GPU server and cabinet
By designing multiple vertically connected GPU boards in the GPU server and setting connectors on each GPU board to connect different GPUs, the problems of limited number of GPUs and low communication performance in the prior art are solved, and efficient computing power density and communication performance are achieved.
Patent Information
- Application Number
- CN202411774046.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-03-07
AI Technical Summary
In the prior art, the number of GPUs inside the server is limited, resulting in low computing power density, and GPUs in different servers communicate through the network, affecting communication performance, and thus affecting the training or inference efficiency of AI models.
By designing a GPU server, in which multiple GPU boards are connected perpendicularly to the backplane, the GPU board is equipped with a GPU and a connector. The connector is used to connect different GPUs to realize point-to-point communication and improve communication performance.
It improves the computing power density of GPU devices and the communication performance between GPUs, and improves the training or inference efficiency of AI models.
Smart Images
Figure CN119576086A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to technical fields such as cloud computing and intelligent hardware, and more particularly to a GPU server and cabinet. Background Art
[0002] Servers are the basic equipment for building cloud computing. They are equipped with central processing units (CPU), memory, hard disk, power supply and other components. With the development of artificial intelligence (AI) technology, graphics processing units (GPU) have become one of the important processors. For this reason, AI servers are usually equipped with multiple GPUs to meet AI needs. Summary of the invention
[0003] The present disclosure provides a GPU server and a cabinet.
[0004] According to one aspect of the present disclosure, a GPU server is provided, comprising: a backplane; a plurality of GPU boards, wherein the GPU boards are vertically connected to the backplane and the height of the GPU boards is a preset height; a GPU and a connector are arranged on the GPU board, wherein the connector is used to connect different GPUs.
[0005] According to another aspect of the present disclosure, a cabinet is provided, comprising: a plurality of GPU servers, each GPU server being as described in any one of the above aspects.
[0006] According to the technical solution disclosed in the present invention, the computing power density of GPU devices can be improved, and the communication performance between GPUs can be improved.
[0007] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.
[0009] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;
[0010] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;
[0011] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure;
[0012] Figure 4 is a schematic diagram of a communication method between GPUs provided according to an embodiment of the present disclosure;
[0013] Figure 5 is a schematic diagram of another communication method between GPUs provided according to an embodiment of the present disclosure;
[0014] Figure 6 is a schematic diagram according to a fourth embodiment of the present disclosure;
[0015] Figure 7 is a schematic diagram according to a fifth embodiment of the present disclosure;
[0016] Figure 8 is a schematic diagram of a replacement of a cable tray provided according to an embodiment of the present disclosure;
[0017] Fig. 9 is a schematic diagram according to a sixth embodiment of the present disclosure;
[0018] Fig.10 is a schematic diagram according to a seventh embodiment of the present disclosure;
[0019] Fig.11 is a schematic diagram of connecting multiple GPUs and multiple switches according to an embodiment of the present disclosure;
[0020] Fig.12 yes Fig.11 A schematic diagram of the connection between the corresponding GPU data transmission port and the switch;
[0021] Fig.13 is a schematic diagram of connecting multiple GPUs to a single switch according to an embodiment of the present disclosure;
[0022] Fig.14 yes Fig.13 Schematic diagram of the connection between the corresponding GPU data transmission port and the switch. DETAILED DESCRIPTION
[0023] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.
[0024] As AI models become larger and contain more parameters, more GPUs are needed to meet greater computing power requirements.
[0025] In the related art, within the same server, the GPU is generally plugged into the motherboard through a peripheral component interconnect express (PCIe) slot. PCIe is a high-speed serial computer expansion bus standard with the advantage of high data transmission rate.
[0026] GPUs in different servers generally communicate via a network (eg, Ethernet).
[0027] However, with this layout, the number of GPUs inside each server is limited, resulting in low computing density. GPUs in different servers communicate through the network, which will affect the communication performance between GPUs and thus affect the training or reasoning efficiency of the AI model.
[0028] In order to improve computing power density and communication performance, the present disclosure provides the following embodiments.
[0029] Figure 1 It is a schematic diagram according to the first embodiment of the present disclosure. This embodiment provides a GPU server.
[0030] like Figure 1 As shown, the GPU server 100 includes: a backplane 101 and a plurality of GPU boards 102a-102n. The backplane 101 is located inside the server chassis and is usually a printed circuit board (PCB) for signal transmission.
[0031] The GPU board is vertically connected to the backplane 101. Generally speaking, the backplane is vertically arranged inside the chassis, specifically perpendicular to the motherboard inside the chassis. Thus, since the GPU board is perpendicular to the backplane, the GPU board is usually horizontally arranged inside the chassis.
[0032] The GPU board is provided with a GPU and a connector, and the connector is used to connect different GPUs.
[0033] Specifically, the bottom layer of the GPU board is a PCB, and a slot can be set on the PCB, through which the GPU is plugged in, or the GPU can be integrated on the PCB, so that the GPU is set on the GPU board by plugging or integrating. A connector can also be set on the PCB, through which different GPUs are connected.
[0034] The height of the GPU board is a preset height, for example, the height of the GPU board is 1 U. Wherein, U is a height unit, 1U=1.75 inches, approximately 4.445 centimeters.
[0035] The number of GPU boards is multiple, and multiple means at least two. The specific number can be set according to the needs and the height of the GPU server where it is located.
[0036] like Figure 1 As shown, it is assumed that the GPU boards in the GPU server are represented by 102a~102n respectively, and these GPU boards are parallel to each other and vertically connected to the backplane to realize the layered arrangement of different GPU boards.
[0037] In terms of specific connection methods, the GPU board can be plugged into the backplane or connected via a cable.
[0038] Taking the GPU plugged into the GPU board as an example, each GPU board is provided with a preset number of slots, such as PCIe slots, into which the GPU is plugged. For example, the preset number is 4, so each GPU board can be plugged with a maximum of 4 GPUs.
[0039] Each GPU board is also provided with a connector, and the connector is used to connect different GPUs. The different GPUs may be located on the same GPU board or on different GPU boards.
[0040] In order to achieve high-speed communication between GPUs, the connector may specifically be a high-speed connector for transmitting data signals at a relatively high rate (eg, greater than 20 Gbps), where 20 Gbps refers to a data volume of 20 G bits per second.
[0041] In this embodiment, multiple GPU boards are vertically connected to the backplane, so that a layered setting between the multiple GPU boards can be implemented. Multiple layers of GPU boards can be set up as needed to improve computing power density. Different GPUs are connected through connectors, so that communication between GPUs can be achieved through connectors. In this way, point-to-point communication can be achieved, which can improve communication performance compared to network communication.
[0042] The above embodiments show GPU-related components. In actual scenarios, the GPU server may further include other components, such as a CPU, etc. For this purpose, the present disclosure also provides other embodiments.
[0043] Figure 2 It is a schematic diagram according to the second embodiment of the present disclosure. This embodiment provides a GPU server.
[0044] Take two GPU boards as an example. Figure 2 As shown, the GPU server 200 includes: a backplane 201 and a plurality of GPU boards 202a-202b, and also includes: a CPU board 203, which is vertically connected to the backplane. The CPU board may be provided with components such as a CPU.
[0045] In some embodiments, the GPU server may further include a switch board 204, which is vertically connected to the backplane. A network card is provided on the switch board, and different GPU servers communicate with each other through the switch board.
[0046] In terms of specific connection methods, the CPU board, GPU board and switch board can be plugged into the backplane, or they can be connected to the backplane through cables.
[0047] In this embodiment, by setting a CPU board, a switch board, etc. in the server, it can meet the needs of different scenarios and improve the server performance.
[0048] Figure 3 1 is a schematic diagram according to a third embodiment of the present disclosure, and this embodiment provides a GPU server. In order to more clearly understand the internal structure of the GPU server, this embodiment shows an exploded view of the internal components of the GPU server.
[0049] like Figure 3 As shown, the GPU server includes: a CPU board, a backplane, and a GPU board. The CPU board is provided with a CPU, and this embodiment takes two CPUs as an example. During installation, the backplane is vertically connected to the CPU board, and each GPU board is vertically connected to the backplane, so as to realize the layered arrangement of the GPU boards.
[0050] This embodiment takes two GPU boards as an example. Multiple slots are set on each GPU board so that multiple GPUs can be plugged in. In this embodiment, 4 GPUs are plugged in each GPU board as an example.
[0051] On each GPU board, multiple GPUs are arranged horizontally on the GPU board, for example, see Figure 3 , 4 GPUs are plugged horizontally into the GPU board.
[0052] In this embodiment, by horizontally arranging multiple GPUs on the GPU board, the number of GPUs can be increased, and the computing power density can be further improved. The horizontal arrangement means that the plane where these GPUs are located is horizontal relative to the GPU board, and the plane where these GPUs are located is vertical relative to the backplane. In addition, multiple slots can be set on the GPU board to plug in multiple GPUs.
[0053] Each GPU board is also provided with a connector. This embodiment takes a high-speed connector as an example, that is, a connector that can transmit data signals at a relatively high rate (such as greater than 20Gbps). Each GPU corresponds to at least one high-speed connector, and different GPUs are connected through high-speed connectors to transmit data signals.
[0054] The high-speed connectors may be connected via a cable, and the cable may be a cable for transmitting high-speed data signals, such as a PCIe cable.
[0055] For easier cable management, these cables can be placed in cable assemblies, such as Figure 3 As shown, the cable device may specifically be a cable tray, which is a device for supporting and protecting cables and is commonly used in electrical engineering.
[0056] In this embodiment, by providing a cable device, it is possible to facilitate management of cables and improve data transmission performance.
[0057] Furthermore, by using the cable tray as a cable device, the cables can be managed efficiently and conveniently.
[0058] Figure 4 It is a schematic diagram of a communication method between GPUs provided according to an embodiment of the present disclosure.
[0059] like Figure 4 As shown, assuming that the two GPUs communicating with each other are respectively the first GPU and the second GPU, and the high-speed connectors corresponding to the two GPUs are respectively called the first high-speed connector and the second high-speed connector, then the first GPU and the second GPU communicate through the first high-speed connector and the second high-speed connector. Specifically, the first high-speed connector and the second high-speed connector are connected by a cable, and the cable can specifically be a signal line for transmitting high-speed signals, such as a PCIe cable.
[0060] Taking the transmission of data signals from the first GPU to the second GPU as an example, the data signal generated by the first GPU is first transmitted to the first high-speed connector, and the first high-speed connector then transmits the data signal to the second high-speed connector through the cable placed in the cable tray, and the second high-speed connector then transmits the data signal to the second GPU.
[0061] When two GPUs communicate, they can Figure 4 As shown, direct point-to-point communication is performed. In a scenario where the number of GPU ports is limited, in order to enable a single GPU to communicate with any other GPU, a switch can be set between the GPUs to enable communication between the GPUs.
[0062] Figure 5 It is a schematic diagram of another communication method between GPUs provided according to an embodiment of the present disclosure.
[0063] like Figure 5 As shown, each GPU is connected to a switch via a high-speed connector and a cable, so that different GPUs can communicate through the switch to meet the communication needs when the number of GPU ports is small.
[0064] The above embodiments describe a GPU server, which is usually placed in a cabinet.
[0065] Figure 6 It is a schematic diagram according to a fourth embodiment of the present disclosure. This embodiment provides a cabinet.
[0066] like Figure 6 As shown, the cabinet 600 includes: a plurality of GPU servers 601a, 601b~601m.
[0067] Each GPU server is as shown in any of the above embodiments.
[0068] In this embodiment, by setting the above-mentioned GPU server in the cabinet, the computing power density of the cabinet can be increased, and the communication performance between GPUs can be improved.
[0069] The GPUs in the same GPU server mentioned above are connected via connectors. In a cabinet scenario, GPUs in different GPU servers are also connected via connectors. At this time, the connectors can also be connected via cables. Accordingly, these cables can also be managed by cable devices.
[0070] Figure 7 It is a schematic diagram according to the fifth embodiment of the present disclosure. This embodiment provides a cabinet.
[0071] like Figure 7 As shown, the cabinet includes multiple GPU servers, and a cable device, such as a cable tray, can also be installed on the cabinet, for example, fixed to the rear of the cabinet with screws.
[0072] In this way, GPUs in different GPU servers can communicate through connectors and cables placed in cable trays.
[0073] In this embodiment, by providing a cable device, it is possible to facilitate management of cables and improve data transmission performance.
[0074] Furthermore, by using the cable tray as a cable device, the cables can be managed efficiently and conveniently.
[0075] Multiple GPUs are usually arranged on each GPU board, and these GPUs can be divided into multiple groups. Accordingly, multiple cable devices can be arranged, and each cable device corresponds to a group of GPUs, that is, the cables corresponding to each group of GPUs are placed in each cable device respectively.
[0076] For example, each GPU board is provided with 4 GPUs, and these 4 GPUs can be divided into two groups, the two GPUs on the left side are one group, and the two GPUs on the right side are another group. Figure 7As shown, two cable trays may be provided, the cable tray on the left being used to place cables corresponding to the two GPUs on the left, and the cable tray on the right being used to place cables corresponding to the two GPUs on the right.
[0077] In this embodiment, by configuring the cable device into multiple groups, different cable devices can be used to place cables corresponding to different groups of GPUs, which can avoid signal transmission loss caused by excessively long cables and improve data transmission efficiency.
[0078] In the above example, each group of GPUs corresponds to one cable device. Alternatively, multiple GPUs may correspond to the same cable device, that is, there are multiple GPUs and one cable device, and the cables corresponding to the multiple GPUs are all placed in the one cable device.
[0079] For example, if only a single GPU server is needed, such as a single GPU server including two layers of GPU boards, each GPU board is provided with 4 GPUs, that is, a scenario where 8 GPUs are needed, the cables can be placed through a cable device.
[0080] In this embodiment, by corresponding one cable device to multiple GPUs, resources and space can be saved and resource utilization can be improved.
[0081] In a scenario where more GPUs are required, multiple GPU servers can be set up, each GPU server includes multiple GPU boards, and each GPU board is equipped with multiple (such as 4) GPUs. The GPUs in different GPU servers communicate through high-speed connectors and cables, and the cables are placed in a cable tray.
[0082] At this time, multiple GPU servers are layered in the cabinet along the up-down direction (vertical direction), and multiple GPU boards are also layered in the GPU server along the up-down direction (vertical direction). Figure 7 As shown, a taller cable tray is required.
[0083] In scenarios where fewer GPUs are needed, for example, only GPUs in a single GPU server are needed, such as the single GPU server includes two layers of GPU boards, each GPU board is equipped with 4 GPUs, that is, 8 GPUs are needed, the cable tray can be replaced by a smaller cable tray to replace the larger cable tray.
[0084] In this way, the cable tray can be replaced in different scenarios.
[0085] Figure 8 It is a schematic diagram of a replacement of a cable tray provided according to an embodiment of the present disclosure.
[0086] like Figure 8 As shown, when the number of GPUs required changes from more to less, the larger cable tray corresponding to more GPUs (represented by the first cable tray) can be removed first and replaced with a smaller cable tray corresponding to fewer GPUs (represented by the second cable tray). The second cable tray is used to place the cables corresponding to the two layers of GPU boards in a single GPU server.
[0087] On the contrary, if the number of GPUs required increases from a small number, the second cable tray can be replaced by the first cable tray to connect a larger number of GPUs.
[0088] In this way, by replacing the cable tray, it can be flexibly adapted to various scenarios.
[0089] When different GPUs communicate with each other, they can use Figure 4 Alternatively, the direct communication method shown in Figure 5 The communication method shown is through the switch.
[0090] To this end, a switch may also be provided in the cabinet to implement communication between GPUs through the switch.
[0091] The switch can exchange data signals for GPUs in different GPU servers, or can also exchange data signals for GPUs in the same GPU server. Based on different scenarios, the setting position of the switch can be different.
[0092] In this embodiment, by setting a switch, the switch can be used to exchange signals between different GPUs, which is suitable for scenarios with a large number of GPUs but a small number of ports, and can meet large-scale GPU communication requirements.
[0093] Fig. 9 It is a schematic diagram according to the sixth embodiment of the present disclosure. This embodiment provides a cabinet.
[0094] In this embodiment, a switch is used as an example to exchange data signals between GPUs in different GPU servers. At this time, the switch is set in the middle of multiple GPU servers.
[0095] like Fig. 9 As shown, taking four GPU servers as an example, represented by the first GPU server to the fourth GPU server respectively, the switch is placed in the middle of these GPU servers, that is, two GPU servers (the first GPU server and the second GPU server) are placed above the switch, and two GPU servers (the third GPU server and the fourth GPU server) are placed below the switch.
[0096] In this embodiment, when the switch exchanges signals between GPUs in different GPU servers, the switch is set in the middle of these GPU servers, which can ensure that different GPU servers are evenly distributed on both sides of the switch, thereby reducing signal transmission loss.
[0097] Fig.10 It is a schematic diagram according to the seventh embodiment of the present disclosure. This embodiment provides a cabinet.
[0098] In this embodiment, a switch performs data signal exchange on GPUs in the same GPU server as an example. At this time, the switch is set at an adjacent position of the GPU server.
[0099] For example, see Fig.10 , assuming that the switch needs to exchange data signals with the GPU in the first GPU server, at this time, the positions of the first GPU server and the second GPU server can be swapped so that the first GPU server is adjacent to the switch. It is understandable that the positions of the second GPU server and the switch can also be swapped so that the first GPU server is adjacent to the switch.
[0100] In this embodiment, when the switch exchanges signals between GPUs in the same GPU server, the switch is arranged at an adjacent position of the same GPU server, which can shorten the signal transmission path between GPUs and reduce signal transmission loss.
[0101] To achieve redundant backup, multiple data transmission ports can be set on a single GPU to improve reliability and transmission efficiency.
[0102] For example, each GPU has four data transmission ports. When a data transmission port is busy, other data transmission ports can be used to transmit data signals.
[0103] At this time, multiple switches may be set, for example, 4 switches may be set.
[0104] For multiple data transmission ports and multiple switches, each data transmission port may be connected to a corresponding switch, or all data transmission ports may be connected to the same switch.
[0105] Fig.11 It is a schematic diagram of connecting multiple GPUs and multiple switches according to an embodiment of the present disclosure. Fig.12 yes Fig.11 Schematic diagram of the connection between the corresponding GPU data transmission port and the switch.
[0106] like Fig.11As shown, assuming that 4 GPUs are set on each GPU board and each GPU has 4 data transmission ports, 4 switches are set, represented by the first switch to the fourth switch respectively, then each GPU is connected to all switches, and each switch is connected to all GPUs.
[0107] At this time, if Fig.12 As shown, each data transmission port is connected to a corresponding switch, so that each GPU is connected to each switch, and each switch is connected to each GPU.
[0108] For example, the four data transmission ports are respectively called the first port to the fourth port, the first port is connected to the first switch, the second port is connected to the second switch, and the rest are similar.
[0109] In specific signal transmission, taking the case where the first GPU sends a data signal to the second GPU, and both transmit data through the first port as an example, the data signal output by the first port of the first GPU is first transmitted to the first connector corresponding to the first GPU, and then transmitted to the first switch corresponding to the first port through the first connector and the cable, and then transmitted to the second GPU by the first switch through the cable and the second connector corresponding to the second GPU, specifically transmitted to the first port of the second GPU, thereby realizing signal transmission between the first GPU and the second GPU.
[0110] In this embodiment, data transmission is performed using corresponding data transmission ports and switches, which can achieve redundant backup and improve data transmission efficiency and reliability.
[0111] The above-mentioned method of connecting each data transmission port to a corresponding switch can be applied to large-scale GPU scenarios. When the number of GPUs is small, for example, in the scenario where 8 GPUs in the same GPU server communicate, only a single switch can be used for signal exchange. In this case, multiple data transmission ports on each GPU are connected to the single switch.
[0112] Fig.13 It is a schematic diagram of connecting multiple GPUs to a single switch according to an embodiment of the present disclosure. Fig.14 yes Fig.13 Schematic diagram of the connection between the corresponding GPU data transmission port and the switch.
[0113] like Fig.13 As shown, there are 8 GPUs located in the same GPU server and divided into two layers, that is, the GPU server includes two layers of GPU boards, each layer of GPU boards includes 4 GPUs, and these GPUs are all connected to the same switch.
[0114] Taking the above four switches as an example, the same switch may be any one of the above four switches, or each switch may be used as the same switch.
[0115] At this time, if Fig.14 As shown, each GPU has multiple data transmission ports. This embodiment takes four as an example, and each data transmission port is connected to the same switch, that is, all data transmission ports are connected to the same switch.
[0116] For example, the four data transmission ports are respectively called the first port to the fourth port, and the same switch is the first switch, then the first port to the fourth port are all connected to the first switch.
[0117] In this embodiment, all data transmission ports of the GPU are connected to the same switch, which can be applicable to scenarios where the GPU is relatively small and improve data transmission efficiency.
[0118] It can be understood that in the embodiments of the present disclosure, the same or similar contents in different embodiments can be referenced to each other.
[0119] It can be understood that the “first”, “second”, etc. in the embodiments of the present disclosure are only used for distinction and do not indicate the degree of importance, time sequence, etc.
[0120] It can be understood that in the embodiments of the present disclosure, the directions or positional relationships indicated by "up", "down", "left", "right", "front", "back", "inside", "outside", "top", "bottom", "vertical", "horizontal", etc. are based on the directions or positional relationships shown in the accompanying drawings, and are only used to illustrate the relative positional relationships between the various components or components, and do not particularly limit the specific installation orientation of the various components or components.
[0121] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those skilled in the art, the specific meanings of these terms in this disclosure can be understood according to specific circumstances.
[0122] In addition, the terms "installed", "set", "provided with", "provided with", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those skilled in the art, the specific meanings of the above terms in the embodiments of the present disclosure can be understood according to specific circumstances.
[0123] In addition, the shapes, structures, proportions, sizes, etc. drawn in the drawings of the present disclosure are only used to match the contents disclosed in the specification so that those skilled in the art can understand and read them, and are not used to limit the conditions under which the present disclosure can be implemented. Any modification of shape or structure, change in proportion or adjustment of size should still fall within the scope of the technical contents disclosed in the present disclosure without affecting the effects and purposes that can be achieved by the present disclosure.
[0124] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A GPU server, comprising: Back panel; A plurality of GPU boards, wherein the GPU boards are vertically connected to the backplane, and the height of the GPU boards is a preset height; The GPU board is provided with a GPU and a connector, and the connector is used to connect different GPUs.
2. The server according to claim 1, further comprising: CPU board, and / or switch board; The CPU board and / or the switch board are vertically connected to the backplane.
3. The server according to claim 1, wherein: The GPU board is provided with a plurality of slots for plugging in a plurality of GPUs; The multiple GPUs are horizontally arranged on the GPU board.
4. A cabinet comprising: Multiple GPU servers; Each GPU server is as described in any one of claims 1-3.
5. The cabinet according to claim 4, wherein: The connectors are connected via cables; The cabinet also includes: The cable device is used for placing the cable.
6. The cabinet according to claim 5, wherein: The cable device comprises: Cable tray.
7. The cabinet according to claim 6, wherein: There are multiple GPUs, and the multiple GPUs are divided into multiple groups. There are multiple cable devices, and cables corresponding to each group of GPUs are placed in each cable device respectively. or, There are multiple GPUs and one cable device, and cables corresponding to the multiple GPUs are all placed in the one cable device.
8. The cabinet according to claim 4, further comprising: A switch, wherein the GPUs in the GPU server are all connected to the switch through the connector.
9. The cabinet according to claim 8, wherein: The switch is used to perform signal exchange between GPUs in different GPU servers; The switch is located in the middle of the different GPU servers.
10. The cabinet according to claim 9, wherein: The GPU has a plurality of data transmission ports, and the number of switches is multiple; Each data transmission port of the GPU is connected to a corresponding switch.
11. The cabinet according to claim 8, wherein: The switch is used to perform signal exchange between GPUs in the same GPU server; The switch is located at an adjacent position to the same GPU server.
12. The cabinet according to claim 11, wherein: The GPU has a plurality of data transmission ports, and the switch is one; The multiple data transmission ports of the GPU are all connected to the one switch.