CN109710557B

CN109710557B - Server structure supporting 16GPU

Info

Publication number: CN109710557B
Application number: CN201811531884.8A
Authority: CN
Inventors: 吕振山; 蔡积淼
Original assignee: Zhengzhou Yunhai Information Technology Co Ltd
Current assignee: Zhengzhou Yunhai Information Technology Co Ltd
Priority date: 2018-12-14
Filing date: 2018-12-14
Publication date: 2021-11-26
Anticipated expiration: 2038-12-14
Also published as: CN109710557A

Abstract

A server structure supporting 16GPU comprises 8U cases, 4U servers, bridging modules, bolts and screws, wherein each 8U case is provided with a front window structure and a rear window structure and comprises a base, an upper cover, a middle partition plate, a rear window perforated plate and a rear window bracket, the invention designs a 4U server structure capable of accommodating eight GPUs, and simultaneously designs an 8U case capable of storing two 4U servers, namely a server structure supporting 16 GPUs, the GPUs of the two 4U servers are connected through the bridging modules at high-density ports, so that the two 4U servers and sixteen GPUs work cooperatively to meet the space requirement of a novel server, meanwhile, the special structural design of an air guide cover improves the heat dissipation performance of the server, ensures that the servers operate cooperatively in a stable and high-performance mode, the rear window perforated plate is connected with the base through the bolts on a rear window, the connection between the upper cover and the base is strengthened, the heat dissipation of the server can be guaranteed, and meanwhile, technicians can conveniently observe whether the server has abnormal conditions or not on the rear window.

Description

Server structure supporting 16GPU

Technical Field

The invention relates to the technical field of servers, in particular to a server structure supporting a 16 GPU.

Background

With the rapid development of AI (artificial intelligence), the application of artificial intelligence technologies such as image recognition and voice recognition is widely popularized, and the structure and performance of a GPU server as a core computing power providing device face significant examination.

In the prior art, a GPU server mainly comprises 4GPU and 8GPU, a chassis adopts 2U and 4U structural design, an HGX-2 ultra-strong computing platform pushed out by NVIDIA adopts sixteen Tesla V100 GPUs, and twelve NVSwitch are used to enable each GPU to communicate with other GPUs at the speed of 2.4TB per second.

Disclosure of Invention

In order to overcome the defects in the prior art, the invention provides a server structure supporting 16 GPUs; the invention designs an 8U case capable of storing two 4U servers, and connects the GPU high-density ports of the two 4U servers through a bridge module, so that the two 4U servers and sixteen GPUs work cooperatively to meet the requirement of a novel server on space; the power assisting mechanism ensures the fastening connection between the PCIE module and the GPU, reduces the probability of server failure caused by insufficient connection between high-density ports, and greatly improves the stability of the server.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a server structure supporting a 16GPU comprises an 8U case, a 4U server and a bridging module, wherein the 8U case comprises a base, an upper cover, a middle partition plate, a rear window perforated plate and a rear window support, the upper cover is arranged above the base and forms a four-sided closed structure with the base, the front end and the rear end of the four-sided closed structure are open, the front window is arranged at the front end of the base, the rear window is arranged at the rear end of the base, a slide rail is respectively arranged at the middle positions of the inner walls of the left side plate and the right side plate of the base, the planes of the two slide rails are parallel to the bottom surface of the base and the upper cover, the middle partition plate is arranged between the two side plates of the base along the direction of the slide rail, a support mounting seat is arranged at the rear window at the bottom surface of the base, the rear window perforated plate is arranged at the upper part of the rear window of the 8U case and is connected with the upper cover and the left side plate and the right side plate of the base, one end of the rear window support is connected with the rear window perforated plate, and the support mounting seat of the base, the 4U server comprises a 4U server base, a PCIE module, a GPU board and an air guide cover, the GPU board is installed in the 4U server base, the air guide cover is installed on the upper portion of the 4U server base, the PCIE module is connected with the GPU board and installed at the front end of the 4U server base, one 4U server is installed in a space between the middle partition plate and the upper cover, one 4U server is installed in a space between the middle partition plate and the bottom face of the base, and the two 4U servers are connected through the bridge module at the rear window.

Preferably, the bottom plate and the left and right side plates of the base are of an integrated structure, and the rear window support is of a long strip structure.

Preferably, the inner wall of the bottom surface of the base is provided with a plurality of anti-skidding bulges, the inner walls of the left side plate and the right side plate at one end of the base, which is close to the rear window, are respectively provided with an upper limiting nail and a lower limiting nail, and the middle partition plate is of a rectangular thin-wall structure.

Preferably, 4U server base's the left and right sides board front end is equipped with a plurality of screw thread mounting holes, and wherein the inner wall bottom of left side board front end is equipped with the rotation fulcrum, and rotation fulcrum protrusion in side inner wall plane, the rear end outer wall of the board of the 4U server base left and right sides all is equipped with spacing hole, and the 4U server is installed at 8U quick-witted incasement fixed positions with the spacing nail cooperation on spacing hole and the base, the middle part of the board inner wall of the 4U server base left and right sides is equipped with a guide rail respectively, and the rear end of 4U server base is equipped with Power clip and ventilating board.

Preferably, the GPU board comprises a GPU board PCIE high-density port, a GPU high-density port male head, a power port and eight GPUs, the GPU board PCIE high-density port and the GPU high-density port male head are installed at two ends of the GPU board, the eight GPUs are installed on the upper surface of the GPU board, and the power port is installed at one side of the upper surface of the GPU board.

Preferably, the PCIE module comprises a GPU high-density port of the PCIE module, a handle, a power-assisted mechanism, a fastening threaded hole and a side mounting hole, the GPU high-density port of the PCIE module is arranged at the inner side of the PCIE module, the handle is arranged at the outer side of the PCIE module, the bottom of the power-assisted mechanism is provided with a hinge joint and a rotating port, the power-assisted mechanism is hinged on the left side of the PCIE module through a hinge joint, a bolt is arranged on the upper part of the power-assisted mechanism, a fastening threaded hole is arranged on the outer side of the PCIE module, the side mounting holes are arranged at the left side and the right side of the PCIE module, the PCIE module is connected with the PCIE high-density port of the GPU board through the PCIE high-density port along the direction of the guide rail and is arranged at the front end of the 4U server base, the power assisting mechanism is installed on the rotating fulcrum of the 4U server base through the meshing of the rotating port at the lower part, and the power assisting mechanism is installed on the outer side of the PCIE module through the matching of the bolt at the upper part and the fastening threaded hole at the outer side of the PCIE module.

Preferably, two wind scooper installation buckles are respectively arranged on the inner walls of two side faces of the 4U server base, each wind scooper comprises a wind scooper fixing position, a wire clamping groove and a wind guiding opening, and the wind scoopers are installed on the upper portion of the 4U server base in a matched mode through the wind scooper fixing positions and the wind scooper installation buckles.

Preferably, the bridging module include female head of high-density mouthful, bridging board, support, safety cover, guide post, the female head of high-density mouthful install the both ends at the bridging board, the support on be provided with two guide posts from top to bottom respectively, install the bridging board of the female head of two high-density mouths and install the both sides face at the support, the support mounting who installs two bridging boards is in the safety cover.

Compared with the prior art, the invention has the beneficial effects that:

1. the invention designs a 4U server structure capable of accommodating eight GPUs, and simultaneously designs an 8U case capable of storing two 4U servers, namely a server structure supporting 16 GPUs, wherein the GPUs of the two 4U servers are connected through a bridge module at high density, so that the two 4U servers and sixteen GPUs work cooperatively, the requirement of a novel server on space is met, meanwhile, the special structural design of the air guide cover improves the heat dissipation performance of the server, the stable and high-performance cooperative operation of the plurality of servers is ensured, a rear window perforated plate is connected with an upper cover and a base through a rear window through a bolt, the connection between the upper cover and the base is reinforced, the heat dissipation of the server is ensured, and meanwhile, a technician can conveniently observe whether the server has abnormal conditions or not at the rear window.

2. The bottom plate and the left and right sides board formula structure as an organic whole of base have simplified the installation flow, save plenty of time and human cost, have improved the tightness and the durability of quick-witted case simultaneously, and the back window support adopts vertical rectangular structure, connects back window trompil board and base bottom, both can further strengthen being connected of base and upper cover, provides sufficient space for installing bridging module at the back window position between two servers simultaneously.

3. The middle partition plate for separating the two 4U servers is arranged in the middle of the 8U case and is of a rectangular thin-wall structure, so that the two servers are compactly installed in the 8U case, a certain space is provided for ensuring the heat dissipation effect of the servers, and the bottom surface of the base is provided with the anti-skid protrusions, so that the position deviation of the internal servers in the process of moving the case is effectively reduced.

4. The threaded mounting holes of the 4U server base are formed in the front ends of the left side plate and the right side plate, space resources are fully utilized, the 4U server is installed at a fixed position in the 8U case through the matching of the limiting holes in the 4U server base and the limiting nails on the base, the server is prevented from being installed out of position, and space is provided for the bridge module to connect the two servers on the rear window.

5. The power assisting mechanism is installed on the PCIE module in a hinged mode, a rotating fulcrum on the 4U server base is hooked through a rotating port when the PCIE module is installed, through the manual rotating power assisting mechanism, the PCIE module is further pushed into the 4U server base through lever force generated between a hinged joint and the rotating fulcrum, a high-density port of a GPU (graphics processing unit) of the PCIE module is fully combined with a PCIE high-density port of a GPU (graphics processing unit) board on the GPU board, a fastening state is kept through matching of a bolt and a fastening threaded hole, fastening connection between the high-density ports is guaranteed, probability of server failure caused by insufficient connection between the high-density ports is reduced, and stability of the server is greatly improved.

6. 8 GPUs are installed on the upper surface of the GPU board, plane resources are fully utilized, the structure is compact, the power supply port is installed on one side of the upper surface of the GPU board and keeps a certain distance from the GPU, and temperature and signal interference is avoided.

7. The wind scooper sets up the wind-guiding mouth structure, wind energy in the make full use of environment carries out effective wind-guiding heat dissipation to GPU, increase server life, the wind scooper adopts injection molding process one shot forming, high machining efficiency, the wind scooper installation lock through wind scooper fixed position and 4U server base is buckled the cooperation, once push down and to target in place the mountable, remove numerous and diverse bolted connection from, the installation effectiveness is greatly improved, save the assemble duration, reduce the production human cost, the cable that leaks outward can be fixed to the card wire casing, improve system stability when guaranteeing beautifully.

8. Be equipped with the guide post on the support among the bridging module, the clearance fit installation between guide post and two upper and lower 4U servers effectively reduces the emergence probability that the dislocation was connected to the high-density mouthful in the bridging process, four high-density mouthful are passed through the support and the bridging board is firm to be connected, install in the safety cover, the safety cover passes through the high-density mouthful of semi-enclosed mode protection, effectively avoid the bridge module that foreign object touching caused trouble such as not hard up, guarantee the bridging quality, still be equipped with the vent simultaneously and guarantee the heat dissipation.

Drawings

Figure 1 is a schematic structural view of the present invention,

figure 2 is a schematic diagram of the front end structure of the 8U chassis of the present invention,

figure 3 is a schematic diagram of the rear end structure of the 8U chassis of the present invention,

figure 4 is a schematic diagram of a 4U server architecture,

figure 5 is a schematic view of the installation of the wind scooper,

FIG. 6 is a schematic structural diagram of a 4U server near the GPU high-density port,

figure 7 is a schematic diagram of a GPU board structure,

FIG. 8 is a schematic view of the operation principle of the power assisting mechanism,

figure 9 is a schematic structural diagram of the rear end of an assembled 8U chassis,

figure 10 is a schematic diagram of a bridge module configuration,

in the figure, 1, 8U chassis, 2, 4U server, 3, bridge module, 4, bolt, 5, screw, 101, base, 102, upper cover, 103, front window, 104, rear window, 105, middle partition, 106, rear window perforated plate, 107, rear window bracket, 10101, slide rail, 10102, anti-skid projection, 10103, limit nail, 10104, bracket mounting seat, 201, PCIE module, 202, GPU board, 203, 4U server base, 204, wind scooper, 20101, PCIE module GPU high-density port, 20102, handle, 20103, Power assisting mechanism, 20104, fastening screw hole, 20105, side mounting hole, 20201, GPU board PCIE high-density port, 20202, GPU high-density port male, 20203, Power supply port, 20204, GPU, 20301, threaded mounting hole, 20302, limit hole, ip20303, Power clcl, 04, ventilation board, 05, mounting buckle, 20401, 20402, wind scooper mounting position, 20403, 20402, wind scooper, 20402, 20307. rotation fulcrum 301, high-density port female head 302, bridging board 303, support 304, safety cover 305, guide post 2010301, articulated joint 2010302 and rotation port.

Detailed Description

For a better understanding of the present invention, embodiments thereof are explained in detail below with reference to the accompanying drawings.

As shown in fig. 1 to 4, a server structure supporting 16GPU includes an 8U chassis 1, a 4U server 2, and a bridge module 3, where the 8U chassis includes a base 101, an upper cover 102, a middle partition 105, a rear window opening plate 106, and a rear window support 107, the upper cover 102 is installed above the base 101, and forms a four-sided closed structure with an open front end and a open rear end with the base 101, the front window 103 is located at the front end of the base 101, the rear window 104 is located at the rear end of the base 101, a slide rail 10101 is respectively disposed at the middle positions of inner walls of left and right side plates of the base 101, the planes of the two slide rails 10101 are parallel to the bottom surface of the base 101 and the upper cover 102, the middle partition 105 is installed between the two side plates of the base 101 along the direction of the slide rail 10101, the bottom surface of the base 101 is provided with a support mounting seat 10104 at the rear window 104, the rear window opening plate 106 is installed on the upper portion of the rear window 104 of the 8U chassis 1, and connects the upper cover 102 with the left and right side plates of the base 101, one end of the rear window support 107 is connected with the lower portion of the rear window perforated plate 106, the other end of the rear window support is connected with the support mounting seat 10104 of the base 101, the 4U server 2 comprises a 4U server base 203, a PCIE module 201, a GPU plate 202 and an air guide cover 204, the GPU plate 202 is installed in the 4U server base 203, the air guide cover 204 is installed on the upper portion of the 4U server base 203, the PCIE module 201 is connected with the GPU plate 202 and installed at the front end of the 4U server base 203, one 4U server 2 is installed in the space between the middle partition plate 105 and the upper cover 102, one 4U server 2 is installed in the space between the middle partition plate 105 and the bottom face of the base 101, and the two 4U servers 2 are connected with each other through the bridge module 3 at the rear window 104.

The invention designs a 4U server 2 structure capable of containing eight GPUs 20204, and designs an 8U case 1 capable of storing two 4U servers 2, namely a server structure supporting 16 GPUs, wherein the GPUs of the two 4U servers 2 are connected at a high-density port through a bridge module 3, so that the two 4U servers 2 and the sixteen GPUs 20204 work cooperatively to meet the space requirement of a novel server, meanwhile, the special structural design of the air guide cover 204 improves the heat dissipation performance of the server, the stable and high-performance cooperative operation of the servers is ensured, the rear window perforated plate 106 is connected with the upper cover 102 and the base 101 through the bolts 4 at the rear window 104, the connection between the upper cover 102 and the base 101 is strengthened, the heat dissipation performance of the server can be ensured, and meanwhile, a technician can conveniently observe whether the server has abnormal conditions at the rear window 104.

As shown in fig. 2 and 3, the bottom plate and the left and right side plates of the base 101 are integrated, so as to simplify the installation process, save a lot of time and labor cost, and improve the robustness and durability of the chassis, the rear window bracket 107 is a long structure and connects the rear window opening plate 106 with the bottom of the base 101, which can further strengthen the connection between the base 101 and the upper cover 102, and provide enough space for installing the bridge module 3 between two servers at the position of the rear window 104.

As shown in fig. 2, the inner wall of the bottom surface of the base 101 is provided with a plurality of anti-slip protrusions, so that the position deviation of the internal server in the case moving process is effectively reduced, the inner walls of the left and right side plates of the base 101 near one end of the rear window 104 are respectively provided with an upper limit nail 10103 and a lower limit nail 10103, the middle partition plate 105 is of a rectangular thin-wall structure, so that the two servers are compactly installed in the 8U case 1, and a certain space is provided to ensure the heat dissipation effect of the servers.

As shown in fig. 4 to 6, the front ends of the left and right side plates of the 4U server base 203 are provided with a plurality of threaded mounting holes 20301, space resources are fully utilized, wherein the bottom of the inner wall of the front end of the left side plate is provided with a rotation fulcrum 20307, the rotation fulcrum 20307 protrudes out of the plane of the inner wall of the side surface, the outer walls of the rear ends of the left and right side plates of the 4U server base 203 are provided with limiting holes 20302, the limiting holes 20302 are matched with a limiting nail 10103 on the base 101 to mount the 4U server 2 at a fixed position in the 8U case 1, so as to prevent the server from being mounted out of position, so as to provide space for the bridge module 3 to connect the two servers at the rear window 104, the middle parts of the inner walls of the left and right side plates of the 4U server base 203 are respectively provided with a guide rail 20306, and the rear ends of the 4U server base 203 are provided with Power clip20303 and a ventilation plate 20304.

As shown in fig. 7, the GPU board 202 includes a GPU board PCIE high-density port 20201, a GPU high-density port male head 20202, a power port 20203, and eight GPUs 20204, the GPU board PCIE high-density port 20201 and the GPU high-density port male head 20202 are installed at two ends of the GPU board 202, the eight GPUs 20204 are installed on the upper surface of the GPU board 202, so that planar resources are fully utilized, the structure is compact, and the power port 20203 is installed at one side of the upper surface of the GPU board 202 and keeps a certain distance from the GPU20204, thereby avoiding temperature and signal interference.

As shown in fig. 4 to 8, the PCIE module 201 includes a PCIE module GPU high-density port 20101, a handle 20102, a power-assisted mechanism 20103, a fastening screw hole 20104, and a side-mounting hole 20105, the PCIE module GPU high-density port 20101 is disposed on the inner side of the PCIE module 201, the handle 20102 is disposed on the outer side of the PCIE module 201, the bottom of the power-assisted mechanism 20103 is provided with a hinge 2010301 and a rotation port 2010302, the power-assisted mechanism 20103 is hinged on the left side of the PCIE module 201 through a hinge 2010301, the upper portion of the power-assisted mechanism 20103 is provided with a bolt 4, the fastening screw hole 20104 is disposed on the outer side of the PCIE module 201, the side-mounting hole 20105 is disposed on the left and right sides of the PCIE module 201, the PCIE module 201 is connected with the GPU board PCIE high-density port 20201 through the PCIE module GPU high-density port 20101 along the direction 20306, and is mounted at the front end of the U server base 203 of the PCIE server 2034U 203, the rotation-assisted mechanism 20103 is mounted on the rotation fulcrum 203 of the U server base 203 through the rotation port 2010302 at the lower portion 20307, the boosting mechanism 20103 is installed outside the PCIE module 201 through the bolt 4 at the upper part and the fastening screw hole 20104 at the outside of the PCIE module 201,

when the PCIE module 201 is installed, the rotating fulcrum 20307 on the 4U server base 203 is hooked by the rotating port 2010302, the PCIE module 201 is pushed into the 4U server base 203 by the lever force generated between the hinge 2010301 and the rotating fulcrum 20307 through the manual rotation assisting mechanism 20103, so that the PCIE high-density ports 20101 of the PCIE module GPU 20101 and the GPU board PCIE high-density ports 20201 on the GPU board 202 are fully combined, and the fastening state is maintained by the matching of the bolt 4 and the fastening threaded hole 20104, thereby ensuring the fastening connection between the high-density ports, reducing the probability of server failure caused by insufficient connection between the high-density ports, and greatly improving the stability of the server.

As shown in fig. 4 and 5, two wind scooper installation buckles 20305 are respectively arranged on inner walls of two side surfaces of the 4U server base 203, each wind scooper 204 comprises a wind scooper fixing position 20401, a wire clamping groove 20402 and a wind scooper opening 20403, the wind scooper 204 is installed on the upper portion of the 4U server base 203 through the matching of the wind scooper fixing position 20401 and the wind scooper installation buckle 20305 and can be installed in place by pressing down once, complicated bolt 4 connection is omitted, the installation efficiency is greatly improved, the assembly time is saved, the production labor cost is reduced, the wind scooper 204 is provided with a wind scooper structure, wind energy in the environment is fully utilized to effectively guide wind and dissipate heat of the GPU20204, the service life of the server is prolonged, the wind scooper 204 is formed by injection molding once, the processing efficiency is high, the wire clamping groove 20402 can fix an external leakage cable, and the system stability is improved while the appearance is ensured.

As shown in fig. 9 and 10, the bridge module 3 includes a high-density female head 301, a bridge plate 302, a support 303, a protective cover 304, and guide posts 305, the high-density female head 301 is installed at two ends of the bridge plate 302, two guide posts 305 are respectively installed above and below the support 303, the bridge plate 302 installed with the two high-density female heads 301 is installed at two side surfaces of the support 303, the support 303 installed with the two bridge plates 302 is installed in the protective cover 304, and the guide posts 305 are installed in clearance fit with the upper and lower 4U servers 2, so as to effectively reduce the occurrence probability of high-density connection dislocation during bridging, the four high-density ports are stably connected with the bridge plate 302 through the support 303 and are installed in the protective cover 304, the protective cover 304 protects the high-density ports through a semi-surrounding manner, effectively avoid the occurrence of failures such as loosening of the bridge module 3 caused by the touch of foreign objects, and ensure the bridging quality, meanwhile, a vent is arranged to ensure heat dissipation.

When the installation is carried out, firstly, the upper cover 102 is installed on the upper part of the base 101 through the bolt 4, the middle partition plate 105 is installed in the base 101 along the sliding rail 10101, the rear window perforated plate 106 is installed at the position of the rear window 104 through the bolt 4, the upper cover 102 and the base 101 are connected, the rear window support 10104 is installed between the base 101 and the rear window perforated plate 106, the base 101 and the rear window perforated plate 106 are connected, and the installation of the 8U case 1 is completed; secondly, the GPU board 202 is installed in the 4U server base 203, the PCIE module GPU high-density port 20201 and the GPU board PCIE high-density port 20201 on the GPU board 202 are connected and installed at one end of the 4U server base 203 and are clamped through the power assisting mechanism 20103, the air guide cover 204 is pressed downwards and installed on the upper portion of the 4U server base 203, and the installation of the 4U server 2 is completed; then, pushing and installing one 4U server 2 below the middle partition plate 105 of the 8U case 1, and pushing and installing one 4U server 2 above the middle partition plate 105 until the limiting holes 20302 on the two 4U servers 2 are matched with the limiting nails 10103 on the base 101 in place; thirdly, mounting two high-density female heads 301 on two bridge plates 302 through screws 5, mounting the two bridge plates 302 provided with the high-density female heads 301 on two sides of a support 303 through the screws 5, and mounting the support 303 provided with the two bridge plates 302 in a protective cover 304 through the screws 5; finally, the bridging module 3 is used for connecting the GPU high-density port male heads 20202 of the upper and lower 4U servers 2 at the position of the rear window 104 of the 8U case 1 through the upper and lower high-density port female heads 301, and the bridging module 3 is fixed on the rear window perforated plate 106 through the bolts 4, so that communication interconnection between the two servers is realized.

It will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in the embodiments described above without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims.

Claims

1. A server structure supporting a 16GPU comprises an 8U case, a 4U server and a bridging module, and is characterized in that the 8U case comprises a base, an upper cover, a middle partition plate, a rear window perforated plate and a rear window support, wherein the upper cover is arranged above the base and forms a four-sided closed structure with the base, the front end and the rear end of the four-sided closed structure are open, the front window is arranged at the front end of the base, the rear window is arranged at the rear end of the base, the middle positions of the inner walls of the left side plate and the right side plate of the base are respectively provided with a slide rail, the planes of the two slide rails are parallel to the bottom surface of the base and the upper cover, the middle partition plate is arranged between the two side plates of the base along the direction of the slide rails, the bottom surface of the base is provided with a support mounting seat at the rear window, the rear window perforated plate is arranged at the upper part of the rear window of the 8U case and is connected with the upper cover and the left side plate and the right side plate of the base, one end of the rear window support is connected with the support mounting seat of the base, the 4U server comprises a 4U server base, a PCIE module, a GPU board and an air guide cover, wherein the GPU board is installed in the 4U server base, the air guide cover is installed on the upper portion of the 4U server base, the PCIE module is connected with the GPU board and installed at the front end of the 4U server base, one 4U server is installed in a space between the middle partition plate and the upper cover, one 4U server is installed in a space between the middle partition plate and the bottom surface of the base, and the two 4U servers are connected through the bridge module at the rear window;

the bottom plate and the left and right side plates of the base are of an integrated structure, and the rear window support is of a long strip structure.

2. The server structure supporting a 16GPU as claimed in claim 1, wherein the inner wall of the bottom surface of the base is provided with a plurality of anti-skid protrusions, the inner walls of the left and right side plates at one end of the base close to the rear window are respectively provided with an upper limit pin and a lower limit pin, and the middle partition plate is of a rectangular thin-wall structure.

3. The server structure supporting 16 GPUs as claimed in claim 1, wherein the front ends of the left and right side plates of the 4U server chassis are provided with a plurality of threaded mounting holes, wherein the bottom of the inner wall of the front end of the left side plate is provided with a rotation fulcrum, the rotation fulcrum protrudes out of the plane of the inner wall of the side surface, the outer walls of the rear ends of the left and right side plates of the 4U server chassis are provided with limiting holes, the limiting holes are matched with limiting pins on the chassis to mount the 4U server at a fixed position in the 8U chassis, the middle parts of the inner walls of the left and right side plates of the 4U server chassis are respectively provided with a guide rail, and the rear end of the 4U server chassis is provided with a Power clip and a ventilation plate.

4. The server structure supporting 16 GPUs according to claim 1, wherein the GPU board comprises a GPU board PCIE high-density port, a GPU high-density port male power port and eight GPUs, the GPU board PCIE high-density port and the GPU high-density port male are installed at two ends of the GPU board, the eight GPUs are installed on the upper surface of the GPU board, and the power port is installed on one side of the upper surface of the GPU board.

5. The structure of claim 1, wherein the PCIE module comprises a PCIE module GPU high-density port, a handle, a boosting mechanism, a fastening threaded hole and a side mounting hole, the PCIE module GPU high-density port is arranged on the inner side of the PCIE module, the handle is arranged on the outer side of the PCIE module, the boosting mechanism is hinged on the left side of the PCIE module through a hinge joint, the bottom of the boosting mechanism is provided with a hinge joint and a rotating port, the upper portion of the boosting mechanism is provided with a bolt, the fastening threaded hole is arranged on the outer side of the PCIE module, the side mounting hole is arranged on the left side and the right side of the PCIE module, the PCIE module is connected with the PCIE high-density port of the GPU board through the PCIE high-density port in the direction of the guide rail and is arranged at the front end of the 4U server base, the boosting mechanism is arranged on the rotating fulcrum of the 4U server base through the rotating port on the lower portion, the power-assisted mechanism is installed in the PCIE module outside through the bolt cooperation in upper portion and the fastening screw hole in the PCIE module outside.

6. The server structure supporting a 16GPU as claimed in claim 1, wherein two wind scooper mounting buckles are respectively arranged on the inner walls of two side faces of the 4U server base, each wind scooper comprises a wind scooper fixing position, a wire clamping groove and a wind guiding opening, and the wind scoopers are mounted on the upper portion of the 4U server base through the wind scooper fixing positions and the wind scooper mounting buckles in a matched mode.

7. The server structure supporting a 16GPU according to claim 1, wherein the bridge module comprises high-density female heads, bridge plates, supports, protective covers and guide posts, the high-density female heads are installed at two ends of each bridge plate, the supports are respectively provided with two guide posts from top to bottom, the bridge plates provided with the two high-density female heads are installed on two side surfaces of each support, and the supports provided with the two bridge plates are installed in the protective covers.