Server system structure

Abstract

The invention discloses a server system structure, and relates to the technical field of server system architecture design. Comprising a 6U configuration case, GPU modules and an MLB module are arranged in the case, HDD modules and OCP / IO modules are arranged on the front portion of the case, Nvme modules, rear OCP / IO modules and fan modules are arranged on the rear portion of the case, an upper cover is arranged on the upper portion of the case, the multiple GPU modules, the multiple HDD modules and the multiple Nvme modules are connected through a GPU adapter plate, and the multiple GPU adapter plates are connected through a GPU adapter plate. The plurality of HDD modules are connected through the HDD backboard, the plurality of Nvme modules are connected through the Nvme backboard, and the GPU adapter plate, the OCP / IO module, the Nvme backboard, the rear OCP / IO module and the HDD backboard are all connected with the MLB module. According to the invention, the problem that an existing 6U server is difficult to give consideration to the industry pain point of high computing power and large-capacity storage is solved, meanwhile, various configuration combinations can be supported, specific requirements of different application scenes can be adapted, and diversified choices are provided for users.

Description

Technical Field

[0001] This invention relates to the field of server system architecture design technology, specifically to a server system structure. Background Technology

[0002] With the rapid development of emerging application scenarios such as artificial intelligence (AI), high-performance computing (HPC), and big data analytics, unprecedented demands are being placed on the computing power, storage capacity, and scalability of server systems. Especially in the context of the "AI+" era, continuous breakthroughs in general artificial intelligence technology have driven an urgent need for high-density GPU deployment and high-speed storage architectures.

[0003] Most mainstream 6U servers on the market today adopt a single-function-oriented design: either focusing on large-scale GPU-accelerated computing at the expense of large-capacity hard drives (such as 3.5-inch HDDs), or prioritizing high-density storage configurations but unable to accommodate multiple high-performance GPU cards simultaneously. This fragmented design forces users to use multiple devices to meet computing and storage needs separately when deploying mixed workloads such as AI training, scientific computing, or video rendering. This not only increases procurement costs and operational complexity but also limits overall system efficiency.

[0004] Chinese invention patent application CN119828862A discloses an AI server architecture system, including a chassis. A CPU module is installed on the lower layer of the chassis, comprising a motherboard, a first NVMe module, an I / O board, a first PCIe module, and a SATA module. A GPU module is installed on the upper layer of the chassis, comprising a UBB board, a mounting frame, and multiple AI boards. Each AI board is connected to a slot on the UBB board via a dual-density connector, and adjacent AI boards are signal-connected. A HIB module and a QSFP-56 module are also installed on the upper layer of the chassis. The HIB module is connected to the UBB board via a BTB connector and is signal-connected to the motherboard, a second PCIe module, and a second NVMe module. The QSFP-56 module is signal-connected to the UBB board. This invention improves server computing and image processing capabilities, saves on switch and network card costs, and offers convenient expansion and high flexibility.

[0005] However, the above and similar technical solutions still have the following shortcomings: existing 6U servers are unable to meet the comprehensive needs of high-density GPU computing, large-capacity HDD storage and high-speed NVMe expansion in a limited space. As a result, they lack an integrated architecture that takes into account performance, heat dissipation, modularity and efficient power supply / signal management, which restricts the deployment efficiency and cost optimization in AI and high-performance computing scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a server system architecture to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a server system architecture, including a 6U configuration chassis, wherein the chassis internally houses a GPU module and an MLB module, an HDD module and an OCP / IO module are arranged at the front, an NVMe module, a rear-mounted OCP / IO module and a fan module are arranged at the rear, and a top cover is arranged at the top. Multiple GPU modules, HDD modules, and NVMe modules are provided. Multiple GPU modules are connected via GPU adapter boards, multiple HDD modules are connected via HDD backplanes, and multiple NVMe modules are connected via NVMe backplanes. Simultaneously, the GPU adapter boards, OCP / IO modules, NVMe backplanes, rear-mounted OCP / IO modules, and HDD backplanes are all connected to the MLB modules.

[0008] Furthermore, it also includes a PSU power supply, which is connected to the rear of the chassis via a PSU backplane and is electrically connected to the GPU adapter board, OCP / IO module, NVMe backplane, rear OCP / IO module, and HDD backplane.

[0009] Furthermore, the fan module includes multiple fans, which are fixedly connected to each other via a fan backplate, and the fan backplate is electrically connected to the PSU power supply via a PSU backplate.

[0010] Furthermore, the fan backplane has multiple hot-swappable redundant power connectors, which are connected to the GPU adapter board, NVMe backplane, fan backplane, and HDD backplane.

[0011] Furthermore, the GPU adapter board has multiple PCIe 5.0 x8 slots, which are connected to the GPU module;

[0012] The NVMe backplane has multiple second SATA slots, which are connected to the NVMe module.

[0013] The HDD backplane has multiple first SATA slots, which are connected to the HDD module.

[0014] Furthermore, the MLB module has multiple MCIOx8 slots, which are connected to the PCIe 5.0x8 slot, the first SATA slot, and the second SATA slot via signal cables. The MCIOx8 slots and PCIe 5.0x8 slots are configured in a one-to-one correspondence, the MCIOx8 slots and the first SATA slots are configured in a one-to-four correspondence, and the MCIOx8 slots and the second SATA slots are configured in a one-to-two correspondence.

[0015] Furthermore, it is compatible with at least one of the following configurations, including:

[0016] 10 GPU modules, 12 HDD modules, OCP / IO modules, 8 NVMe modules, and a rear OCP / IO module;

[0017] 10 GPU modules, 12 HDD modules, OCP / IO modules or rear-mounted OCP / IO modules;

[0018] 10 GPU modules, OCP / IO modules, 32 NVMe modules, and a rear OCP / IO module;

[0019] 10 GPU modules, 24 NVMe modules, OCP / IO modules or rear OCP / IO modules;

[0020] It has 13 GPU modules, OCP / IO modules, 8 NVMe modules, and a rear OCP / IO module.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] Firstly, this invention can integrate at least 10 GPU modules, 12 HDD modules, and 8 NVMe modules within a standard 6U chassis height, thereby solving the industry pain point that existing 6U servers cannot balance high computing power and large-capacity storage. At the same time, the server system structure is not a single fixed configuration, but can support multiple configuration combinations, adapt to the specific needs of different application scenarios, and provide users with diversified choices.

[0023] Secondly, this invention provides multiple PCIe 5.0 x8 slots through the GPU adapter board, which can make full use of the high bandwidth of the PCIe 5.0 protocol, thereby ensuring high-speed data interaction between the GPU adapter board and the MLB module and reducing performance bottlenecks;

[0024] Thirdly, the PSU power supply of the present invention directly supplies power to the GPU module, NVMe module and fan module through the PSU backplane, reducing the power loss and complexity of the MLB module, improving power supply efficiency and stability, and providing a reliable guarantee for the continuous high-load operation of the system.

[0025] Fourthly, this invention provides effective air cooling for high-power GPU modules and dense hardware through a fan module composed of multiple fans, ensuring stable operation of the system under long-term high load.

[0026] Fifthly, this invention, through the direct connection of multiple hot-swappable redundant power supplies, can still operate normally when a single power supply fails, thereby avoiding the downtime of the entire server due to power supply problems and meeting the high availability requirements of data centers. Attached Figure Description

[0027] Figure 1 This is an exploded view of the server system structure in this invention;

[0028] Figure 2 This is a schematic diagram of the motherboard of the chassis in this invention;

[0029] Figure 3 This is an internal sectional view of the chassis in this invention;

[0030] Figure 4 This is another internal sectional view of the chassis in this invention;

[0031] Figure 5 This is an overall structural diagram of the 6U configuration chassis in this invention;

[0032] Figure 6 This is a schematic diagram of the signal connections of the GPU module in this invention;

[0033] Figure 7 This is a schematic diagram of the power connection of the GPU module in this invention;

[0034] Figure 8 This is a schematic diagram of the signal connection of the HDD module in this invention;

[0035] Figure 9 This is a schematic diagram of the power connection of the HDD module in this invention;

[0036] Figure 10 This is a schematic diagram of the power connection of the fan module in this invention;

[0037] Figure 11 This is a schematic diagram of the signal connection of the Nvme module in this invention;

[0038] Figure 12 This is a schematic diagram of the power connection of the NVMe module in this invention;

[0039] The component names corresponding to each number in the diagram are as follows:

[0040] 1. Top cover; 2. GPU module; 3. GPU adapter board; 4. Chassis; 5. HDD module; 6. OCP / IO module; 7. NVMe module; 8. NVMe backplate; 9. PSU power supply; 10. Rear OCP / IO module; 11. Fan module; 12. Fan backplate; 13. PSU backplate; 14. MLB module; 15. HDD backplate. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] refer to Figures 1-5 This embodiment provides a server system architecture, which includes a 6U configuration chassis. The 6U configuration chassis includes a top cover 1 and a chassis 4. The top cover 1 is located at the top of the chassis 4 and is used to seal and fix the interior of the chassis 4. The interior of the chassis 4 houses circuit boards, a GPU module 2, an HDD module 5, an OCP / IO module 6, an NVMe module 7, a rear OCP / IO module 10, and an MLB module 14. Specifically, the interior of the chassis 4 houses the GPU module 2 and the MLB module 14, the front of the chassis 4 houses the HDD module 5 and the OCP / IO module 6, and the rear of the chassis 4 houses the NVMe module 7, the rear OCP / IO module 10, and the fan module 11. In other words, in the 6U configuration chassis, HDD module 5 and OCP / IO module 6 are both located at the front of GPU module 2 and MLB module 14, while NVMe module 7, rear OCP / IO module 10, and fan module 11 are all located at the rear of GPU module 2 and MLB module 14. Specifically, in this embodiment, multiple GPU module 2, HDD module 5, OCP / IO module 6, NVMe module 7, and rear OCP / IO module 10 are provided.

[0043] Furthermore, GPU module 2 is mounted on GPU adapter board 3 for primary computing and graphics processing. HDD module 5 is mounted on HDD backplane 15 for read / write operations, data storage, and as a system disk. OCP / IO module 6 is located below HDD module 5 to enhance data center flexibility and efficiency through hardware standardization, supporting modular expansion and high-performance network requirements. Meanwhile, NVMe module 7 is mounted on NVMe backplane 8, similar to HDD module 5, for read / write operations, data storage, and as a system disk. A rear-mounted OCP / IO module 10 is located to the side of NVMe module 7 to further enhance data center flexibility and efficiency through hardware standardization, supporting modular expansion and high-performance network requirements.

[0044] Furthermore, a PSU power supply 9 is located at the rear of chassis 4 via a PSU backplate 13, and this PSU power supply 9 supplies power to the GPU module 2, HDD module 5, OCP / IO module 6, NVMe module 7, rear OCP / IO module 10, fan module 11, and MLB module 14. The fan module 11 is located at the rear of chassis 4 via a fan backplate 12, below the PSU power supply 9. The fan module 11 includes multiple fans for air cooling of the interior of chassis 4, and all fans are fixedly connected via the fan backplate 12. In other words, multiple fans are fixed to a single fan backplate 12, thus securing multiple fans to the rear of chassis 4 through this backplate.

[0045] refer to Figure 6 In this embodiment, the GPU adapter board 3 has multiple PCIe 5.0 x8 slots, and the number of PCIe 5.0 x8 slots is no less than the number of GPU modules 2. That is, the ends of multiple GPU modules 2 are inserted into their respective PCIe 5.0 x8 slots, so that multiple GPU modules 2 are concentrated on one GPU adapter board 3 through the multiple PCIe 5.0 x8 slots. Meanwhile, the MLB module 14 has multiple signal line interfaces, and each interface can connect to one GPU adapter board 3. Specifically, this embodiment has 10 GPU modules 2 and 20 PCIe 5.0 x8 slots, and the 10 GPU modules 2 can be connected to 10 of the 10 PCIe 5.0 x8 slots.

[0046] refer to Figure 8In this embodiment, the HDD backplane 15 has multiple first SATA slots, and the number of first SATA slots is not less than the number of HDD modules 5. That is, the ends of the multiple HDD modules 5 are respectively inserted into the corresponding first SATA slots, so that the multiple HDD modules 5 are concentrated on one HDD backplane 15 through the multiple first SATA slots. Meanwhile, the MLB module 14 can be connected to the HDD backplane 15 through the provided signal line interface. Specifically, this embodiment has 12 HDD modules 5 and 12 first SATA slots, with each of the 12 HDD modules 5 connected to one of the 12 first SATA slots.

[0047] refer to Figure 11 In this embodiment, the NVMe backplane 8 has multiple second SATA slots, and the number of second SATA slots is no less than the number of NVMe modules 7. That is, the ends of the multiple NVMe modules 7 are inserted into their respective second SATA slots, so that the multiple NVMe modules 7 are concentrated on one NVMe backplane 8 through the multiple second SATA slots. Meanwhile, the MLB module 14 can be connected to the NVMe backplane 8 through the provided signal cable interface. Specifically, this embodiment has 8 NVMe modules 7 and 8 second SATA slots, with each of the 8 NVMe modules 7 connected to one of the 8 second SATA slots.

[0048] refer to Figure 7 , Figure 9 , Figure 10 and Figure 12 In this embodiment, the PSU backplane 13 has multiple hot-swappable redundant power connectors, which are electrically connected to the GPU adapter board 3, NVMe backplane 8, fan backplane 12, and HDD backplane 15. In other words, the PSU power supply 9 provides power to the GPU adapter board 3, NVMe backplane 8, fan backplane 12, and HDD backplane 15 through the hot-swappable redundant power connectors on the PSU backplane 13, thus powering the GPU module 2, HDD module 5, NVMe module 7, and fan module 11.

[0049] Specifically, in this embodiment, the MLB module 14 has multiple MCIOx8 slots. These MCIOx8 slots are connected to the PCIe 5.0x8 slots on the GPU adapter board 3, the first SATA slot on the HDD backplane 15, and the second SATA slot on the NVMe backplane 8 via signal cables. It is worth noting that in this embodiment, the MCIOx8 slots and PCIe 5.0x8 slots are connected one-to-one, the MCIOx8 slots and the first SATA slots are connected in a one-to-four configuration, and the MCIOx8 slots and the second SATA slots are connected in a one-to-two configuration. In other words, this embodiment has 21 MCIOx8 slots, 10 PCIe 5.0x8 slots, 12 first SATA slots, and 8 second SATA slots. Among them, 10 MCIOx8 slots are connected to 10 PCIe 5.0x8 slots, 3 MCIOx8 slots are connected to 12 first SATA slots, and 4 MCIOx8 slots are connected to 8 second SATA slots. Thus, the remaining MCIOx8 slots can be used for reserved expansion and other configuration needs.

[0050] In other words, this embodiment can connect at least 10 GPU modules 2, 12 HDD modules 5, 1 OCP / IO module 6, 8 NVMe modules 7, and 1 rear OCP / IO module 10. Alternatively, it can connect at least 10 GPU modules 2, 12 HDD modules 5, and 1 OCP / IO module 6 or rear OCP / IO module 10. It can also connect at least 10 GPU modules 2, 1 OCP / IO module 6, 32 NVMe modules 7, and 1 rear OCP / IO module 10. Alternatively, it can connect at least 10 GPU modules 2, 24 NVMe modules 7, and 1 OCP / IO module 6 or rear OCP / IO module 10. It can also connect at least 13 GPU modules 2, 1 OCP / IO module 6, 8 NVMe modules 7, and 1 rear OCP / IO module 10.

[0051] Furthermore, the specific installation of the server system architecture provided in this embodiment is as follows:

[0052] Insert the GPU adapter board 3 into the MCIOx8 slot on the MLB module 14, and secure the GPU adapter board 3 with its accompanying heatsink (such as the included cooling fan) using screws. Then, install the MLB module 14 with the GPU adapter board 3 installed into the chassis 4, and install 32 DDR5 4800 REG memory modules in a 12-channel configuration to ensure a stable connection between the 32 DDR5 4800 REG memory modules and the MLB module 14.

[0053] Furthermore, multiple GPU modules 2 are installed on the GPU adapter board 3 via PCIe 5.0 x8 slots, and the OCP / IO module 6 is assembled inside the chassis 4. At the same time, multiple HDD modules 5 are installed on the HDD backplane 15 via the first SATA slot, and the HDD backplane 15 is connected to the hot-swappable redundant power supply on the PSU backplane 13. The HDD backplane 15 is also connected to the MCIO x8 slot on the MLB module 14.

[0054] Furthermore, the fan backplate 12 is connected to the rear of the chassis 4 using thumb screws, and multiple fans in the fan module 11 are fixed to the fan backplate 12. Simultaneously, the fan backplate 12 is connected to the hot-swappable redundant power supply on the PSU backplate 13.

[0055] Furthermore, the rear OCP / IO module 10 is positioned at the upper rear of the chassis 4 and connected to the hot-swappable redundant power connector on the PSU backplate 13 and the MCIOx8 slot on the MLB module 14. Simultaneously, an NVMe backplate 8 is installed on the side of the rear OCP / IO module 10, and multiple NVMe modules 7 are fixed to the NVMe backplate 8 via the second SATA slot. The NVMe backplate 8 is then connected to the hot-swappable redundant power connector on the PSU backplate 13 and the MCIOx8 slot on the MLB module 14. Finally, the top cover 1 is placed on the top of the chassis 4 and secured with screws, completing the installation.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. A server system architecture, comprising a 6U configuration chassis, characterized in that, The chassis (4) is equipped with a GPU module (2) and an MLB module (14) inside, an HDD module (5) and an OCP / IO module (6) at the front, an Nvme module (7), a rear OCP / IO module (10) and a fan module (11) at the rear, and a top cover (1) at the top. Multiple GPU modules (2), HDD modules (5) and Nvme modules (7) are provided. Multiple GPU modules (2) are connected through a GPU adapter board (3), multiple HDD modules (5) are connected through an HDD backplane (15), and multiple Nvme modules (7) are connected through an Nvme backplane (8). At the same time, the GPU adapter board (3), OCP / IO module (6), Nvme backplane (8), rear OCP / IO module (10) and HDD backplane (15) are all connected to the MLB module (14).

2. The server system architecture according to claim 1, characterized in that, It also includes a PSU power supply (9), which is connected to the rear of the chassis (4) via a PSU backplane (13) and is electrically connected to the GPU adapter board (3), the OCP / IO module (6), the NVMe backplane (8), the rear OCP / IO module (10), and the HDD backplane (15).

3. The server system architecture according to claim 1, characterized in that, The fan module (11) includes multiple fans, which are fixedly connected to each other via a fan backplate (12), and the fan backplate (12) is electrically connected to the PSU power supply (9) via a PSU backplate (13).

4. A server system architecture according to claim 2 or 3, characterized in that, The fan backplate (12) has multiple hot-swappable redundant power connectors, which are connected to the GPU adapter board (3), NVMe backplate (8), fan backplate (12) and HDD backplate (15).

5. A server system architecture according to claim 1 or 2, characterized in that, The GPU adapter board (3) has multiple PCIe 5.0 x8 slots, which are connected to the GPU module (2). The Nvme backplane (8) is provided with multiple second SATA slots, which are connected to the Nvme module (7); The HDD backplane (15) has multiple first SATA slots, which are connected to the HDD module (5).

6. A server system architecture according to claim 5, characterized in that, The MLB module (14) has multiple MCIOx8 slots. The MCIOx8 slots are connected to the PCIe 5.0x8 slot, the first SATA slot, and the second SATA slot via signal lines. The MCIOx8 slots and the PCIe 5.0x8 slots are configured in a one-to-one correspondence. The MCIOx8 slots and the first SATA slots are configured in a one-to-four correspondence. The MCIOx8 slots and the second SATA slots are configured in a one-to-two correspondence.

7. A server system architecture according to claim 6, characterized in that, It is compatible with at least one of the following configurations, including: 10 GPU modules (2), 12 HDD modules (5), OCP / IO modules (6), 8 NVMe modules (7) and rear OCP / IO modules (10). 10 GPU modules (2), 12 HDD modules (5), OCP / IO modules (6) or rear OCP / IO modules (10). 10 GPU modules (2), OCP / IO modules (6), 32 NVMe modules (7) and rear OCP / IO modules (10); 10 GPU modules (2), 24 NVMe modules (7), OCP / IO modules (6) or rear OCP / IO modules (10). 13 GPU modules (2), OCP / IO modules (6), 8 NVMe modules (7) and rear OCP / IO modules (10).

Images