Server system architecture

Abstract

The invention relates to the technical field of server design, in particular to a server system architecture which comprises a lower case, an upper case, a mainboard, a power supply adapter plate, a PDB adapter plate, two GPU modules and two GPU adapter plates, the mainboard is installed in the lower case, the power supply adapter plate is installed in the lower case, the PDB adapter plate is installed in the lower case, and the two GPU adapter plates are installed in the upper case. The two GPU adapter plates are respectively mounted in the lower case and the upper case, and the two GPU modules are respectively connected with the two GPU adapter plates; the mainboard is used for installing a CPU (Central Processing Unit) and a memory to form an electronic system, and is provided with a plurality of PCIe5.0 * 16 slots and a plurality of PCIe5.0 * 8 slots; the power supply adapter plate is used for shunting a PSU direct-current power supply to other plates in the server; the PDB adapter plate is used for being connected with a GPU power interface and supplying power to the GPU; the GPU module is used for calculation and graphic processing; and the GPU adapter plate is used for connecting the GPU and the PCIe and transmitting the CPU signal to the GPU. And server performance and cost optimization are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of server design, and in particular to a server system architecture. Background Art

[0002] With the advent of the digital age, artificial intelligence technology has become a new productivity that promotes the transformation and upgrading of all industries. The continuous breakthroughs in general artificial intelligence technology have greatly expanded the application scenarios and scope of "artificial intelligence +", and we have truly entered a new stage of "empowering new quality with artificial intelligence". In this context, the demand for computing power infrastructure and innovative applications has grown rapidly, and higher requirements have been placed on server systems in the fields of high-performance computing (HPC) and artificial intelligence (AI).

[0003] Existing server architectures mostly use dual-node designs such as 2 / 4 / 6 / 8U. These architectures cannot meet customers' needs for cost savings while using functions.

[0004] The number of nodes placed in conventional cabinets on the market is even, resulting in the lack of a single-node 9U architecture that is compatible with 12 GPUs. This limits the improvement of server performance and cost optimization. Summary of the invention

[0005] The purpose of the present invention is to provide a server system architecture, aiming to solve the problem that the number of nodes of the server placed in the existing conventional cabinets on the market is even, resulting in the failure to adopt a single-node 9U architecture that is compatible with 12 GPUs, which limits the improvement of server performance and cost optimization.

[0006] To achieve the above-mentioned object, the present invention provides a server system architecture, comprising a lower chassis and an upper chassis, wherein the upper chassis is located above the lower chassis, and further comprises a motherboard, a power adapter board, a PDB adapter board, two GPU modules and two GPU adapter boards, wherein the motherboard is installed in the lower chassis, the power adapter board is installed in the lower chassis, the PDB adapter board is installed in the lower chassis, the two GPU adapter boards are respectively installed in the lower chassis and the upper chassis, and the two GPU modules are respectively connected to the two GPU adapter boards;

[0007] The motherboard is used to install the CPU and memory to form an electronic system, and has multiple PCIe5.0x16 slots and multiple PCIe5.0x8 slots;

[0008] The power adapter board is used to shunt the PSU DC power to other boards inside the server;

[0009] The PDB adapter board is used to connect to the GPU power interface to supply power to the GPU;

[0010] The GPU module is used for computing and graphics processing;

[0011] The GPU adapter board is used to connect the GPU and PCIe to transmit CPU signals to the GPU.

[0012] The server system architecture further includes an NVME module and a PCle module, wherein the NVME module is installed on the front window of the lower chassis, and the PCle module is inserted into the card slot of the motherboard;

[0013] The NVME module is used for reading and writing, storing data and serving as a system disk;

[0014] The PCle module is used for signal conversion and transmission, and multiple PCIe5.0 adapter cards are inserted.

[0015] The server system architecture further includes a power supply and a hard disk backplane, wherein the power supply is installed at the rear of the lower chassis, and the hard disk backplane is installed behind the NVME module;

[0016] The power supply is used to convert the external AC power into DC power;

[0017] The hard disk backplane is used to supply power to the NVME module and connect the signal to the motherboard, and to install multiple hard disks.

[0018] Wherein, the server system architecture further includes a plurality of fans, and the fans are disposed on both the front and rear sides of the upper chassis and the lower chassis.

[0019] Among them, the server system architecture also includes an upper chassis cover and a lower chassis cover, the upper chassis cover is detachably connected to the upper header box and is located on the top of the upper chassis, and the lower chassis cover is detachably connected to the lower chassis and is located on a side of the lower chassis close to the upper chassis.

[0020] The GPU module is composed of at least one of a mixed GPU combination, a full double-width GPU combination, and a full triple-width GPU combination.

[0021] A server system architecture of the present invention supports the fourth-generation AMD EPYC (Xiaolong) 9004 series processors, with up to 96 cores per chip, and also supports Intel Whitley and Eagle Stream, is compatible with multiple platforms, and has the advantages of high-performance computing, storage, high reliability, flexible configuration, and easy management; supports 12 RTX double-width gaming graphics cards, each card communication reaches PCIe5.0x8, and makes full use of the PCIe5.0 high-speed bus protocol to achieve high-speed data transmission. The high bandwidth of PCIe5.0x8 ensures the data interaction efficiency between the GPU and other components. This unique GPU configuration and parallel computing method can play a huge role in scenarios such as artificial intelligence training; it can also support up to 10 triple-width GPU cards; or 6 double-width + 6 triple-width GPU cards. It supports up to 12 PCS fans, with good heat dissipation performance, to ensure stable performance. It supports 24 DDR54800REG memories, 12 channels, and a maximum memory capacity of 6TB. The high frequency and large capacity support of DDR5 memory ensure the efficient operation of the server when processing large-scale data. It has five 3.5-inch hot-swappable hard disk slots and supports SATA and U.2 interfaces, allowing users to flexibly configure storage devices according to their needs to meet the storage capacity and read / write speed requirements of different application scenarios. The modular and tool-free structural design of the 9U12GPU is designed to bring customers a comfortable and convenient operation and maintenance experience. It is particularly worth mentioning that the 9U12GPU separates the host node and the GPU node through an upper and lower layer allocation method, and creates multiple modules for key components such as fans, power supplies, network cards, etc., allowing customers to maintain specific components in an easier way, improving server performance and cost optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 It is a schematic diagram of the overall structure of the server system architecture of the present invention.

[0024] Figure 2 It is a front view layout diagram of the server system architecture of the present invention.

[0025] Figure 3 It is a rear view layout diagram of the server system architecture of the present invention.

[0026] Figure 4 It is an exploded view of the server system architecture of the present invention.

[0027] Figure 5 It is a structural schematic diagram of the mainboard of the present invention.

[0028] Figure 6 It is a schematic diagram of the installation of the power supply of the present invention.

[0029] Figure 7 It is a schematic diagram of installing the GPU module of the present invention.

[0030] Figure 8 It is a schematic diagram of the structure inside the upper chassis of the present invention.

[0031] Fig. 9 It is a schematic diagram of the installation of the fan of the present invention.

[0032] Fig.10 It is a schematic diagram showing signal lines of the control system of the present invention.

[0033] Fig.11 It is a schematic diagram of the power line of the control system of the present invention.

[0034] Fig.12 It is a schematic diagram of the hybrid insertion combination of 6 double-width + 6 triple-width GPU combination of the present invention.

[0035] Fig.13 Schematic diagram of a combination of 12 double-width GPUs of the present invention.

[0036] Fig.14 1 is a schematic diagram of a triple-wide GPU combination of 10 according to the present invention.

[0037] In the figure: 101-lower chassis, 102-upper chassis, 103-motherboard, 104-power adapter board, 105-PDB adapter board, 106-GPU module, 107-GPU adapter board, 108-NVME module, 109-PCle module, 110-power supply, 111-hard disk backplane, 112-fan, 113-upper chassis cover, 114-lower chassis cover. DETAILED DESCRIPTION

[0038] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0039] See also Figures 1 to 14 ,in, Figure 1 It is a schematic diagram of the overall structure of the server system architecture of the present invention. Figure 2 It is a front view layout diagram of the server system architecture of the present invention. Figure 3It is a rear view layout diagram of the server system architecture of the present invention. Figure 4 It is an exploded view of the server system architecture of the present invention. Figure 5 It is a schematic structural diagram of the mainboard 103 of the present invention. Figure 6 FIG. 1 is a schematic diagram of the installation of the power supply 110 of the present invention. Figure 7 It is a schematic diagram of the installation of the GPU module 106 of the present invention. Figure 8 It is a schematic diagram of the structure inside the upper chassis 102 of the present invention. Fig. 9 It is a schematic diagram of the installation of the fan 112 of the present invention. Fig.10 It is a schematic diagram showing signal lines of the control system of the present invention. Fig.11 It is a schematic diagram of the power line of the control system of the present invention. Fig.12 It is a schematic diagram of the hybrid insertion combination of 6 double-width + 6 triple-width GPU combination of the present invention. Fig.13 Schematic diagram of a combination of 12 double-width GPUs of the present invention. Fig.14 1 is a schematic diagram of a triple-wide GPU combination of 10 according to the present invention.

[0040] The present invention provides a server system architecture, comprising a lower chassis 101, an upper chassis 102, a mainboard 103, a power adapter board 104, a PDB adapter board 105, two GPU modules 106, two GPU adapter boards 107, an NVME module 108, a PCle module 109, a power supply 110, a hard disk backplane 111, a plurality of fans 112, an upper chassis cover 113 and a lower chassis cover 114.

[0041] According to this specific embodiment, the motherboard 103 is installed in the lower chassis 101, the power adapter board 104 is installed in the lower chassis 101, the PDB adapter board 105 is installed in the lower chassis 101, the two GPU adapter boards 107 are respectively installed in the lower chassis 101 and the upper chassis 102, and the two GPU modules 106 are respectively connected to the two GPU adapter boards 107;

[0042] The motherboard 103 is used to install the CPU and memory to form an electronic system, and has multiple PCIe5.0x16 slots and multiple PCIe5.0x8 slots;

[0043] The power adapter board 104 is used to shunt the PSU DC power to other boards inside the server;

[0044] The PDB adapter board 105 is used to connect to the GPU power interface to supply power to the GPU;

[0045] The GPU module 106 is used for computing and graphics processing;

[0046] The GPU adapter board 107 is used to connect the GPU and PCIe to transmit CPU signals to the GPU.

[0047] The NVME module 108 is installed on the front window of the lower chassis 101, and the PCle module 109 is inserted into the card slot of the motherboard 103;

[0048] The NVME module 108 is used for reading and writing, storing data and serving as a system disk;

[0049] The PCle module 109 is used for signal conversion and transmission, and multiple PCIe5.0 adapter cards are inserted.

[0050] Secondly, the power supply 110 is installed at the rear of the lower chassis 101, and the hard disk backplane 111 is installed behind the NVME module 108;

[0051] The power supply 110 is used to convert the external AC power into DC power;

[0052] The hard disk backplane 111 is used to supply power to the NVME module 108 and connect the signal to the motherboard 103, and to install multiple hard disks.

[0053] Meanwhile, the fans 112 are disposed on both the front and rear sides of the upper chassis 102 and the lower chassis 101 .

[0054] In addition, the upper chassis cover 113 is detachably connected to the upper header box and is located on the top of the upper chassis 102 , and the lower chassis cover 114 is detachably connected to the lower chassis 101 and is located on a side of the lower chassis 101 close to the upper chassis 102 .

[0055] Hardware assembly: Install the motherboard 103 (MLB) into the lower chassis 101, install the fourth-generation AMD EPYC (Xiaolong) 9004 series CPU on the corresponding slot of the MLB, ensure that the CPU and MLB are in good contact, and install the heat sink. Install the two PDB adapter boards 105 and the power adapter board 104 (GPU PDB), and fix them with screws. According to the design requirements, insert the five PCIe5.0 adapter cards into the PCIe5.0x16 slots of the motherboard 103 respectively to ensure that the graphics card is firmly installed. At the same time, install 24 DDR54800REG memory sticks and install them in a 12-channel configuration to ensure that the memory and MLB are stably connected. Figure 5 .

[0056] Install five 3.5-inch hard disks on the hard disk slots of the hard disk backplane 111. Select hard disks with SATA or U.2 interfaces as required, and connect data cables and power cables. Install an optional 2400W / 2700W / 3000W, 4+1 hot-swappable redundant power supply, i.e., the power supply 110, and connect the power supply lines to ensure a stable power supply. Figure 6 .

[0057] Assemble the lower GPU. First, assemble the GPU adapter board 107 that fixes the GPU, and then insert the five GPU cards into the slots of the GPU adapter board 107. Fix the GPU with the front and back screws to ensure that the graphics card is firmly installed. Figure 7 .

[0058] Assemble the upper GPU module 106. The upper layer is mainly composed of the upper chassis 102, the GPU adapter board 107, the GPU, etc., forming a separate module for easy maintenance and assembly, and connected to the lower chassis 101 by screws. The assembly method of the upper GPU is consistent with that of the lower GPU, such as Figure 8 .

[0059] The overall heat dissipation system is composed of 12 fans 112 of the 1256 standard, which are respectively distributed at the front and rear ends of the upper chassis 102 and the lower chassis 101. Fig. 9 .

[0060] The upper chassis cover 113 and the lower chassis cover 114 are assembled and fixed by screws to ensure the safety and sealing of the chassis, thereby completing the assembly relationship of the whole machine.

[0061] Control system connection instructions:

[0062] According to the design requirements, 10 GPUs are respectively inserted into the GPU adapter board 107 with a PCIe5.0x8 slot. Each GPU has a separate GPU adapter board 107, which does not affect each other, ensuring a firm installation. The motherboard 103 is designed with 6 PCIe5.0x16 slots and one PCIe5.0x8 slot (for reserved expansion and other configuration requirements). The connection between the motherboard 103 and the GPU signal requires inserting a PCIe5.0x16 board into the PCIe5.0x16 slot of the ATX motherboard, and connecting it to the signal connector on the GPU adapter board 107 through a signal line. Each PCIe board is equipped with two signal line interfaces, which can connect 2 GPUs, so the 6 PCIe5.0x16 slots of the ATX motherboard are fully sufficient to support 10 RTX graphics cards. Fig.10 .

[0063] The connection of the GPU graphics card power supply is achieved by transmitting the power adapter board 104 to the PDB adapter board 105, and then the PDB adapter board 105 is connected to the GPU adapter board 107. Fig.11 .

[0064] Software Configuration:

[0065] ①. Install the server operating system, such as Linux, and perform necessary system settings and optimizations, including kernel parameter adjustment, driver installation, etc.

[0066] ②. According to the application scenario, install corresponding software tools and libraries, such as deep learning frameworks (Tensor Flow, PyTorch, etc.), engineering analysis software, image recognition libraries, encoding and decoding software, 3D modeling software, etc.

[0067] ③. Configure the server resource management system to reasonably allocate and schedule GPU, memory, storage and other resources to meet the operating requirements of different applications.

[0068] Performance testing and optimization:

[0069] ①.Perform performance tests on the assembled servers, including GPU computing capability tests, memory read and write speed tests, storage read and write performance tests, etc.

[0070] ②. Based on the test results, optimize the server architecture, such as adjusting GPU parallel computing parameters, optimizing memory access patterns, upgrading storage devices, etc., to further improve server performance.

[0071] Finally, the GPU module 106 is composed of at least one of a GPU mixed insertion combination, a full double-width GPU combination, and a full triple-width GPU combination.

[0072] Supports mixed insertion combination of 6 double-width + 6 triple-width GPU combinations, such as Fig.12 . Supports 12 double-width GPU combinations, such as Fig.13 . Supports 10 triple-width GPU combinations, such as Fig.14 .

[0073] A server system architecture using this embodiment supports the fourth-generation AMD EPYC (Xiaolong) 9004 series processors, with up to 96 cores per chip, and also supports Intel Whitley and Eagle Stream, which is compatible with multiple platforms and has the advantages of high-performance computing, storage, high reliability, flexible configuration, and easy management; it supports 12 RTX double-width gaming graphics cards, and each card communicates up to PCIe5.0x8, making full use of the PCIe5.0 high-speed bus protocol to achieve high-speed data transmission. The high bandwidth of PCIe5.0x8 ensures the efficiency of data interaction between the GPU and other components. This unique GPU configuration and parallel computing method can play a huge role in scenarios such as artificial intelligence training; it can also support up to 10 triple-width GPU cards; or 6 double-width + 6 triple-width GPU cards. It supports up to 12 PCS fans 112, with good heat dissipation performance, to ensure stable performance. It supports 24 DDR54800REG memory, 12 channels, and a maximum memory capacity of 6TB. The high frequency and large capacity support of DDR5 memory ensure the efficient operation of the server when processing large-scale data. It has five 3.5-inch hot-swappable hard disk slots and supports SATA and U.2 interfaces, allowing users to flexibly configure storage devices according to their needs to meet the storage capacity and read / write speed requirements of different application scenarios. The modular and tool-free structural design of the 9U12GPU is designed to bring customers a comfortable and convenient operation and maintenance experience. It is particularly worth mentioning that the 9U12GPU separates the host node and the GPU node through an upper and lower layer allocation method, and creates multiple modules for key components such as fans 112, power supplies, network cards, etc., allowing customers to maintain specific components in an easier way, improving server performance and cost optimization.

[0074] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A server system architecture, comprising a lower chassis and an upper chassis, wherein the upper chassis is located above the lower chassis, characterized in that: It also includes a mainboard, a power adapter board, a PDB adapter board, two GPU modules and two GPU adapter boards, wherein the mainboard is installed in the lower chassis, the power adapter board is installed in the lower chassis, the PDB adapter board is installed in the lower chassis, the two GPU adapter boards are installed in the lower chassis and the upper chassis respectively, and the two GPU modules are connected to the two GPU adapter boards respectively; The motherboard is used to install the CPU and memory to form an electronic system, and has multiple PCIe5.0x16 slots and multiple PCIe5.0x8 slots; The power adapter board is used to shunt the PSU DC power to other boards inside the server; The PDB adapter board is used to connect to the GPU power interface to supply power to the GPU; The GPU module is used for computing and graphics processing; The GPU adapter board is used to connect the GPU and PCIe to transmit CPU signals to the GPU.

2. A server system architecture as claimed in claim 1, characterized in that: The server system architecture further includes an NVME module and a PCle module, wherein the NVME module is installed on the front window of the lower chassis, and the PCle module is inserted into the card slot of the motherboard; The NVME module is used for reading and writing, storing data and serving as a system disk; The PCle module is used for signal conversion and transmission, and multiple PCIe5.0 adapter cards are inserted.

3. A server system architecture as claimed in claim 2, characterized in that: The server system architecture further includes a power supply and a hard disk backplane, wherein the power supply is installed at the rear of the lower chassis, and the hard disk backplane is installed behind the NVME module; The power supply is used to convert the external AC power into DC power; The hard disk backplane is used to supply power to the NVME module and connect the signal to the motherboard, and to install multiple hard disks.

4. A server system architecture as claimed in claim 1, characterized in that: The server system architecture further includes a plurality of fans, and the fans are disposed on both the front and rear sides of the upper chassis and the lower chassis.

5. A server system architecture as claimed in claim 1, characterized in that: The server system architecture also includes an upper chassis cover and a lower chassis cover, wherein the upper chassis cover is detachably connected to the upper header box and is located on the top of the upper chassis, and the lower chassis cover is detachably connected to the lower chassis and is located on a side of the lower chassis close to the upper chassis.

6. A server system architecture as claimed in claim 1, characterized in that: The GPU module is composed of at least one of a GPU mixed insertion combination, a full double-width GPU combination, and a full triple-width GPU combination.

Images