Ethernet card, communication module, and server

Abstract

Provided in the present disclosure are an Ethernet card, a communication module, and a server. The Ethernet card comprises a plurality of network card chips and a plurality of connectors, wherein each of the connectors is used for connecting an optical module, any of the connectors is connected to at least two network card chips, and any of the network card chips is connected to at least two connectors. A single network card integrates a plurality of independent network card chips and a plurality of connectors; any of the connectors is connected to at least two network card chips, any of the network card chips is connected to at least two connectors, and the plurality of network card chips are cross-connected to a plurality of optical module connectors, such that electrical channels of the plurality of network card chips are cross-aggregated on the plurality of connectors. When a failure occurs in an optical module on a link where any connector is located, only part of the bandwidth of the connected network card chip is affected, without causing communication interruption to any network card chip, thereby reducing the impact of the failure in the optical module or optical link on a multi-GPU parallel computing task, and improving the efficiency of the multi-GPU parallel computing task.

Description

Ethernet cards, communication modules, and servers

[0001] This disclosure claims priority to Chinese Patent Application No. 202410950126.9, filed on July 15, 2024, entitled "Ethernet Card, Communication Module and Server", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to computer technology, and more particularly to an Ethernet card, a communication module, and a server. Background Technology

[0003] In complex tasks such as training large AI models, multiple graphics processing units (GPUs) are typically used for parallel computation. After completing one round of computation, they begin to communicate with each other to synchronize the computation results. The next round of computation can only begin after all GPUs have completed the synchronization of computation results, and this process continues until the entire task is completed.

[0004] Currently, in multi-GPU parallel computing architectures, one GPU is typically configured to correspond to one network card, which is connected to an optical module. This optical module forms an optical link with the optical module on the switch. The GPUs communicate with each other through the network link to achieve synchronization of computing results.

[0005] If a network link fails during the communication process of synchronizing computation results, such as a failure of the optical module of a switch or a network card, it will affect the synchronization task of the GPU corresponding to that network link. This will affect all GPUs from starting the next stage of computation, severely impacting the overall execution efficiency of the task. Summary of the Invention

[0006] This disclosure provides an Ethernet card, a communication module, and a server to address the problem in multi-GPU parallel computing architectures where a failure of the optical module in the network link affects the synchronous task of the GPU corresponding to that network link, thus affecting all GPUs from starting the next stage of computation and severely impacting the overall task execution efficiency.

[0007] In a first aspect, this disclosure provides an Ethernet card, comprising:

[0008] Multiple network interface card (NIC) chips and multiple connectors, the connectors being used to connect optical modules;

[0009] Each of the connectors is connected to at least two of the network interface card (NIC) chips, and each of the NIC chips is connected to at least two of the connectors.

[0010] In an alternative embodiment, the connector includes a plurality of first ports for transmitting electrical signals, and the connector is connected to at least two of the network interface card chips through the plurality of first ports;

[0011] The network interface card (NIC) chip includes multiple second ports, and the NIC chip is connected to at least two of the connectors through the multiple second ports.

[0012] In an alternative embodiment, the plurality of first ports of any of the connectors are respectively connected to different network interface card (NIC) chips, wherein the number of the plurality of NIC chips is greater than or equal to the number of the first ports of any of the connectors.

[0013] In an optional embodiment, the number of first ports of any connector is equal to the number of the plurality of network interface card (NIC) chips, and any connector is connected to the plurality of NIC chips respectively through the plurality of first ports;

[0014] The number of second ports of any of the network interface cards (NICs) is equal to the number of the plurality of connectors, and each of the NICs is connected to the plurality of connectors through the plurality of second ports respectively.

[0015] In one alternative embodiment, the number of connectors is greater than or equal to the number of network interface card (NIC) chips.

[0016] Secondly, this disclosure provides an Ethernet card, including:

[0017] Multiple network interface card (NIC) chips and multiple connectors, the connectors being used to connect optical modules;

[0018] Each of the connectors is connected to one of the plurality of network interface card (NIC) chips, and each of the NIC chips is connected to one of the plurality of connectors.

[0019] In an optional embodiment, the connector includes a plurality of first ports for transmitting electrical signals, the number of the first ports of the connector being equal to the number of the plurality of network interface card (NIC) chips, and any one of the connectors is connected to the plurality of NIC chips respectively through the plurality of first ports;

[0020] The network interface card (NIC) chip includes multiple second ports, the number of which is equal to the number of connectors. Each NIC chip is connected to the multiple connectors through the multiple second ports.

[0021] In one optional embodiment, the Ethernet card includes: four network card chips and four connectors, the connectors being used to connect to an optical module;

[0022] Each of the connectors is connected to one of the four network interface card (NIC) chips, and each of the NIC chips is connected to one of the four connectors.

[0023] In one alternative embodiment, the Ethernet card includes: four network card chips and eight connectors, the connectors being used to connect to an optical module;

[0024] Each of the connectors is connected to one of the four network interface card (NIC) chips, and each of the NIC chips is connected to one of the eight connectors.

[0025] Thirdly, this disclosure provides a communication module, including:

[0026] Multiple optical modules, and an Ethernet card as described in any of the foregoing aspects;

[0027] The plurality of optical modules are respectively connected to the plurality of connectors of the Ethernet card.

[0028] Fourthly, this disclosure provides a GPU server, including:

[0029] Multiple GPUs, and the Ethernet card described in either aspect;

[0030] The plurality of GPUs correspond to the plurality of network interface cards (NICs) in the Ethernet card, and each GPU transmits data through its corresponding NIC chip, wherein any NIC chip corresponds to at least one GPU.

[0031] In an optional embodiment, the GPU server further includes a plurality of optical modules, each of which is connected to a plurality of connectors of the Ethernet card.

[0032] Fifthly, this disclosure provides a server, including:

[0033] Multiple processors, multiple optical modules, and Ethernet cards as described in any of the foregoing aspects;

[0034] The plurality of optical modules are respectively connected to the plurality of connectors of the Ethernet card.

[0035] The plurality of processors correspond to the plurality of network interface cards (NICs) in the Ethernet card, and each processor transmits data through its corresponding NIC, wherein any NIC corresponds to at least one processor.

[0036] The Ethernet card, communication module, and server disclosed herein include an Ethernet card comprising multiple network interface card (NIC) chips and multiple connectors, the connectors being used to connect optical modules; wherein any connector is connected to at least two NIC chips, and any NIC chip is connected to at least two connectors, a single NIC integrating multiple independent NIC chips and multiple connectors, and by connecting any connector to at least two NIC chips, and any NIC chip to at least two connectors, multiple NIC chips are cross-interconnected with multiple optical module connectors, so that the electrical channels of multiple NIC chips are cross-aggregated on multiple connectors. When an optical module on a link where any connector is located fails, only a portion of the bandwidth of the connected NIC chip is affected, without causing communication interruption of any NIC chip, thereby reducing the impact of optical module or optical link failures on multi-GPU parallel computing tasks and improving the efficiency of multi-GPU parallel computing tasks. Attached Figure Description

[0037] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0038] Figure 1 is a schematic diagram of the architecture of a traditional Ethernet card in a GPU server provided in an embodiment of this disclosure;

[0039] Figure 2 is a schematic diagram of the structure of an Ethernet card provided in an exemplary embodiment of this disclosure;

[0040] Figure 3 is a schematic diagram of the structure of an Ethernet card provided in another exemplary embodiment of this disclosure;

[0041] Figure 4 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0042] Figure 5 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0043] Figure 6 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0044] Figure 7 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0045] Figure 8 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0046] Figure 9 is a schematic diagram of the structure of an Ethernet card provided in yet another exemplary embodiment of this disclosure;

[0047] Figure 10 is a schematic diagram of a typical Ethernet card provided in an exemplary embodiment of this disclosure;

[0048] Figure 11 is a schematic diagram of another typical Ethernet card provided in an exemplary embodiment of the present disclosure;

[0049] Figure 12 is a schematic diagram of the structure of a communication module provided in an exemplary embodiment of the present disclosure;

[0050] Figure 13 is a schematic diagram of a typical communication module provided in an exemplary embodiment of the present disclosure;

[0051] Figure 14 is a schematic diagram of the structure of a GPU server provided in an exemplary embodiment of this disclosure;

[0052] Figure 15 is a schematic diagram of the structure of a GPU server provided in another exemplary embodiment of this disclosure;

[0053] Figure 16 is a schematic diagram of the structure of a server provided in an exemplary embodiment of this disclosure.

[0054] The accompanying drawings have illustrated specific embodiments of this disclosure, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this disclosure to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0056] It should be noted that the user information (including but not limited to user device information, user attribute information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0057] First, let's explain the terms used in this disclosure:

[0058] Ethernet switches: used for building data center networks and forwarding messages between servers, based on the Ethernet protocol.

[0059] Ethernet network interface card (NIC): also known as Ethernet card or network card, it is used by servers to connect to Ethernet switches, thereby enabling the transmission of data packets between servers.

[0060] Optical module: A photoelectric conversion module used for long-distance communication, used in conjunction with optical fiber.

[0061] Large models refer to deep learning models with a massive number of parameters, typically containing hundreds of millions, tens of billions, or even trillions of parameters. Large models are also known as foundation models (FM), which are pre-trained on large-scale unlabeled corpora to produce pre-trained models with hundreds of millions of parameters. These models can adapt to a wide range of downstream tasks and have good generalization ability. Examples include Large Language Models (LLMs) and Multi-modal Pre-training Models.

[0062] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as NLP tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios for large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0063] With the development of large-scale AI models, they have been widely applied in various fields and scenarios. In complex scenarios such as training large-scale AI models, multiple GPUs are typically used for parallel computation. After completing one round of computation, they begin to communicate with each other to synchronize the computation results. The next round of computation can only continue after all GPUs have completed the synchronization of computation results, and this process is repeated until the entire training task is completed.

[0064] In multi-GPU parallel computing architectures, each GPU is typically configured with a dedicated network interface card (NIC). This NIC connects to an optical module, which in turn forms an optical link with an optical module on a switch. GPUs communicate with each other via this network link to synchronize computation results. If the network link corresponding to a GPU fails—for example, if the optical module on the switch or the Ethernet NIC malfunctions—the link will be interrupted, affecting the synchronization task of the GPU connected to that link. This, in turn, will prevent all GPUs from starting the next stage of computation, severely impacting the overall task's execution efficiency.

[0065] Therefore, multi-GPU parallel computing tasks such as AI large model training are very sensitive to link failures. Failure of the optical module connected to the switch or network card port can cause the multi-GPU parallel computing task to be interrupted, thus affecting training efficiency.

[0066] For example, taking a server configured with 8 GPUs and 400G Ethernet cards as an example, as shown in Figure 1, a typical configuration is one 400G Ethernet card (containing a network interface card chip) for each GPU. The Ethernet card is connected to an optical module, and the four 100G electrical signals of the Ethernet card are transmitted to the corresponding optical module. The optical module connected to the Ethernet card is connected to an optical module on the switch through an optical link (such as fiber optic cable or optical channel), forming a network link for the GPU to communicate with the switch. In Figure 1, eight different numbers, "#1, #2, #3, #4, #5, #6, #7, #8", are used to distinguish different GPUs and their corresponding Ethernet cards and network interface cards. The Ethernet card and network interface card chip corresponding to the GPU have the same number as the GPU. For example, Ethernet card #1 is the network card corresponding to GPU #1, and Ethernet card #1 contains network interface card chip #1. The other numbers have the same function, and the steps are described in detail here.

[0067] If the optical module connected to the network card chip #1 in the network link corresponding to GPU #1 in Figure 1 fails, or if the optical module of the switch on the network link fails, the network link for GPU #1 to interact with the outside world will be completely interrupted. This will prevent GPU #1 from synchronizing the calculation results, which will in turn affect all GPUs from starting the next stage of calculation, and seriously affect the execution efficiency of the entire task.

[0068] This disclosure provides a novel Ethernet card, comprising: multiple network interface card (NIC) chips and multiple connectors, wherein the connectors are used to connect optical modules; any connector is connected to at least two NIC chips, and any NIC chip is connected to at least two connectors. The Ethernet card provided in this embodiment integrates multiple independent NIC chips and multiple connectors in a single NIC. By connecting any connector to at least two NIC chips and any NIC chip to at least two connectors, the multiple NIC chips are cross-interconnected with multiple optical module connectors, allowing the electrical channels of the multiple NIC chips to be cross-aggregated on multiple connectors. When an optical module on a link connected to any connector fails, only a portion of the bandwidth of the connected NIC chip is affected, without causing communication interruption for any NIC chip. This reduces the impact of optical module or optical link failures on multi-GPU parallel computing tasks and improves the efficiency of multi-GPU parallel computing tasks.

[0069] The technical solutions of this disclosure and how they solve the aforementioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this disclosure will now be described with reference to the accompanying drawings.

[0070] Figure 2 is a schematic diagram of the structure of an Ethernet card provided in an exemplary embodiment of the present disclosure. As shown in Figure 2, the Ethernet card 20 provided in the present disclosure includes multiple network card chips 21 and multiple connectors 22.

[0071] Connector 22 is used to connect an optical module. The dashed box in Figure 2 indicates that connector 22 can connect to an optical module; the Ethernet card 20 itself does not include an optical module. Connector 22 of the Ethernet card 20 can be packaged in the same form as the optical module to be connected, including but not limited to OSFP (Octal Small Form Factor Pluggable), QSFP-DD (Quad Small Form Factor Pluggable-Double Density), QSFP (Quad Small Form Factor Pluggable), and OSFP-XD (Octal Small Form Factor Pluggable-Extra Density). The specific form of connector 22 in the Ethernet card 20 can be determined based on the form of the optical module to be connected in the actual application scenario; this embodiment does not impose a specific limitation.

[0072] In this embodiment, in order to distinguish between the optical module connected to the Ethernet card 20 and the optical module of the switch, the optical module connected to the Ethernet card 20 (that is, the optical module connected to the connector 22 in the Ethernet card 20) is called the first optical module, and the optical module of the switch (that is, the optical module connected to the first optical module in the same optical fiber) is called the second optical module.

[0073] As shown in Figure 2, in the Ethernet card 20 provided in this embodiment, any connector 22 is connected to at least two network interface card (NIC) chips 21, and any NIC chip 21 is connected to at least two connectors 22. This connection method can distribute the communication bandwidth of the same optical module connected to the connector 22 to multiple NIC chips 21, so that any NIC chip 21 only uses a portion of the bandwidth of the optical module; and any NIC chip 21 is connected to multiple different optical modules, and the communication bandwidth of any NIC chip 21 is provided by multiple optical modules. In this way, when the optical link connected to any connector 22 in the Ethernet card 20 (such as the first optical module connected to the connector 22 of the Ethernet card 20, the optical fiber, or the second optical module of the switch) fails, only a portion of the bandwidth of each NIC chip 21 connected to the connector 22 is lost, while these NIC chips 21 can still communicate through the optical links of other connectors 22, without interrupting the network communication of any NIC chip 21. This can avoid interruption of the external network communication of the GPU corresponding to the NIC chip 21, reduce the impact of optical module or optical link failures on multi-GPU parallel computing tasks, and improve the efficiency of multi-GPU parallel computing tasks.

[0074] It should be noted that Figure 2 only illustrates an example of one connector 22 connecting two network interface card (NIC) chips 21 and one NIC chip 21 connecting two connectors 22. In practical applications, one connector 22 can connect two or more NIC chips 21. The communication bandwidth of different connectors 22 can be different, and the number of NIC chips 21 connected to different connectors 22 can also be different. Similarly, one NIC chip 21 can connect two or more connectors 22. The communication bandwidth of different NIC chips 21 can be different, and the number of connectors 22 connected to different NIC chips 21 can also be different. The number of NIC chips 21, the number of connectors 22, and the specific connection method between connectors 22 and NIC chips 21 in the Ethernet card 20 can be designed and configured according to the needs of the actual application scenario, and are not specifically limited here. For an Ethernet card 20 including multiple NIC chips 21 and multiple optical modules, the aforementioned technical effect can be achieved as long as any connector 22 is connected to at least two NIC chips 21, and any NIC chip 21 is connected to at least two connectors 22.

[0075] The Ethernet card 20 provided in this embodiment integrates multiple independent network card chips 21 and multiple connectors 22. Each network card 20 is connected to at least two network card chips 21 through any connector 22, and each network card chip 21 is connected to at least two connectors 22. This achieves cross-interconnection between multiple network card chips 21 and multiple optical module connectors 22, so that the electrical channels of multiple network card chips 21 are cross-aggregated on multiple connectors 22. When the optical module on the link where any connector 22 is located fails, only part of the bandwidth of the connected network card chip 21 is affected, and no communication interruption of any network card chip 21 will occur. This can reduce the impact of optical module or optical link failures on multi-GPU parallel computing tasks and improve the efficiency of multi-GPU parallel computing tasks.

[0076] Figure 3 is a schematic diagram of the structure of an Ethernet card provided in an exemplary embodiment of the present disclosure. In practical applications, as shown in Figure 3, the connector 22 includes a plurality of first ports 221 for transmitting electrical signals, and the network card chip 21 includes a plurality of second ports 211.

[0077] In this embodiment, connector 22 is connected to at least two network interface card (NIC) chips 21 through multiple first ports 221, and NIC chips 21 are connected to at least two connectors 22 through multiple second ports 211. Any first port 221 of connector 22 can be connected to any second port 211 of NIC chip 21, and one of the first ports 221 is used to receive one electrical signal from NIC chip 21.

[0078] It should be noted that Figure 3 is only used as an example to illustrate that one connector 22 contains two first ports 221 and one network card chip 21 contains two second ports 211. The number of first ports 221 of the connector 22 and the number of second ports 211 of the network card chip 21 in the Ethernet card 20 can be designed and configured according to the needs of the actual application scenario, and no specific limitation is made here.

[0079] For any connector 22, if the number of first ports 221 of the connector 22 is equal to 2, the two first ports 221 of the connector 22 are respectively connected to two different network interface card (NIC) chips 21. If the number of first ports 221 of the connector 22 is greater than 2, the connector 22 can be connected to two or more NIC chips 21 through these first ports 221, and the number of NIC chips 21 connected to the connector 22 through the first ports 221 is less than or equal to the number of first ports 221. If the number of NIC chips 21 connected to the connector 22 through the first ports 221 is less than the number of first ports 221, there are at least two different first ports 221 in the connector 22 connected to the same NIC chip 21 (different first ports 221 are connected to different second ports 211 of the same NIC chip 21).

[0080] For example, Figure 4 provides a schematic diagram of an Ethernet card structure, assuming the Ethernet card includes three connectors (connector #1, connector #2, and connector #3 as shown in Figure 4) and three network interface card (NIC) chips (NIC chip #1, NIC chip #2, and NIC chip #3 as shown in Figure 4), each connector containing three first ports, and each NIC chip containing three second ports. This example illustrates the connection relationship between the connectors and NIC chips in the Ethernet card. As shown in Figure 4, connector #1 connects to NIC chips #1 and #2 through its three first ports. The number of NIC chips connected to connector #1 is less than the number of first ports on connector #1. Two of the first ports of connector #1 are connected to two different second ports of NIC chip #1, and the remaining third first port is connected to one second port of NIC chip #2. Connector #2 connects to NIC chips #1, #2, and #3 through its three first ports, each of which is connected to one first port of one of the three NIC chips. Connector #3 connects to network interface card (NIC) chips #2 and #3 via three first ports. The number of NIC chips connected to connector #3 is less than the number of first ports on connector #3. Two of the first ports of connector #3 are connected to two different second ports of NIC chip #3, and the remaining first port is connected to one second port of NIC chip #2. Furthermore, Figure 4 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0081] In one alternative embodiment, the number of network interface cards (NICs) included in the Ethernet card is greater than or equal to the number of first ports of any connector. The multiple first ports of any connector are respectively connected to different NICs. By connecting the multiple first ports of the connector to different NICs, the impact of a failure in the optical module or optical link connected to the connector can be distributed to as many NICs as possible, thereby reducing the impact on the communication bandwidth of each NIC.

[0082] In this embodiment, the number of first ports of any connector in the Ethernet card can be less than or equal to the number of network interface card (NIC) chips in the Ethernet card, and the number of second ports of any NIC chip can be less than or equal to the number of multiple connectors. The number of connectors is greater than or equal to the number of NIC chips.

[0083] For example, Figure 5 provides a schematic diagram of an Ethernet card structure, assuming the Ethernet card includes three connectors (connector #1, connector #2, and connector #3 as shown in Figure 5) and three network interface card (NIC) chips (NIC chip #1, NIC chip #2, and NIC chip #3 as shown in Figure 5), each connector containing three first ports, and each NIC chip containing three second ports. In this case, the number of connectors in the Ethernet card equals the number of NIC chips, the number of first ports of any connector equals the number of NIC chips included in the Ethernet card, and the number of second ports of any NIC chip equals the number of connectors included in the Ethernet card. In this configuration, each connector connects to each NIC chip through multiple first ports, and each NIC chip connects to each connector through multiple second ports.

[0084] As shown in Figure 5, the three first ports of each connector can be connected to three different network interface card (NIC) chips. This connection method ensures that if any optical module or optical link connected to any connector fails, the communication bandwidth of each NIC chip will be reduced by 1 / 3, down to 2 / 3 of its original bandwidth, without causing the NIC chip to stop communicating altogether. Furthermore, Figure 5 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0085] For example, Figure 6 provides a schematic diagram of an Ethernet card structure, assuming the Ethernet card includes four connectors (connector #1, connector #2, connector #3, and connector #4 as shown in Figure 6) and two network interface card (NIC) chips (NIC chip #1 and NIC chip #2 as shown in Figure 6). Each connector contains two first ports, and each NIC chip contains four second ports. As shown in Figure 7, the number of connectors in the Ethernet card is greater than the number of NIC chips. The number of first ports of any connector equals the number of NIC chips contained in the Ethernet card, and the number of second ports of any NIC chip equals the number of connectors contained in the Ethernet card. As shown in Figure 7, the two first ports of each connector can be connected to two different NIC chips, and the four second ports of each Ethernet card can be connected to four different connectors. This connection method ensures that when the optical module or optical link connected to any connector fails, the communication bandwidth of each NIC chip is reduced by 1 / 4, down to 3 / 4 of the original communication bandwidth, without causing the NIC chip to stop communicating. Furthermore, Figure 6 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0086] For example, Figure 7 provides a schematic diagram of an Ethernet card structure. The Ethernet card includes three connectors (connector #1, connector #2, and connector #3 as shown in Figure 7) and three network interface card (NIC) chips (NIC chip #1, NIC chip #2, and NIC chip #3 as shown in Figure 7). Each connector contains two first ports, and each NIC chip contains two second ports. The number of connectors in the Ethernet card equals the number of NIC chips. The number of first ports on any connector is less than the number of NIC chips in the Ethernet card, and the number of second ports on any NIC chip is less than the number of connectors in the Ethernet card. As shown in Figure 7, the two first ports of each connector can be connected to two different NIC chips. This connection method ensures that if the optical module or optical link connected to any connector fails, the communication bandwidth of each NIC chip will be reduced by half, to half of its original bandwidth, without causing the NIC chip to stop communicating. Furthermore, Figure 7 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0087] In some alternative implementations, the number of connectors included in the Ethernet card may be less than the number of network interface card (NIC) chips. For example, Figure 8 provides a schematic diagram of an Ethernet card structure, assuming the Ethernet card includes two connectors (connector #1 and connector #2 as shown in Figure 8) and four NIC chips (NIC chip #1, NIC chip #2, NIC chip #3, and NIC chip #4 as shown in Figure 8), each connector includes four first ports, and each NIC chip includes two second ports. As shown in Figure 8, the number of connectors in the Ethernet card is less than the number of NIC chips. The number of first ports of any connector is equal to the number of NIC chips included in the Ethernet card, and the number of second ports of any NIC chip is equal to the number of connectors included in the Ethernet card. As shown in Figure 8, the four first ports of each connector can be connected to four different NIC chips, and the two second ports of each Ethernet card can be connected to two different connectors. This connection method ensures that when the optical module or optical link connected to any connector fails, the communication bandwidth of each NIC chip is reduced by half, to half of its original communication bandwidth, without causing the NIC chip to stop communicating. In addition, Figure 8 uses straight lines of different line types to represent the connections between different connectors and network card chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0088] It should be noted that Figure 8 is only an example of an Ethernet card in which the number of connectors can be less than the number of network interface card (NIC) chips. There are no specific limitations on the number of connectors and NIC chips, as well as the number of first ports of the connectors and the number of second ports of the NIC chips in this type of Ethernet card.

[0089] FIG. 9 is a schematic structural diagram of an Ethernet card provided by an exemplary embodiment of the present disclosure. A typical Ethernet card provided by an embodiment of the present disclosure, as shown in FIG. 9, the Ethernet card includes n network card chips (network card chip #1, network card chip #2,... network card chip #n shown in FIG. 9) and m connectors (connector #1, connector #2,... connector #m shown in FIG. 9), and the connectors are used to connect optical modules. Among them, both n and m are positive integers greater than or equal to 2. Any one of the connectors in the Ethernet card is respectively connected to m network card chips, and any one of the network card chips is respectively connected to n connectors.

[0090] In the Ethernet card provided in this embodiment, the number n of network card chips and the number m of connectors can be equal, or n>m, or n<m, and no specific limitation is made here in this embodiment.

[0091] In an optional implementation manner, the number of the first ports of any one of the connectors is equal to the number m of multiple network card chips, and the connector is respectively connected to m network card chips through m first ports, and each of the first ports is connected to a second port of a network card chip.

[0092] The number of the second ports of any one of the network card chips is equal to the number n of multiple connectors, and the network card chip is respectively connected to n connectors through n second ports. Each of the second ports is connected to a first port of a connector.

[0093] In another optional implementation manner, the number of the first ports of any one of the connectors can be greater than the number m of network card chips, and the connector is respectively connected to m network card chips through multiple first ports, and at least two of the first ports are respectively connected to multiple different second ports of the same network card chip. Or, the connector is respectively connected to m network card chips through m first ports, and the remaining first ports can be idle.

[0094] The number of the second ports of any one of the network card chips can be greater than the number of multiple connectors, and the network card chip is respectively connected to n connectors through multiple second ports. At least two of the second ports are connected to multiple different first ports of the same connector. Or, the network card chip is respectively connected to n network card chips through n first ports, and the remaining second ports can be idle.

[0095] Figure 10 is a schematic diagram of a typical Ethernet card structure provided by an exemplary embodiment of this disclosure. As shown in Figure 10, the Ethernet card provided in this embodiment includes four network card chips (network card chip #1, network card chip #2, network card chip #3 and network card chip #4 shown in Figure 10) and four connectors (connector #1, connector #2, connector #3 and connector #4 shown in Figure 10). The connectors are used to connect optical modules. Each connector is connected to one of the four network card chips, and each network card chip is connected to one of the four connectors.

[0096] As shown in Figure 10, each connector includes four first ports, and each network interface card (NIC) chip includes four second ports. The number of first ports on any connector equals the number of NIC chips contained in the Ethernet card. Each connector's four first ports connect to four different NIC chips. Similarly, each NIC chip has four second ports equal to the number of connectors in the Ethernet card, and each Ethernet card's four second ports connect to four different connectors. This connection method ensures that if any optical module or optical link connected to a connector fails, the communication bandwidth of each NIC chip will only decrease by 1 / 4, reducing it to 3 / 4 of its original bandwidth, without causing the NIC chip's communication to stop. Furthermore, Figure 10 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0097] For example, in Figure 10, each network interface card (NIC) chip outputs four electrical signals through four second ports, each with a bandwidth of 100G. These four electrical signals are transmitted to connectors #1, #2, #3, and #4, respectively. Each connector converts the received electrical signal into an optical signal using an optical module and transmits it outward. If the optical module or optical link fails during the subsequent transmission of one of the electrical signals, it will not affect the transmission of the remaining three signals.

[0098] Figure 11 is a schematic diagram of another typical Ethernet card structure provided by an exemplary embodiment of this disclosure. As shown in Figure 11, the Ethernet card provided in this embodiment includes four network interface card (NIC) chips (NIC chip #1, NIC chip #2, NIC chip #3 and NIC chip #4 shown in Figure 11) and eight connectors (connector #1, connector #2, connector #3, connector #4, connector #5, connector #6, connector #7 and connector #8 shown in Figure 11). The connectors are used to connect optical modules. Each connector is connected to one of the four NIC chips, and each NIC chip is connected to one of the eight connectors.

[0099] As shown in Figure 11, each connector includes four first ports, and each network interface card (NIC) chip includes eight second ports. The number of first ports on any connector equals the number of NIC chips contained in the Ethernet card. Each connector's four first ports connect to four different NIC chips. Similarly, each NIC chip has eight second ports equal to the number of connectors in the Ethernet card, and each Ethernet card's eight second ports connect to eight different connectors. This connection method ensures that if any optical module or optical link connected to a connector fails, the communication bandwidth of each NIC chip will only decrease by 1 / 8, reducing it to 7 / 8 of its original bandwidth, without causing the NIC chip to stop communicating. Furthermore, Figure 11 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0100] For example, in Figure 11, each network interface card (NIC) chip outputs eight electrical signals through eight second ports, each with a bandwidth of 100G. These eight electrical signals are transmitted to connectors #1, #2, #3, #4, #5, #6, #7, and #8, respectively. Each connector converts the received electrical signal into an optical signal using an optical module and transmits it outward. If the optical module or optical link fails during the subsequent transmission of one of the electrical signals, it will not affect the transmission of the remaining seven signals.

[0101] Figure 12 is a schematic diagram of the structure of a communication module provided in an exemplary embodiment of this disclosure. As shown in Figure 12, the communication module 10 includes multiple optical modules 30 and an Ethernet card 20 provided in any of the foregoing embodiments. The Ethernet card 20 includes multiple network card chips 21 and multiple connectors 22. The connectors 22 are used to connect the optical modules 30; any connector 22 is connected to at least two network card chips 21, and any network card chip 21 is connected to at least two connectors 22. The multiple optical modules 30 in the communication module 10 are respectively connected to the multiple connectors 22 of the Ethernet card 20.

[0102] In practical applications, connector 22 includes a slot for inserting optical module 30. Optical module 30 includes gold fingers that match the slot of connector 22. The gold fingers of optical module 30 can be inserted into the slot of connector 22, realizing the connection between connector 22 and optical module 30. The network card chip 21 of Ethernet card 20 can transmit electrical signals to optical module 30 through connector 22. Optical module 30 converts the electrical signals into optical signals and transmits the optical signals through optical fiber.

[0103] It should be noted that the Ethernet card 20 in the communication module 10 can be any of the Ethernet cards provided in the foregoing embodiments, and the communication module 10 also includes optical modules 30 that are plugged into each connector 22 of the Ethernet card 20. Based on the Ethernet card provided in any of the foregoing embodiments, a corresponding communication module can be obtained by plugging in optical modules.

[0104] Figure 13 is a schematic diagram of a typical communication module provided in an exemplary embodiment of this disclosure. As shown in Figure 13, the communication module provided in this embodiment includes an Ethernet card, which includes four network interface card (NIC) chips (NIC chip #1, NIC chip #2, NIC chip #3, and NIC chip #4 as shown in Figure 13) and four connectors (connector #1, connector #2, connector #3, and connector #4 as shown in Figure 13). Each connector is connected to one of the four NIC chips, and each NIC chip is connected to one of the four connectors. The communication module also includes four optical modules connected to the four connectors of the Ethernet card, namely optical module #1, optical module #2, optical module #3, and optical module #4 as shown in Figure 13.

[0105] As shown in Figure 13, each connector includes four first ports, and each network interface card (NIC) chip includes four second ports. The number of first ports on any connector equals the number of NIC chips contained in the Ethernet card. Each connector's four first ports connect to four different NIC chips. Similarly, each NIC chip has four second ports equal to the number of connectors in the Ethernet card, and each Ethernet card's four second ports connect to four different connectors. This connection method ensures that if any optical module or optical link connected to a connector fails, the communication bandwidth of each NIC chip will only decrease by 1 / 4, reducing it to 3 / 4 of its original bandwidth, without causing communication to cease. Furthermore, Figure 13 uses straight lines of different line types to represent the connections between different connectors and NIC chips, making it more intuitive and easier to distinguish the connections between different connectors.

[0106] Figure 14 is a schematic diagram of the structure of a GPU server provided in an exemplary embodiment of this disclosure. As shown in Figure 14, the GPU server includes: multiple GPUs, and an Ethernet card provided in any of the foregoing embodiments. The Ethernet card includes multiple network card chips and multiple connectors, the connectors being used to connect optical modules; any connector is connected to at least two network card chips, and any network card chip is connected to at least two connectors.

[0107] The GPUs in this GPU server correspond to the network interface cards (NICs) in the Ethernet card. Each GPU has a communication link with its corresponding NIC, and each GPU transmits data through its corresponding NIC.

[0108] It should be noted that in this GPU server, each network interface card (NIC) chip corresponds to at least one GPU. Figure 14 is only an illustrative example of the GPU server architecture. One NIC chip in the GPU server can correspond to one or more GPUs, and all one or more GPUs corresponding to a NIC chip transmit data externally through that NIC chip. Furthermore, in some application scenarios, one GPU in the GPU server can also correspond to one or more NIC chips. This embodiment does not specifically limit the exact correspondence between NIC chips and GPUs.

[0109] The GPU in the GPU server and the network card chip in the Ethernet card can communicate with each other through a PCIe (Peripheral Component Interconnect express, high-speed serial computer expansion bus standard) bus or PCIe interface, or other types of buses or interfaces. This embodiment does not make specific limitations here.

[0110] In addition, the GPU server may also include multiple optical modules, which are respectively connected to multiple connectors of the Ethernet card.

[0111] In practical applications, the connector includes a slot for inserting an optical module. The optical module includes gold fingers that match the connector slot. The gold fingers of the optical module can be inserted into the connector slot, thus enabling the connector and the optical module to connect. The Ethernet card's network interface chip can transmit electrical signals to the optical module through the connector. The optical module converts the electrical signals into optical signals and transmits them through the optical fiber.

[0112] It should be noted that the GPU server may include one or more Ethernet cards, and may use the Ethernet cards provided in any of the foregoing embodiments.

[0113] Figure 15 is a schematic diagram of a typical GPU server structure provided by an exemplary embodiment of this disclosure. As shown in Figure 15, taking a GPU server configured with 8 GPUs as an example, the GPU server can use two Ethernet cards as shown in Figure 10. As shown in Figure 15, the GPU server includes 8 GPUs: GPU#1, GPU#2, GPU#3, GPU#4, GPU#5, GPU#6, GPU#7, and GPU#8, and two Ethernet cards: Ethernet card 1 and Ethernet card 2.

[0114] The Ethernet card 1 includes four network interface cards (NICs): NIC #1, NIC #2, NIC #3, and NIC #4, and four connectors: connector #1, connector #2, connector #3, and connector #4. The four connectors are used to connect different optical modules. Each connector in Ethernet card 1 connects to one of the four NICs: NIC #1, NIC #2, NIC #3, and NIC #4. Each NIC chip in Ethernet card 1 connects to one of the four connectors: connector #1, connector #2, connector #3, and connector #4.

[0115] As shown in Figure 15, in Ethernet card 1, each connector includes four first ports, and each network card chip includes four second ports. The number of first ports of any connector is equal to the number of network card chips contained in the Ethernet card. The four first ports of each connector are connected to four different network card chips, and the four second ports of each Ethernet card are connected to four different connectors. This connection method ensures that when the optical module or optical link connected to any connector in Ethernet card 1 fails, the communication bandwidth of network card chips #1, #2, #3, and #4 each loses 1 / 4, reducing it to 3 / 4 of its original bandwidth, without causing any network card chip to stop communicating.

[0116] Ethernet NIC 2 includes four network interface cards (NICs): NIC #5, NIC #6, NIC #7, and NIC #8, and four connectors: connector #5, connector #6, connector #7, and connector #8. The four connectors are used to connect different optical modules. Each connector in Ethernet NIC 2 connects to one of the four NICs: NIC #5, NIC #6, NIC #7, or NIC #8. Each NIC chip in Ethernet NIC 2 connects to one of the four connectors: connector #5, connector #6, connector #7, or connector #8.

[0117] As shown in Figure 15, in Ethernet card 2, each connector includes four first ports, and each network card chip includes four second ports. The number of first ports of any connector is equal to the number of network card chips contained in the Ethernet card. The four first ports of each connector are connected to four different network card chips, and the four second ports of each Ethernet card are connected to four different connectors. This connection method ensures that when the optical module or optical link connected to any connector in Ethernet card 1 fails, the communication bandwidth of network card chips #5, #6, #7, and #8 each loses 1 / 4, reducing it to 3 / 4 of its original bandwidth, without causing any network card chip to stop communicating.

[0118] Furthermore, Figure 15 uses straight lines of different line types to represent the connections between different connectors and network interface card (NIC) chips within the same Ethernet card, making it more intuitive and easier to distinguish the connections between different connectors. Figure 15 only uses a one-to-one correspondence between NIC chips and GPUs as an example to illustrate the architecture of a GPU server. In some application scenarios, one NIC chip in a GPU server can correspond to one or more GPUs, and the one or more GPUs corresponding to the NIC chip all transmit data externally through that NIC chip. In other application scenarios, one GPU in a GPU server can also correspond to one or more NIC chips. This embodiment does not specifically limit the specific correspondence between NIC chips and GPUs.

[0119] The GPU server provided in this embodiment can be used to perform multi-GPU parallel computing tasks that are highly sensitive to link failures, such as training large AI models. In the event of a failure in the optical module or optical link connected to any connector, only a portion of the communication bandwidth of the network interface card (NIC) chip connected to that connector is lost. This does not lead to an interruption of the network link for that NIC chip, and the corresponding GPU can still use the remaining communication bandwidth to complete the synchronization of computation results. This does not affect all GPUs from starting the next stage of computation, thus improving the execution efficiency of multi-GPU parallel computing tasks.

[0120] Figure 16 is a schematic diagram of the structure of a server provided in an exemplary embodiment of this disclosure. As shown in Figure 16, the server includes: multiple processors, and an Ethernet card provided in any of the foregoing embodiments. The Ethernet card includes multiple network card chips and multiple connectors, the connectors being used to connect optical modules; any connector is connected to at least two network card chips, and any network card chip is connected to at least two connectors.

[0121] The server has multiple processors that correspond to multiple network interface cards (NICs) in the Ethernet card. Each processor has a communication link with its corresponding NIC chip, and each processor transmits data through its corresponding NIC chip. Each NIC chip corresponds to at least one processor.

[0122] It should be noted that in this server, each network interface card (NIC) chip corresponds to at least one processor. Figure 16 is only an illustrative example of the server architecture. One NIC chip in the server can correspond to one or more processors, and all one or more processors corresponding to the NIC chip transmit data externally through that NIC chip. Furthermore, in some application scenarios, one processor in the server can also correspond to one or more NIC chips. This embodiment does not specifically limit the exact correspondence between NIC chips and processors.

[0123] The processor in the server and the network card chip in the Ethernet card can communicate with each other through a PCIe (Peripheral Component Interconnect express, high-speed serial computer expansion bus standard) bus or PCIe interface, or other types of buses or interfaces. This embodiment does not make specific limitations here.

[0124] In addition, the server may include multiple optical modules, each of which is connected to a corresponding connector of the Ethernet card.

[0125] In practical applications, the connector includes a slot for inserting an optical module. The optical module includes gold fingers that match the connector slot. The gold fingers of the optical module can be inserted into the connector slot, thus enabling the connector and the optical module to connect. The Ethernet card's network interface chip can transmit electrical signals to the optical module through the connector. The optical module converts the electrical signals into optical signals and transmits them through the optical fiber.

[0126] It should be noted that the server may include one or more Ethernet cards, and may use the Ethernet cards provided in any of the foregoing embodiments.

[0127] Optionally, the server may also include other components such as storage, firewall, load balancer, communication components, and power supply components.

[0128] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules in at least one processor.

[0129] The memory may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0130] The aforementioned memory can be object storage (OSS). This memory can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] The aforementioned communication components are configured to facilitate wired or wireless communication between the device containing the communication components and other devices. The device containing the communication components can access wireless networks based on communication standards, such as mobile hotspots (WiFi), second-generation (2G), third-generation (3G), fourth-generation (4G) / Long Term Evolution (LTE), fifth-generation (5G), or combinations thereof. In one exemplary embodiment, the communication components receive broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication components also include a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), infrared, Ultra Wide Band (UWB), Bluetooth, and other technologies.

[0132] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0133] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0134] The order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, some processes described in the above embodiments and accompanying drawings include multiple operations appearing in a specific order; however, it should be clearly understood that these operations may not be executed in the order they appear herein or may be executed in parallel. The sequence numbers are merely used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types. "Multiple" means two or more, unless otherwise explicitly specified.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solutions of this disclosure, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this disclosure.

[0136] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

[0137] The above are merely preferred embodiments of this disclosure and do not limit the patent scope of this disclosure. Any equivalent structural or procedural transformations made using the content of this disclosure and its drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this disclosure.

Claims

1. An Ethernet card, wherein, Comprising: a plurality of network card chips and a plurality of connectors, the connectors being used to connect optical modules; any one of the connectors being connected to at least two of the network card chips, and any one of the network card chips being connected to at least two of the connectors.

2. The Ethernet card of claim 1, wherein the connectors comprise a plurality of first ports for transmitting electrical signals, the connectors being connected to at least two of the network card chips through the plurality of first ports; the network card chips comprise a plurality of second ports, the network card chips being connected to at least two of the connectors through the plurality of second ports.

3. The Ethernet card of claim 2, wherein the plurality of first ports of any one of the connectors are connected to different network card chips respectively, wherein the number of the plurality of network card chips is greater than or equal to the number of the first ports of any one of the connectors.

4. The Ethernet card of claim 3, wherein the number of the first ports of any one of the connectors is equal to the number of the plurality of network card chips, and any one of the connectors is connected to the plurality of network card chips through the plurality of first ports respectively; the number of the second ports of any one of the network card chips is equal to the number of the plurality of connectors, and any one of the network card chips is connected to the plurality of connectors through the plurality of second ports respectively.

5. The Ethernet card according to any of claims 1-4, wherein, the number of the connectors is greater than or equal to the number of the network card chips.

6. An Ethernet card, wherein, Comprising: a plurality of network card chips and a plurality of connectors, the connectors being used to connect optical modules; any one of the connectors being connected to the plurality of network card chips respectively, and any one of the network card chips being connected to the plurality of connectors respectively.

7. The Ethernet card of claim 6, wherein the connectors comprise a plurality of first ports for transmitting electrical signals, the number of the first ports of the connectors being equal to the number of the plurality of network card chips, and any one of the connectors being connected to the plurality of network card chips through the plurality of first ports respectively; the network card chips comprise a plurality of second ports, the number of the second ports of the network card chips being equal to the number of the plurality of connectors, and any one of the network card chips being connected to the plurality of connectors through the plurality of second ports respectively.

8. The Ethernet card of claim 6 or 7, wherein, The Ethernet card comprises: four network card chips and four connectors, the connectors being used to connect optical modules; any one of the connectors being connected to the four network card chips respectively, and any one of the network card chips being connected to the four connectors respectively.

9. The Ethernet card of claim 6 or 7, wherein, The Ethernet card comprises: four network card chips and eight connectors, the connectors being used to connect optical modules; any one of the connectors being connected to the four network card chips respectively, and any one of the network card chips being connected to the eight connectors respectively.

10. A communications module, wherein, Comprising: a plurality of optical modules, and an Ethernet card as claimed in any one of claims 1-9; the plurality of optical modules being connected to the plurality of connectors of the Ethernet card respectively.

11. A GPU server, wherein, Comprising: a plurality of GPUs, and an Ethernet card as claimed in any one of claims 1-9; the plurality of GPUs corresponding to the plurality of network card chips in the Ethernet card, each of the GPUs transmitting data through the corresponding network card chip, wherein any one of the network card chips corresponds to at least one GPU.

12. The GPU server of claim 11, wherein, Further comprising: A plurality of optical modules are respectively connected with a plurality of connectors of the Ethernet card.

13. A server, wherein, Comprise: A plurality of processors, a plurality of optical modules, and the Ethernet card according to any one of claims 1-9; A plurality of optical modules are respectively connected with a plurality of connectors of the Ethernet card. The plurality of processors correspond to a plurality of network card chips in the Ethernet card, and each processor transmits data through the corresponding network card chip, wherein any network card chip corresponds to at least one processor.

Images