first heat dissipation channel is arranged along the height direction. The liquid refrigerant can enter the cavity from the refrigerant inlet and flow out of the cavity through the refrigerant outlet. After the liquid refrigerant comes into contact with the first heat sink, the temperature of the first heat sink will decrease, while the temperature of the liquid refrigerant will increase. The greater the temperature difference between the liquid refrigerant and the first heat sink, the more significant the cooling effect of the liquid refrigerant on the first heat sink. By guiding the heated liquid refrigerant outside the cavity and cooling it through the heat exchanger, a large temperature difference can always be maintained between the first heat sink and the liquid refrigerant in contact with it, so as to stably cool the GPU and maintain the performance stability of the server. Under the action of the pump, liquid refrigerant continuously enters from the refrigerant inlet and flows out from the refrigerant outlet, giving the liquid refrigerant in the cavity a fixed flow direction, that is, from bottom to top. Since the low-temperature liquid refrigerant has a lower density than the high-temperature liquid refrigerant, setting the refrigerant inlet lower than the refrigerant outlet allows the low-temperature liquid refrigerant to slowly rise and diffuse in all directions after entering the cavity, thus making full contact with the first radiator for heat dissipation.

[0013] Optionally, the server includes multiple GPUs, which are spaced apart along the length of the housing, and multiple refrigerant inlets are also spaced apart along the length of the housing. Using multiple GPUs can improve the server's graphics processing performance, but it also significantly increases heat generation. The cooling method of this application, with multiple GPUs, can quickly cool down the server and maintain optimal performance. Multiple refrigerant inlets can simultaneously provide low-temperature refrigerant to different parts of the cavity, allowing for simultaneous heat dissipation from multiple GPUs and preventing one or more GPUs from overheating.

[0014] Optionally, the height of the first heatsink does not exceed 20mm, and two rows of GPUs are arranged on the outer casing along its thickness direction, with the two rows of GPUs staggered and stacked along the length direction of the outer casing. Due to its high heat dissipation efficiency, the height of the first heatsink can be set to no more than 20mm to reduce the size of the GPUs, thus allowing more GPUs to be placed on a single server and increasing its computing density. Because of the reduced GPU size, two rows of GPUs can be set on a single server, with each row containing at least two GPUs. The staggered stacking of the two rows of GPUs on the outer casing reduces the impact of the liquid coolant, which heats up after flowing through one row of GPUs, on the heat dissipation of the other row of GPUs, ensuring that both rows of GPUs can dissipate heat effectively and reducing the heat dissipation pressure caused by the increased GPU density.

[0015] Optionally, the GPU further includes a bracket fixedly mounted on the housing, a first circuit board mounted on the bracket, and a second circuit board electrically connected to the first circuit board. The GPU chip is disposed on the second circuit board. The server also includes a motherboard and a CPU disposed on the motherboard. The motherboard is provided with terminal blocks. The first circuit board is electrically connected to the terminal blocks via cables. The CPU is immersed in the liquid coolant. Both the GPU and CPU are high-heat-generating hardware components, and both are immersed in the liquid coolant for heat dissipation. Typically, the GPU is fixed to the motherboard by plugging and unplugging. However, in this application, the GPU is connected to the motherboard via cables and terminal blocks, allowing the CPU and GPU to maintain a certain distance and preventing them from affecting each other's heat dissipation efficiency.

[0016] Optionally, the bracket is L-shaped and includes a first frame and a second frame. The first frame is fixed to the bottom of the housing and supports the second circuit board. The first circuit board is fixed to the second frame and has a slot. The second circuit board has an edge connector that engages with the slot. The first and second frames have multiple through holes. The bracket allows the GPU to be fixed to the housing. The through holes in the bracket allow liquid coolant to pass through the bracket and make full contact with the first circuit board, the second circuit board, and the first heatsink, reducing the obstruction to the flow of liquid coolant.

[0017] Optionally, the server further includes a CPU immersed in the liquid refrigerant. The CPU includes a CPU chip and a second heat sink. The second heat sink includes multiple spaced second heat dissipation channels for the liquid refrigerant to pass through. The second heat dissipation channels are arranged along the flow direction of the liquid refrigerant.

[0018] Optionally, the housing is open to allow the liquid coolant to flow through the GPU; and / or, the housing has multiple through-holes through which the liquid coolant penetrates the housing. The open housing allows for rapid replacement of the internal liquid coolant, ensuring the GPU's heat dissipation efficiency; the through-holes allow the liquid coolant to enter and exit the server, facilitating rapid replacement of the internal liquid coolant and ensuring the GPU's heat dissipation efficiency.

[0019] These features and advantages of this utility model will be disclosed in detail in the following specific embodiments and accompanying drawings. The preferred embodiments or means of this utility model will be shown in detail in conjunction with the accompanying drawings, but are not intended to limit the technical solutions of this utility model. In addition, each of these features, elements and components appearing in the following text and drawings is multiple and is labeled with different symbols or numbers for convenience, but all represent parts with the same or similar structure or function. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the server system in this utility model.

[0021] Figure 2 This is a schematic diagram of the server structure in this utility model.

[0022] Figure 3 This is a schematic diagram of the server structure hidden behind the first heat sink in this utility model.

[0023] Figure 4 for Figure 3 Enlarged diagram of point A in the middle.

[0024] Figure 5 This is a schematic diagram of the structure of the first radiator in this utility model.

[0025] Figure label: Cabinet 100, cavity 110, refrigerant inlet 120, refrigerant outlet 130; Outer shell 200, second through hole 210; GPU 300, GPU chip 310, first heat sink 320, first heat dissipation channel 321, first heat sink base 322, first heat dissipation fins 323, first heat pipe 324, edge connector 330, bracket 340, first frame 341, second frame 342, first through hole 343, first circuit board 350, slot 351, second circuit board 360; Second radiator 400; Mainboard 500, terminal block 510; Interface component 600. Detailed Implementation

[0026] The embodiments of this utility model are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described are intended to explain this utility model and should not be construed as limiting it.

[0027] The terms "an embodiment," "example," or "trademark" used in this specification refer to a particular feature, structure, or characteristic described in connection with the embodiment itself that may be included in at least one embodiment disclosed in this utility model. The phrase "in an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0028] Reference Figures 1 to 5This utility model discloses a liquid-cooled server system, including a rack 100 and servers placed on the rack 100. To reduce the impact of external impurities (such as dust, insects, etc.) on the server's performance, the rack 100 is provided with a cavity 110 to house the server, thus blocking most external factors that can affect the server's performance. The server is actually a computer system, including a casing 200, at least one GPU 300, at least one CPU, a motherboard 500, memory modules, and interface components 600, etc. The GPU 300 includes a GPU chip 310, and the CPU includes a CPU chip. Both the GPU chip 310 and the CPU chip contain billions of transistors, which generate a large amount of heat during operation. If the heat is not dissipated in time, it will cause the performance of the GPU 300 and the CPU to degrade, affecting the performance of the server.

[0029] To efficiently cool the GPU 300 and CPU, the cavity 110 contains a liquid coolant (such as perfluoropolyether, hydrofluoroether, or mineral oil). The rack 100 is equipped with a drive mechanism to propel the liquid coolant within the cavity 110. Both the GPU 300 and CPU are immersed in the liquid coolant. The liquid coolant comes into contact with the GPU 300 and CPU to dissipate heat. Liquid coolant has a significantly higher cooling performance than airflow, enabling rapid cooling of the server. Furthermore, liquid coolant possesses excellent insulation properties, ensuring it does not interfere with the normal operation of the CPU and GPU 300. The drive mechanism propels the liquid coolant to flow continuously across the GPU 300 and CPU, thereby improving cooling efficiency.

[0030] Interface component 600 includes various interfaces: interfaces for connecting the server to external devices, network cable ports, power interfaces, etc. The heat generated by interface component 600 is much lower than that of GPU 300 and CPU, and generally does not require special cooling for interface component 600. For ease of use of the server, interface component 600 is positioned higher than GPU 300 and CPU. At the same time, when the server is installed in cavity 110, interface component 600 is also positioned above the liquid refrigerant level and is not submerged by the liquid refrigerant, which facilitates the plugging and unplugging of interface component 600's cables. Therefore, the various cables connected to interface component 600 will not be affected by the flow of liquid refrigerant due to immersion in liquid refrigerant.

[0031] The GPU chip 310 and the CPU chip are small in size but generate a lot of heat. In order to dissipate heat quickly, the GPU 300 also includes a first heat sink 320 that contacts the GPU chip 310 for heat conduction, and the CPU also includes a second heat sink 400 that contacts the CPU chip for heat conduction. The first heat sink 320 is larger than the GPU chip 310, and the second heat sink 400 is larger than the CPU chip. The GPU chip 310 and the CPU chip disperse heat by transferring heat to the first heat sink 320 and the second heat sink 400 respectively, avoiding heat concentration and heat dissipation difficulties, and effectively improving heat dissipation efficiency.

[0032] The first heat sink 320 and the second heat sink 400 have the same structure. Taking the first heat sink 320 as an example (the following embodiments are mainly described using the first heat sink 320; the second heat sink 400 has the same structure as the first heat sink 320 and will not be elaborated further): The first heat sink 320 includes multiple spaced first heat dissipation channels 321 for liquid refrigerant to pass through. The first heat dissipation channels 321 are arranged along the flow direction of the liquid refrigerant. As the first heat sink 320 absorbs heat from the GPU chip 310 and continuously heats up, the efficiency of heat transfer will decrease. However, when the liquid refrigerant comes into contact with the first heat sink 320, it can exchange heat with the first heat sink 320, cooling the first heat sink 320. Thus, the first heat sink 320 can continue to quickly absorb heat from the GPU chip 310. The first heat dissipation channels 321 on the first heat sink 320 allow liquid refrigerant to pass through, increasing the contact area between the liquid refrigerant and the first heat sink 320 and improving heat dissipation efficiency. Because the first heat dissipation channel 321 is oriented in the same direction as the liquid refrigerant flow, the obstruction of the liquid refrigerant flow by the sidewalls of the first heat dissipation channel 321 is reduced, allowing the low-temperature liquid refrigerant to continuously and rapidly pass through the first heat dissipation channel 321, ensuring heat dissipation efficiency. Thus, the server possesses good heat dissipation performance, ensuring the server remains in optimal working condition. Similarly, the second heat sink 400 includes multiple spaced second heat dissipation channels for the passage of liquid refrigerant, and these second heat dissipation channels are arranged along the flow direction of the liquid refrigerant.

[0033] In addition to the GPU chip 310, the GPU 300 also includes heat-generating components such as capacitors and memory. Therefore, the GPU 300 generates more heat than the CPU. The cooling priority of the GPU 300 is higher than that of the CPU. In the flow path of the liquid coolant, the CPU is located downstream of the GPU 300. That is, if the liquid coolant flows from bottom to top, the CPU is located higher than the GPU 300. Of course, if there is enough space, the CPU and GPU 300 can also be placed side by side.

[0034] Multiple servers can be installed within the rack 100, and liquid refrigerant can simultaneously cool multiple servers.

[0035] Based on the above embodiments, in one embodiment of this utility model, the first heat dissipation channel 321 is configured to extend through both sides in a direction perpendicular to the flow direction of the liquid refrigerant (i.e., the thickness direction of the first heat sink 320). With this configuration, the liquid refrigerant on both sides of the first heat dissipation channel 321 can flow through it, improving heat exchange efficiency and enabling rapid cooling of the GPU 300.

[0036] Specifically: The first heat sink 320 includes a first heat sink base 322 attached to the GPU chip 310 and multiple spaced first heat sink fins 323 connected to the first heat sink base 322. A first heat dissipation channel 321 is formed between two adjacent first heat sink fins 323. The first heat sink 320 also includes multiple first heat pipes 324 arranged along the length of the first heat sink fins 323. The first heat pipes 324 connect the first heat sink base 322 and the multiple first heat sink fins 323.

[0037] The heat generated by the GPU chip 310 is first transferred directly to the first heat sink 322 through contact, and then from the first heat sink 322 to the first heat pipe 324 and the first heat sink fins 323. The heat transferred to the first heat pipe 324 can also be transferred to the first heat sink fins 323. Finally, the heat is dissipated through the liquid refrigerant flowing through the first heat dissipation channel 321 to fully contact the first heat sink fins 323. The first heat sink fins 323 have a relatively long length, thus having a large contact area with the liquid refrigerant and enabling rapid heat dissipation. Multiple first heat pipes 324 are provided to transfer the temperature to different parts of multiple first heat sink fins 323, so as to avoid excessive temperature concentration on the first heat sink fins 323, which would affect the heat dissipation efficiency.

[0038] The first radiator 320 is made of metal material, preferably a copper alloy with excellent thermal conductivity.

[0039] The first heat dissipation fin 323 has a sheet-like structure with a very small thickness and a long length, which can quickly diffuse heat across the entire first heat dissipation fin 323, and can also perform high-efficiency heat exchange with liquid refrigerant.

[0040] The multiple first heat pipes 324 are independent of each other to avoid heat exchange between them. The multiple first heat pipes 324 are connected to different parts of the first heat sink 322, so as to transfer the temperature of different parts of the first heat sink 322 to different parts of the first heat sink fins 323.

[0041] The above describes the specific structure of the first heat sink 320. The second heat sink 400 has the same structure, including a second heat sink base, a second heat sink fin, a second heat pipe, and a second heat dissipation channel, which will not be described in detail here.

[0042] Based on the above embodiments, in one embodiment of this utility model, the cabinet 100 is further provided with a refrigerant inlet 120, a refrigerant outlet 130 and a heat exchanger disposed outside the cavity 110. The driving component includes a pump body that drives liquid refrigerant to flow from the refrigerant inlet 120 into the cavity 110, and the liquid refrigerant flowing out of the cavity 110 flows through the heat exchanger for cooling.

[0043] Liquid refrigerant can enter the cavity 110 through the refrigerant inlet 120 and flow out of the cavity 110 through the refrigerant outlet 130. After the liquid refrigerant comes into contact with the first heat sink 320, the temperature of the first heat sink 320 will decrease, while the temperature of the liquid refrigerant will increase. The greater the temperature difference between the liquid refrigerant and the first heat sink 320, the more obvious the cooling effect of the liquid refrigerant on the first heat sink 320. By guiding the heated liquid refrigerant to the outside of the cavity 110 and cooling it through the heat exchanger, the first heat sink 320 and the liquid refrigerant in contact with it can always maintain a large temperature difference, so as to stably cool the GPU 300 and maintain the performance stability of the server.

[0044] Heat exchangers can take many forms. For example, a pipe installed outside the cavity 110 on the cabinet 100 allows liquid refrigerant to flow within the pipe after leaving the cavity 110. A fan guides the airflow towards the pipe, lowering the temperature of the pipe and the liquid refrigerant inside. Alternatively, a semiconductor heat exchanger can be installed along the flow path of the pipe. Another option is to install an air compressor on the cabinet 100 to cool the liquid refrigerant outside the cavity 110.

[0045] Based on the above embodiments, in one embodiment of this utility model, the refrigerant outlet 130 is disposed above the refrigerant inlet 120, and the first heat dissipation channel 321 is disposed along the height direction.

[0046] Under the action of the pump, liquid refrigerant continuously enters from the refrigerant inlet 120 and flows out (overflows) from the refrigerant outlet 130, so that the liquid refrigerant in the cavity 110 has a fixed flow direction, that is, it flows from bottom to top. Since the low-temperature liquid refrigerant has a smaller density than the high-temperature liquid refrigerant, the refrigerant inlet 120 is set to be lower than the refrigerant outlet 130, so that the low-temperature liquid refrigerant can slowly rise and diffuse to the surroundings after entering the cavity 110, and can fully contact the first radiator 320 to dissipate heat.

[0047] The server includes multiple GPUs 300, which are spaced apart at the bottom of the housing 200 along its length. The CPU is located above the multiple GPUs 300 at the bottom of the cavity 110. Multiple refrigerant inlets 120 are spaced apart along the length of the housing 200.

[0048] Setting up multiple GPUs 300 can improve the server's graphics processing performance, but it also increases heat generation exponentially. The cooling method described in this application, with multiple GPUs 300, can quickly cool them down, maintaining optimal server performance. Multiple coolant inlets 120 can simultaneously provide low-temperature coolant to different parts of the cavity 110, dissipating heat from multiple GPUs 300 at the same time and preventing one or more GPUs 300 from overheating.

[0049] Based on the above embodiments, in one embodiment of this utility model, the height of the first heat sink 320 does not exceed 20mm (the total height of the first heat sink base 322, the first heat pipe 324, and the first heat sink fins 323), and two rows of GPUs 300 are arranged on the outer shell 200 along its thickness direction (the thickness direction of the outer shell 200 is the same as the thickness direction of the GPUs 300), and the number of GPUs 300 in each row is not less than two, and the two rows of GPUs 300 are staggered and stacked along the length direction of the outer shell 200.

[0050] Due to its high heat dissipation efficiency, the thickness of the first heatsink 320 can be set to no more than 20mm to reduce the size of the GPU 300. This allows for the placement of more GPUs 300 on a single server, increasing its computing density. Because of the reduced GPU 300 size, two rows of GPUs 300 can be placed on a single server, with at least two GPUs in each row. The two rows of GPUs 300 are staggered on the casing 200, reducing the impact of the liquid coolant heating up after flowing through one row on the heat dissipation of the other row. This ensures that both rows of GPUs 300 can dissipate heat effectively, reducing the heat dissipation pressure caused by the increased GPU 300 density.

[0051] The number of CPUs is much smaller than the number of GPUs 300, usually only 1 to 2 are needed. Therefore, if 2 CPUs are arranged side by side, there is enough space in the thickness direction of the casing. Thus, the thickness of the second heatsink 400 does not need to be reduced to the same size as the first heatsink 320.

[0052] Based on the above embodiments, in one embodiment of this utility model, the GPU 300 is usually connected to the card slot on the motherboard 500 through the edge connector 330 (also known as the gold finger). In this application, there are a large number of GPUs 300. If they are all connected to the motherboard 500 in the traditional way, the motherboard 500 will need to be larger in size, and it will also affect the position layout of the GPUs 300 on the casing 200.

[0053] The GPU 300 also includes a bracket 340 fixedly mounted on the housing 200, a first circuit board 350 mounted on the bracket 340, and a second circuit board 360 electrically connected to the first circuit board 350. The GPU chip 310 is disposed on the second circuit board 360. The motherboard 500 is provided with a terminal block 510. The first circuit board 350 is provided with a slot 351. An edge connector 330 is disposed on the second circuit board 360. The second circuit board 360 establishes an electrical connection with the first circuit board 350 by inserting and cooperating with the slot 351 through the edge connector 330. The first circuit board 350 is electrically connected to the terminal block 510 through a cable.

[0054] The above method connects the GPU 300, CPU, and motherboard 500, allowing the CPU and GPU 300 to maintain a certain distance to prevent them from affecting each other's heat dissipation efficiency. The GPU 300 can also be freely positioned and its spacing adjusted on the casing 200 as needed for heat dissipation.

[0055] The bracket 340 is L-shaped and includes a first frame 341 and a second frame 342 connected to each other. The first frame 341 is fixed to the bottom of the outer casing 200 by screws. The second frame 342 is arranged in a vertical direction. The first circuit board 350 is fixed on the second frame 342. The first frame 341 can support the second circuit board 360 after the second circuit board 360 is inserted into the first circuit board 350.

[0056] Both the first frame 341 and the second frame 342 are provided with multiple through holes 343. Under the action of the bracket 340, the GPU 300 can be fixed on the housing 200. The through holes 343 on the bracket 340 allow liquid coolant to pass through the bracket 340 and make full contact with the first circuit board 350, the second circuit board 360 and the first heat sink 320, reducing the obstruction of the bracket 340 to the flow of liquid coolant.

[0057] Based on the above embodiments, in one embodiment of this utility model, the housing 200 is open to allow liquid refrigerant to flow through the GPU 300 and CPU. The open housing 200 allows for rapid replacement of the internal liquid refrigerant, ensuring efficient heat dissipation for the GPU 300 and CPU.

[0058] In addition, the housing 200 is provided with multiple through-holes 210, through which liquid refrigerant penetrates the housing 200. The through-holes 210 on the housing 200 allow liquid refrigerant to enter and exit the server, facilitating rapid replacement of the liquid refrigerant inside the server and ensuring the heat dissipation efficiency of the GPU 300 and CPU.

[0059] This embodiment discloses two convenient ways for liquid refrigerant to flow through the first radiator and the second radiator. Combining the two methods has a better effect, but they can also be implemented individually.

[0060] The above are merely specific embodiments of this utility model, but the scope of protection of this utility model is not limited thereto. Those skilled in the art should understand that this utility model includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments above. Any modifications that do not depart from the functional and structural principles of this utility model will be included within the scope of the claims.

Claims

1. A liquid-cooled server system, comprising a rack (100), a cavity (110) formed on the rack (100), and a server housed in the cavity (110), the server comprising a casing (200), at least one GPU (300), and an interface assembly (600), characterized in that, The cavity (110) is also provided with liquid refrigerant. The cabinet (100) is provided with a drive unit for driving the liquid refrigerant to flow in the cavity (110). The GPU (300) is immersed in the liquid refrigerant. The interface component (600) is positioned above the liquid surface of the liquid refrigerant. The GPU (300) includes a GPU chip (310) and a first heat sink (320) that contacts and conducts heat with the GPU chip (310). The first heat sink (320) includes multiple spaced first heat dissipation channels (321) for the liquid refrigerant to pass through. The first heat dissipation channels (321) are arranged along the flow direction of the liquid refrigerant.

2. The liquid-cooled server system according to claim 1, characterized in that, The first heat dissipation channel (321) is provided through both sides in a direction perpendicular to the flow direction of the liquid refrigerant.

3. The liquid-cooled server system according to claim 2, characterized in that, The first heat sink (320) includes a first heat sink base (322) attached to the GPU chip (310) and multiple spaced first heat sink fins (323) connected to the first heat sink base (322). A first heat dissipation channel (321) is formed between two adjacent first heat sink fins (323). The first heat sink (320) also includes multiple first heat pipes (324) arranged along the length of the first heat sink fins (323). The first heat pipes (324) connect the first heat sink base (322) and the multiple first heat sink fins (323).

4. The liquid-cooled server system according to claim 1, characterized in that, The cabinet (100) is also provided with a refrigerant inlet (120), a refrigerant outlet (130) and a heat exchanger disposed outside the cavity (110). The driving component includes a pump body that drives the liquid refrigerant to flow from the refrigerant inlet (120) into the cavity (110). The liquid refrigerant flowing out of the cavity (110) flows through the heat exchanger for cooling. The refrigerant outlet (130) is disposed above the refrigerant inlet (120). The first heat dissipation channel (321) is disposed along the height direction.

5. The liquid-cooled server system according to claim 4, characterized in that, The server includes multiple GPUs (300) that are spaced apart along the length of the housing (200), and multiple refrigerant inlets (120) are spaced apart along the length of the housing (200).

6. The liquid-cooled server system according to claim 5, characterized in that, The first heat sink (320) has a height of no more than 20mm, and two rows of GPUs (300) are arranged on the outer shell (200) along its thickness direction. The two rows of GPUs (300) are staggered and stacked along the length direction of the outer shell (200).

7. The liquid-cooled server system according to any one of claims 1 to 6, characterized in that, The GPU (300) also includes a bracket (340) fixedly mounted on the housing (200), a first circuit board (350) mounted on the bracket (340), and a second circuit board (360) electrically connected to the first circuit board (350). The GPU chip (310) is disposed on the second circuit board (360). The server also includes a motherboard (500) and a CPU (400) disposed on the motherboard (500). The motherboard (500) is provided with a terminal block (510). The first circuit board (350) is electrically connected to the terminal block (510) via a cable. The CPU (400) is immersed in the liquid refrigerant.

8. The liquid-cooled server system according to claim 7, characterized in that, The bracket (340) is L-shaped and includes a first frame (341) and a second frame (342). The first frame (341) is fixed to the bottom of the outer shell (200) and supports the second circuit board (360). The first circuit board (350) is fixed on the second frame (342) and has a slot (351). The second circuit board (360) has an edge connector (330) that is inserted into the slot (351). The first frame (341) and the second frame (342) have multiple through holes (343).

9. The liquid-cooled server system according to any one of claims 1 to 6, characterized in that, The server also includes a CPU immersed in the liquid refrigerant. The CPU includes a CPU chip and a second heat sink (400). The second heat sink (400) includes multiple spaced second heat dissipation channels for the liquid refrigerant to pass through. The second heat dissipation channels are arranged along the flow direction of the liquid refrigerant.

10. The liquid-cooled server system according to any one of claims 1 to 6, characterized in that, The housing (200) is open to allow the liquid refrigerant to flow through the GPU (300); and / or, the housing (200) is provided with a plurality of through second holes (210) through which the liquid refrigerant penetrates the housing (200).

Images