Server architecture system

Abstract

The invention relates to the technical field of servers, in particular to a server architecture system which comprises a case, a lower-layer module, a middle-layer module and an upper-layer module, and the middle-layer module comprises an HDD group, an HDD TRAY, a wind scooper, a straight-through plate group, a switch group, an M.2 group and a wire arrangement frame baffle; the middle layer module comprises eight 2.5 HDD groups and is fixed on an HDD TRAY disk through I-shaped nails and thumb screws in a single module mode, even-number HDD configuration can be realized by interchanging the HDD groups with HDD dummy when less than eight HDD are used, odd-number HDD configuration can be realized by interchanging the single 2.5 HDD with 2.5 HDD driver dummy, so that the cost is saved on the premise of meeting the user requirements, and the PCIe Gen 5 switch group is matched, so that the use efficiency is improved, and the service life of the HDD TRAY disk is prolonged. The method supports single and double wide GPU cards, adapts to the newest GPU of NVIDIA and domestic mainstream computing card schemes, ensures high stability and reliability of the system, meets service requirements of different users, and solves the problem that an existing server double-node architecture cannot save cost under the condition that function use cannot be met.

Description

Technical Field

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

[0002] The digital era is accelerating, and artificial intelligence has become a new productivity that promotes the transformation and upgrading of all industries. In particular, the continuous breakthroughs in general artificial intelligence technology have greatly expanded the application scenarios and scope of "artificial intelligence +", truly entering a new stage of "empowering new quality with artificial intelligence".

[0003] In order to meet the rapidly growing demand for computing infrastructure and innovative applications in the "AI+" era, the 5U12GPU is launched, which is specially designed for the needs of AI and high-performance computing HPC fields. It has extremely high flexibility and comprehensive and thoughtful design, which can meet the high-performance computing needs of different scales and provide faster, more stable and more flexible computing services. This server is based on the latest processor platform of Intel EGS (Eagle Stream) and has excellent computing performance; the GPU supports two connection methods, PCI-E direct connection or Switch, which can be selected according to the actual application scenario, adapt to various workloads, and give full play to the performance of the GPU; it supports up to 12 active or passive cooling GPUs, up to 450W. In addition, the server supports DDR5 and PCI-e 5.0 technology, providing powerful expansion capabilities for various high-performance applications.

[0004] Since the number of nodes placed in conventional cabinets on the market is even, the server architecture on the market is a dual-node architecture such as 2 / 4 / 6 / 8U. There is no single-node 5U architecture that is compatible with 12GPUs, which cannot save costs while meeting functional requirements. Summary of the invention

[0005] The purpose of the present invention is to provide a server architecture system, aiming to solve the problem of cost saving when the existing server dual-node architecture cannot meet the functional usage.

[0006] To achieve the above-mentioned objectives, the present invention provides a server architecture system, comprising a chassis, a lower module, a middle module and an upper module, wherein the lower module is arranged at the bottom of the chassis, the middle module is arranged at the top of the lower module, and the upper module is arranged at the top of the middle module, and the middle module comprises an HDD group, an HDD TRAY, an air guide cover, a pass-through plate group, a switch group, an M.2 group and a cable management rack baffle; the HDD group is connected to the HDD TRAY, the air guide cover is fixed to the chassis, the pass-through plate group and the switch group are interchangeable and are located on one side of the chassis, the M.2 group is connected to the chassis, and the cable management rack baffle is connected to the chassis.

[0007] The middle-layer module further includes a HDD dummy, and the HDD dummy can be interchanged with the HDD group to achieve an even number of HDD configurations.

[0008] The lower module includes a power supply module, an IO board, a COM interface and a heat dissipation module, and the power supply module, the IO board, the COM interface and the heat dissipation module are respectively connected to the MLB.

[0009] Wherein, the heat dissipation module includes a hot-swappable fan group, an aluminum radiator and a liquid cooling plate. The hot-swappable fan group is connected to the chassis and is located on one side of the chassis. The aluminum radiator and the liquid cooling plate are respectively connected to the MLB.

[0010] Wherein, the upper module includes a front GPU group, a front GPU bracket, an upper fan group, a rear GPU bracket, a rear GPU group and a rear GPU pressure plate group, the front GPU bracket is connected to the rack, the front GPU group is connected to the front GPU bracket, the upper fan group is connected to the chassis and is located at an end of the front GPU bracket away from the front GPU group, the rear GPU bracket is connected to the chassis and is located at an end of the upper fan group away from the front GPU bracket, the rear GPU group is fixedly connected to the rear GPU bracket and is located at an end of the rear GPU bracket away from the upper fan group, and the rear GPU pressure plate group is arranged on one side of the rear GPU module.

[0011] The server architecture system of the present invention, the chassis provides support and installation conditions for the lower module, the middle module and the upper module, the lower module provides power, various required signals and heat dissipation for the entire chassis, and the middle module and the upper module release the performance required by the entire server. The middle module includes 8 2.5 HDD groups, which are fixed to the HDD TRAY disk by I-pins and thumb screws in a single module, and when less than 8 HDDs are used, the HDD group can be interchanged with the HDD dummy to achieve an even number of HDD configurations, and a single 2.5 HDD can be interchanged with a 2.5 HDD carrier dummy to achieve an odd number of HDD configurations, and it is equipped with a PCIe Gen 5 switch, supports single and double width GPU cards, and is compatible with NVIDIA's latest GPU and domestic mainstream computing card solutions, ensuring the high stability and reliability of the system and meeting the business needs of different users. In this way, costs can be saved while meeting user needs. The air guide cover is fixed to the corresponding I-nail on the chassis through the positioning groove thereon. The GPU pass-through board of the rear window GPU is locked on the pass-through board TRAY through 9 screws to form the pass-through board group. The position of the pass-through board group can also be used to install the switch group. The M.2PCA card is fixed to the corresponding I-nail and stud on the M.2 bracket through 2 gourd holes and 1 screw expansion hole. This board supports M.2 cards with a length of 22110 or 2280 to form the M.2 group. The cable management rack baffle is initially fixed by 1 U-shaped groove and 1 positioning PIN on both sides and 1 I-nail and 1 positioning hole on both sides of the chassis, and then the cable management rack baffle is limited to rotate up and down by the limited space distance between the HDD TRAY disk in front of the cable management rack baffle and the M.2 bracket at the back to achieve precise fixation, thereby solving the problem of cost saving when the existing server dual-node architecture cannot meet the functional usage. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] 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.

[0013] Figure 1 It is an exploded diagram of a lower layer module of a server architecture system provided by the present invention.

[0014] Figure 2 This is a top view of the lower module (with aluminum radiator installed).

[0015] Figure 3 This is a top view of the lower module (liquid cooling plate).

[0016] Figure 4 This is a schematic diagram of the lower fan PCA module.

[0017] Figure 5 This is the schematic diagram of the MLB module.

[0018] Figure 6 It is the PDB-BRIDGE BOARD-MLB assembly relationship diagram.

[0019] Figure 7 This is the installation diagram of the OCP bracket, OCP card and OCP cable head.

[0020] Figure 8 This is the installation diagram of the IO board, COM interface and IO-OCP cage.

[0021] Fig. 9 This is a schematic diagram of the IO-OCP module.

[0022] Fig.10 This is a top view of the lower module layout.

[0023] Fig.11 It is a side view of the lower module layout.

[0024] Fig.12 This is an exploded view of the middle-level module.

[0025] Fig.13 This is an exploded view of the GPU pass-through board module.

[0026] Fig.14 This is the exploded diagram of the switch group.

[0027] Fig.15 This is an exploded view of the GPU M.2 group.

[0028] Fig.16 This is the layout diagram of the middle layer high-speed signal cable.

[0029] Fig.17 This is the layout diagram of the middle layer power supply cable.

[0030] Fig.18 This is the layout diagram of the middle layer Sideband Signal cables.

[0031] Fig.19 This is an exploded view of the upper module.

[0032] Fig. 20 It is a side view of the upper module.

[0033] Fig.21 This is the layout diagram of the upper power supply cables.

[0034] Fig. 22 This is the layout diagram of the upper-layer PCIe and Sideband signals.

[0035] Fig.23 It is a box ear wiring diagram layout diagram.

[0036] Fig.24 This is the side view of the 5U12 GPU.

[0037] Fig.25 This is the MLB circuit connection diagram.

[0038] Fig.26 It is a schematic diagram of MLB.

[0039] In the figure: 1-chassis, 2-PSU, 3-PDB, 4-BRIDGE BOARD, 5-MLB, 6-aluminum radiator, 7-Lower FCBPCA, 8-hot-swappable fan assembly, 9-IO board, 10-COM interface, 11-liquid cooling plate, 12-PSU cage, 13-IO-OCP cage, 14-MLB tray, 15-lower fan PCA TRAY, 16-lower FCB backplane module, 17-lower fan cage, 18-OCP tray, 19-OCP card, 20-OCP cable head, 21-IO-OCP module, 22-MLB tray module, 23-HDD group, 24-upper cover, 25-HDDtray, 26-HDD dummy, 27-air guide cover, 28-GPU pass-through board, 29-pass-through board tray, 30-pass-through board group, 31-switch group, 32-switch PCA, 33-switch tray, 34-M.2 bracket, 35-M.2 PCA, 36-M.2 group, 37-cable management bracket baffle, 38-NVME HDD PCA, 39-front GPU bracket, 40-front GPU group, 41-upper fan group, 42-rear GPU bracket, 43-rear GPU group, 44-rear GPU pressure plate group, 45-PCIe riser, 46-GPU card, 47-upper FCB PCA, 48-left box ear module, 49-right box ear module, 50-left box ear PCA, 51-right box ear PCA. DETAILED DESCRIPTION

[0040] 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.

[0041] See also Figures 1 to 26 The present invention provides a server architecture system, comprising a chassis 1, a lower module, a middle module and an upper module, wherein the lower module is arranged at the bottom of the chassis 1, the middle module is arranged at the top of the lower module, and the upper module is arranged at the top of the middle module, and the middle module comprises an HDD group 23, an HDD tray 25, an air guide cover 27, a pass-through board group 30, a switch group 31, an M.2 group 36 and a cable management bracket baffle 37;

[0042] The HDD group 23 is connected to the HDD tray 25, the air guide cover 27 is fixed to the chassis 1, the pass-through board group 30 and the switch group 31 are interchangeable and are located on one side of the chassis 1, the M.2 group 36 is connected to the chassis 1, and the cable management bracket 37 is connected to the chassis 1.

[0043] In the embodiment of the present invention, the chassis 1 provides support and installation conditions for the lower module, the middle module and the upper module. The lower module provides power, various required signals and heat dissipation for the entire chassis. The middle module and the upper module release the performance required by the entire server. The middle module includes 8 2.5 HDD groups 23, which are fixed to the HDD tray 25 by I-pins and thumb screws in a single module. When less than 8 HDDs are used, the HDD group 23 can be interchanged with the HDD dummy 26 to achieve an even number of HDD configurations, and a single 2.5 HDD can be interchanged with a 2.5 HDD carrier dummy to achieve an odd number of HDD configurations. The PCIe Gen 5 switch is used to support single and double width GPU cards, and is compatible with NVIDIA's latest GPU and domestic mainstream computing card solutions to ensure high stability and reliability of the system and meet the business needs of different users. This saves costs while meeting user needs.

[0044] The air guide cover 27 is fixed to the corresponding I-shaped nails on the chassis 1 through the positioning grooves thereon. The design of the air guide cover 27 achieves screw-free locking by making the positioning grooves inverted by utilizing the plastic elastic deformation of plastic. The GPU pass-through board 28 of the rear window GPU is locked on the pass-through board tray 29 by 9 screws to form the pass-through board group 30. Then, the pass-through board group 30 is placed above the PSU cage 12 of the rear window of the chassis 1, and is positioned through the 4 gourd holes and 2 U-shaped grooves on the pass-through board tray 29 and the 6 I-shaped nails at the corresponding positions on the PSU cage 12, and then fixed through the 2 thumb screws on the pass-through board tray 29 and the corresponding studs on the PSU cage 12. At the same time, the switch group 31 can also be installed at this position. Its assembly method is: the switch PCA 32 is fixed to the switch tray 33 by 13 screws to form the switch group 31, and the 4 gourd holes and 2 U-shaped grooves on the switch tray 33 are positioned with the 6 I-shaped nails at the corresponding positions on the PSU cage 12, and then the thumb screws on the switch tray 33 are fixed with the corresponding studs on the PSU cage 12 to achieve the purpose of fixing;

[0045] The M.2 PCA card 35 is fixed to the corresponding I-nail and stud on the M.2 bracket 34 through two gourd holes and one screw expansion hole. This board supports M.2 cards with a length of 22110 or 2280, forming the M.2 group 36. The M.2 group 36 is fixed to the two I-nail and one positioning hole on both sides of the chassis 1 through two U-shaped grooves and one positioning PIN on both sides of the bracket;

[0046] The cable management rack baffle 37 is initially fixed by positioning with an I-nail and a positioning hole on both sides of the chassis 1 through a U-shaped groove and a positioning PIN on both sides, and then the cable management rack baffle 37 is precisely fixed by limiting the up and down rotation of the cable management rack baffle 37 through the limited space distance between the HDD tray 25 in front of the cable management rack baffle 37 and the M.2 bracket 34 at the back. In particular, a threaded hole is made in the middle of each side of the baffle as the front window of the chassis 1. When the HDD tray 25 is not installed, the cable management rack baffle 37 is fixed by enlarging the corresponding screw holes on both sides of the chassis 1 and locking the screws;

[0047] All nodes of the middle layer group are independently modularized for easy loading and unloading. The power supply and signal passing of the NVME HDDPCA 38, M.2PCA 35, GPU pass-through board 28 and switch PCA 32 in the middle layer are mainly transmitted by MLB5 (motherboard) to the backplane of each module through cables. The 4 x8 PCIe signals in MLB5 are connected to the MCIO connectors on the 4 NVMe HDD PCA38 through cables to provide NVMe HDD PCA38 with PCIe signals of NVMe protocol. The 10 x8PCIe signals are connected to the 10 MCIO connectors of the GPU pass-through board 28 through cables to provide PCIe signals for the GPU pass-through board 28. The 1 x8 PCIe signal is connected to the x8 Slimline connector on the M.2PCA 35 through cables to provide Slimline signals for the M.2PCA 35. The power required by the four NVMe HDD PCA38 is divided into four by MLB5 through cables and connected to the power connector on the NVMe HDD PCA38 to supply power to the four NVMe HDD PCA38. The power required by the GPU pass-through board 28 is provided by three cables on PDB3 connected to the power connector of the GPU pass-through board 28 to supply power to the GPU pass-through board 28. The Sideband Signal required by the four NVMe HDD PCA 38 is provided by a Sideband cable of MLB5 divided into four and connected to the Sideband Signal connectors of the four NVME HDD PCA 38. The two types of Sideband Signal required by the GPU pass-through board 28 are provided by connecting a Sideband cable of MLB5 and a Sideband cable of PDB3 to two different Sideband Signal connectors of the GPU pass-through board 28 respectively. At the same time, a Sideband cable of the GPU pass-through board 28 is connected to the Sideband Signal connector of the Lower FCB PCA 7 to meet the requirements of the Lower FCB PCA7 Sideband Signal.

[0048] Furthermore, the lower module includes a power supply module, an IO board 9, a COM interface 10 and a heat dissipation module, and the power supply module, the IO board 9, the COM interface 10 and the heat dissipation module are connected to MLB5.

[0049] In the embodiment of the present invention, the power supply module is inserted into PSU2 by an external power supply and then PSU2 is inserted into the connection on PDB3, and then the gold finger at one end of BRIDGE BOARD4 is inserted into the connector on PDB3, and the gold finger at the other end is inserted into the connector on MLB5 to supply power to MLB5 through the external power supply, and MLB5 is connected to FCB PCA 7 in the lower fan module through a cable.

[0050] The x4 Slimline signal in MLB5 is connected to the IO board 9 via a cable to provide a signal thereto, and the 2 Sideband Signals in MLB5 are connected to the IO board 9 via a cable, one of which is split into two and the other is connected to the COM interface 10 .

[0051] MLB5 is locked on MLB tray 14 by 8 I-pins and 2 thumb screws, and then 2 cable management clips 18 are respectively locked on the left and right ends of MLB5 by 2 screws, forming the entire MLB tray module 22. MLB tray module 22 is equipped with cable management clips 18 to achieve a regular layout of the cables inside the chassis 1, so that the interior of the chassis 1 is simple and beautiful. Then the entire MLB tray module 22 is first initially positioned by the two left and right slides of MLB tray 14 and the two positioning pins at the corresponding positions on the chassis 1, and then the MLB tray 14 is slid. The gourd holes on the MLB tray 14 are precisely positioned with the 10 I-pins at the corresponding positions of the chassis 1, and then the MLB tray module 22 is fixed on the chassis 1 by the two thumb screws on the MLB tray 14. In this way, independent modularization is achieved, which is convenient for loading and unloading.

[0052] PDB3 is directly fixed to the chassis 1 by 3 I-pins and 4 screws, and BRIDGE BOARD4 is directly fixed to the studs of PDB3 by 2 screws. BRIDGE BOARD4 is directly fixed to PDB3, and the gold fingers of BRIDGE BOARD4 are used to connect MLB5 and PDB3, so that the assembly of BRIDGE BOARD4 does not occupy extra space in the chassis 1, and at the same time reduces the volume of BRIDGE BOARD4 and the use of connectors and wires, achieving the advantage of saving material, manpower and financial resources in the case of modular installation.

[0053] PSU2 is installed in the PSU cage12 module, wherein PSU cage12 is riveted to the chassis 1. The COM interface is directly fixed to the IO-OCP cage13 by two screws, and the IO board 9 is directly fixed to the IO-OCP cage13 by two I-pins and two screws. Position the OCP tray through the L-shaped card slot on the OCP tray18 and the two I-pins on each side of the IO-OCP cage13, and then fix the OCP tray18 to the IO-OCP cage13 by one screw on each side. Fix the OCP cable head 20 to the OCP tray18 by two screws, and then insert the OCP card 19 into the OCP cable head 20 on the OCP tray18 through the guide rail on the OCP tray18, and then rotate the thumb screw on the OCP card 19 and the screw hole of the IO-OCP cage13 to fix it. Finally, the IO-OCP module 21 is locked on the chassis 1 by a screw and an I-nail corresponding to the IO-OCP cage 13U slot on the chassis 1. In this way, independent modularization is achieved, which is convenient for loading and unloading.

[0054] The OCP cable head 20 consists of two PCIe signal cables, one power cable and one Sideband Signal cable, which are respectively connected to two X8 MCIO connectors, a power connector and a Sideband Signal connector of MLB5 so that the motherboard provides PCIe signals, power and Sideband Signal for the OCP card 19.

[0055] Further, the heat dissipation module includes a hot-swappable fan group 8, an aluminum radiator 6 and a liquid cooling plate 11, wherein the hot-swappable fan group 8 is connected to the chassis 1 and is located on one side of the chassis 1. The aluminum radiator 6 and the liquid cooling plate 11 are respectively connected to the MLB 5 and cannot exist at the same time.

[0056] In the embodiment of the present invention, the heat dissipation module includes the hot-swappable fan group 8 supporting five 8056s on the front window, and the front window fan module can be maintained without powering off the server. The CPU of MLB5 is equipped with the aluminum heat sink 6 with a height of 2U and copper tubes and remote heat dissipation fins. In addition to aluminum heat dissipation, the liquid cooling plate 11 can also be installed on the CPU of MLB5 to achieve liquid cooling, so as to provide users with a variety of heat dissipation options.

[0057] The Lower FCB PCA7 is fixed to the lower fan TRAY15 by 5 I-pins and 2 screws to form a lower FCB backplane module 16. This module is installed on the chassis 1 through the positioning pins at the corresponding positions of the chassis 1, and then fixed to the chassis 1 by screws. The front window fan cage17 is fixed to the chassis 1 by riveting, and then the 5 independent 8056 hot-swappable fan groups 8 are inserted into the front window fan cage17 and connected to the connector on the lower FCB7. The 5 independent 8056 hot-swappable fan groups 8 are quick-swappable modules without screw locking, and support hot-swappable. They can be plugged in and out without powering off the server, which is convenient for operation and can form independent modularization for easy loading and unloading.

[0058] Furthermore, the upper module includes a front GPU group 40, a front GPU bracket 39, an upper fan group 41, a rear GPU bracket 42, a rear GPU group 43 and a rear GPU pressure plate group 44, the front GPU bracket 39 is connected to the chassis 1, the front GPU group 40 is connected to the front GPU bracket 39, the upper fan group 41 is connected to the chassis 1 and is located at the end of the front GPU bracket 39 away from the front GPU group 40, the rear GPU bracket 42 is connected to the chassis 1 and is located at the end of the upper fan group 41 away from the front GPU bracket 39, the rear GPU group 43 is fixedly connected to the rear GPU bracket 42 and is located at the end of the rear GPU bracket 42 away from the upper fan group 41, and the rear GPU pressure plate group 44 is arranged on one side of the rear GPU group 43.

[0059] In the embodiment of the present invention, the assembly relationship of the upper module is as follows: the front GPU bracket 39 is placed from top to bottom, and the U-shaped positioning grooves on the left and right sides thereof are first assembled and positioned with the I-shaped nails at the corresponding positions of the chassis 1, and then the thumb screws on the left and right sides are rotated to lock with the studs at the corresponding positions on the chassis 1 to fix the front GPU bracket 39 on the chassis 1; two front GPU groups 40 are placed from top to bottom, one end is fixed to the front GPU bracket 39 through the I-shaped nails and studs on the front GPU bracket 39 and the gourd hole and thumb screw on one side of the front GPU group 40, and the other end is fixed to the front GPU bracket 39 through the HDD The I-shaped nails and studs on tray25 are fixed to HDDtray25 with the U-shaped grooves and thumb screws on the front GPU bracket 39; the upper fan group 41 is placed from top to bottom, and the U-shaped grooves on the lower left and right sides of the upper fan group 41 are first positioned with the corresponding I-shaped nails on the chassis 1, and then two thumb screws on the left and right sides of the fan module are used to lock and fix them with the corresponding studs on the chassis 1; the rear GPU bracket 42 is placed from top to bottom, and the U-shaped grooves on the lower left and right sides of the rear GPU bracket 42 are first positioned with the corresponding I-shaped nails on the chassis 1, and then one thumb screw on the left and right sides of the rear GPU bracket 42 is used to lock and fix it with the corresponding studs on the chassis 1; the 10 rear window GPU groups are respectively inserted from left to right into the X16PCIe slot, and then one end is used to fix the rear window GPU group to the rear GPU bracket 42 through the thumb screws on the rear window GPU module, and the other end is used to fix the rear window GPU group by locking the gourd holes and thumb screws on the rear window GPU pressure plate group with the I-nail and stud at the corresponding positions on the chassis 1.

[0060] Since the DIMM on our MLB5 requires irregular maintenance, and the position of the DIMM in the chassis 1 is just below the upper fan group 41, when we maintain the DIMM, we only need to disassemble the upper fan group 41 as a whole to maintain the DIMM on the motherboard, without having to disassemble the entire upper configuration module, thereby reducing maintenance costs. At the same time, since the entire server has two sets of upper and lower fan modules and they are independently detachable modules, when only one set of fan modules is needed to meet the heat dissipation requirements, one set of fan modules can be quickly removed to achieve the purpose of saving costs while meeting the functions. And since the front and rear GPU modules are independent, our machine can accommodate up to 12 GPU modules, and you can arbitrarily match the number of GPUs you want within 12 GPUs according to user needs.

[0061] The upper front window PCIe riser 45 is powered on by MLB5 transmitting power to the front window PCIe riser 45 through cables, and then the front window PCIe riser 45 transmits power to the GPU through cables. The upper FCB PCA 47 is powered on by PDB3 transmitting power to the upper FCB PCA 47 through cables. The rear window GPU group is powered on by PDB3 transmitting power to the rear window GPU card 46 through cables.

[0062] The two X8 PCIe signals on the MLB5 are transmitted to the two upper front window PCIe risers 45 through two cables, and then the GPU card is directly inserted into the X16 PCIe slot on the PCIe riser 45 to provide PCIe signals for the GPU card. The PCIe signal required by the rear GPU group 43 card is directly connected by the GPU card inserted into the X16 PCIe slot of the GPU pass-through board 28. The Sideband signal required by the upper FCB PCA 47 is transmitted by the GPU pass-through board 28 to the upper FCB PCA 47 through a cable.

[0063] The left ear module 48 on the front window of the chassis 1 supports VGA, and the right ear module 49 supports USB, Power, UID Button and health indicator light. The left and right ear modules 49 on the front window are fixed to the chassis 1 by three screws. The signal required by the left ear module 48 is transmitted by MLB5 through a cable to the left ear PCA 50 to provide a VGA signal. The signal required by the right ear module 49 is welded to the right ear PCA 51 and then connected to MLB5, transmitting the Slimline signal of MLB5 to the right ear PCA 51.

[0064] Finally, assemble the upper cover 24 from top to bottom and then slide it forward until it stops, then press the lock of the upper cover 24 downward, turn the screwdriver, and fix the upper cover 24 and the chassis 1 to form a complete machine.

[0065] The server architecture system is based on Intel's 4th and 5th generation Xeon processors, with excellent computing performance, TDP 385W; DDR5 technology + PCIe Gen5 latest technology support; PCIe pass-through configuration with low latency to create ultimate performance; front and rear IO maintenance compatible design, with wide applicability, meet the high-performance requirements of multi-scenario applications, support up to 12 high-performance GPU cards (including NVIDIA, AMD, INTEL and other GPUs, and deep adaptation with domestic mainstream computing cards); CPU-Switch can choose X32 or X16lane interconnection to achieve high-bandwidth communication; complete complex data processing and model training / inference tasks in the shortest time. In addition, 5U12GPU also provides different cooling solutions, including air cooling and liquid cooling solutions and a variety of power solutions (2+2 or 3+1 redundancy) to achieve the most efficient performance, with PCIe Gen5switch, support single and double-width GPU cards, adapt to NVIDIA's latest GPU and domestic mainstream computing card solutions, ensure the high stability and reliability of the system, and meet the business needs of different users. The tray, modular and tool-free structure design of 5U12GPU is designed to bring users a comfortable and convenient operation and maintenance experience. It is particularly worth mentioning that 5U12GPU separates the host node and GPU node through different trays, and creates multiple modules for key components (such as fans, power supplies, network cards, etc.), allowing users to maintain specific components in an easier way.

[0066] What is disclosed above is only a preferred embodiment of a server architecture system of the present invention. It is certainly not 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 implementing the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. A server architecture system, characterized in that: It includes a chassis, a lower module, a middle module and an upper module, wherein the lower module is arranged at the bottom of the chassis, the middle module is arranged at the top of the lower module, and the upper module is arranged at the top of the middle module, and the middle module includes an HDD group, an HDD TRAY, an air guide cover, a pass-through board group, a switch group, an M.2 group and a cable management bracket baffle; The HDD group is connected to the HDD TRAY, the air guide cover is fixed to the chassis, the pass-through board group and the switch group can be used interchangeably and are located on one side of the chassis, the M.2 group is connected to the chassis, and the cable management bracket is connected to the chassis.

2. The server architecture system according to claim 1, wherein: The middle layer module further includes a HDD dummy, and the HDD dummy is interchangeable with the HDD group to achieve an even number of HDD configurations.

3. The server architecture system according to claim 1, wherein: The lower layer module includes a power supply module, an IO board, a COM interface and a heat dissipation module, and the power supply module, the IO board, the COM interface and the heat dissipation module are respectively connected to the MLB.

4. The server architecture system according to claim 2, wherein: The heat dissipation module includes a hot-swappable fan group, an aluminum radiator and a liquid cooling plate. The hot-swappable fan group is connected to the chassis and is located on one side of the chassis. The aluminum radiator and the liquid cooling plate are respectively connected to the MLB.

5. The server architecture system according to claim 1, wherein: The upper module includes a front GPU group, a front GPU bracket, an upper fan group, a rear GPU bracket, a rear GPU group and a rear GPU pressure plate group. The front GPU bracket is connected to the chassis, the front GPU group is connected to the front GPU bracket, the upper fan group is connected to the chassis and is located at an end of the front GPU bracket away from the front GPU group, the rear GPU bracket is connected to the chassis and is located at an end of the upper fan group away from the front GPU bracket, the rear GPU group is fixedly connected to the rear GPU bracket and is located at an end of the rear GPU bracket away from the upper fan group, and the rear GPU pressure plate group is arranged on one side of the rear GPU module.

Images