Multi-card GPU server

Abstract

The utility model relates to the field of GPU servers, and discloses a multi-card GPU server which comprises a server box, a protective shell, GPU blocks, connecting devices, clamping blocks and fan sets. Connecting holes are formed in the four corners of the end face of the end, correspondingly connected with the protective shell, of the server box; the end face of the end, correspondingly connected with the server box, of the protective shell is provided with connecting holes connected with the connecting holes in a matched mode through adaptive screws, sliding grooves are formed in the positions, correspondingly connected with the GPU blocks, of the interior of the server box at equal intervals, and guide sliding blocks connected with the sliding grooves in a matched and sliding mode are arranged in the positions, correspondingly connected with the interior of the server box, of the GPU blocks. The multi-card GPU server comprises a server box, a plurality of GPU blocks are arranged in the server box, fan sets are arranged at the two ends of each GPU block, eight connecting devices which are distributed at equal intervals are arranged in the server box and connected with the matching faces of the GPU blocks, partition plate sets are arranged in the server box, and clamping blocks which are symmetrically distributed at equal intervals are connected with the GPU blocks in a matched and clamped mode. And the maintenance time and cost are reduced.

Description

Technical Field

[0001] The utility model relates to the field of GPU servers, in particular to a multi-card GPU server. Background Art

[0002] A GPU, or graphics processing unit, offloads an application's graphics processing instructions from the main processor. This process frees up the main processor's resources for other tasks and executes the application's graphics processing instructions on hardware, meeting our current demand for more sophisticated, lifelike rendering, video processing, and visualization. Without a GPU, graphics instructions would require software emulation, which would then occupy the main processor's resources, resulting in unacceptably low performance.

[0003] Most GPU servers on the market only have one GPU module and are fixed with screws. Some are multi-GPU servers. However, continuing to use screws to fix the connection increases installation time and cost, significantly reducing the installation efficiency of the GPU module. To this end, we have proposed a multi-GPU server. Utility Model Content

[0004] (1) Technical problems solved

[0005] In view of the deficiencies in the prior art, the present invention provides a multi-GPU server to solve the above problems.

[0006] (2) Technical solution

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a multi-card GPU server, comprising a server box, a protective shell, a GPU block, a connecting device, a clamping block and a fan group, wherein the server box is provided with connecting holes at four corners of an end face corresponding to one end connected to the protective shell, and the protective shell is provided with connecting holes on an end face corresponding to one end connected to the server box, which are connected to the connecting holes by means of adapter screws, the server box is provided with equidistant sliding grooves at positions corresponding to the positions connected to the GPU blocks, the GPU blocks are provided with guide sliders slidingly connected to the sliding grooves at the interior of the server box corresponding to the interior of the server box, and fan groups are provided at both ends of the GPU blocks, the server box is provided with eight equidistantly distributed connecting devices connected to the mating surfaces of the GPU blocks, and the server box is provided with a partition group with symmetrically and equidistantly distributed clamping blocks that are clamped to the GPU blocks.

[0008] Preferably, the server box as a whole is a rectangular box, and the end of the server box corresponding to the protective shell is an open structure. Three vertical partitions are provided inside the server box, and a slide groove is provided between two adjacent vertical partitions and is axially symmetrically distributed and connected to the GPU block, and rectangular bins are provided at both ends of the two partitions and are connected to the fan group. The partitions inside the server box are provided with equidistantly distributed ventilation holes on both sides parallel to each other, and the end of the server box corresponding to the protective shell is provided with two rows of equidistantly distributed matching holes, and a hollow block is provided inside the matching hole to match the snap-in block.

[0009] Preferably, the connecting device consists of three parts, namely a data connecting rod, a spring part and a matching block, and the connecting device is welded to the end face of the server box facing away from the protective shell, the spring part is concentric with the data connecting rod, and the spring part is welded to the matching block at one end of the protective shell.

[0010] Preferably, the main body of the GPU block is a rectangular plate surface, and the GPU block is provided with two axially symmetrically distributed matching bevel blocks at one end corresponding to the connection device, and a data connection hole for matching and connecting with the data connection rod is provided at the center of one end of the matching bevel block facing away from the GPU block body, the matching bevel block is in an inverted trapezoidal shape, and a sliding bevel block is provided at the end of the matching bevel block facing away from the connection device.

[0011] Preferably, the sliding bevel is trapezoidal as a whole, the sliding bevel is consistent with the external shape of the data connection hole, and a matching groove for sliding with the GPU block is opened at the center of the sliding bevel, and the end of the sliding bevel corresponding to the matching bevel is a large end face.

[0012] Preferably, the fan group consists of two parts, fan 1 and fan 2, and the fan groups are correspondingly distributed at both ends of the GPU block, and both fan 1 and fan 2 are perpendicular to the GPU block.

[0013] Preferably, the snap-in block consists of three parts, namely a connecting block, a matching bevel block 2 and three equidistantly distributed spring groups welded together. The middle part of the snap-in block is the spring group, the end of the snap-in block corresponding to the welded hollow block is the connecting block, and the snap-in block corresponding to the matching bevel block 1 of the GPU block and the sliding bevel block is the matching bevel block 2.

[0014] Preferably, the end face of the mating bevel block 2 facing away from the spring group is an inclined surface, and the angle of the inclined surface is consistent with the side inclined surface of the mating bevel block 1 and the sliding bevel block, and the mating positions of the pointed corners of the inclined surfaces of the mating bevel block 2, the mating bevel block 1 and the sliding bevel block are all arc chamfers.

[0015] (3) Beneficial effects

[0016] Compared with the existing technology, the present invention provides a multi-GPU server with the following advantages:

[0017] The multi-GPU server uses a card-type connection method for GPU connection, which reduces the time cost of fixed installation and improves installation efficiency. The multi-GPU design enables the server to improve computing efficiency, enhance task parallel execution capabilities, strengthen floating-point computing capabilities, and reduce hardware costs and maintenance difficulties. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the structure of a multi-GPU server in this utility model;

[0019] Figure 2 This is a front cross-sectional schematic diagram of the multi-card GPU server of the present invention;

[0020] Figure 3 This is a schematic diagram of a left-side cross-sectional view of a multi-card GPU server of the present invention;

[0021] Figure 4 for Figure 3 A partial enlarged schematic diagram in the middle;

[0022] Figure 5 This is a schematic diagram of the server box of the utility model;

[0023] Figure 6 This is a schematic diagram of the GPU block of the utility model;

[0024] Figure 7 This is a schematic diagram of the sliding inclined block of the utility model;

[0025] Figure 8 This is a schematic diagram of the clamping block of the utility model.

[0026] In the figure: 1. Server box; 2. Protective shell; 3. Fan 1; 4. GPU block; 5. Fan 2; 6. Connecting device; 7. Snap-in block; 8. Partition; 9. Slide groove; 10. Ventilation hole; 11. Hollow block; 12. Matching hole; 13. Connecting hole; 14. Guide slider; 15. Matching inclined block 1; 16. Sliding inclined block; 17. Data connection hole; 18. Matching slide groove; 19. Connecting block; 20. Spring assembly; 21. Matching inclined block 2; 22. Data connecting rod; 23. Spring member; 24. Matching block. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] See also Figure 1-8 , a multi-card GPU server includes a server box 1, a protective shell 2, a GPU block 4, a connecting device 6, a card block 7 and a fan group. The server box 1 is provided with connecting holes 13 at the four corners of the end face corresponding to the end connected to the protective shell 2, and the protective shell 2 is provided with connecting holes on the end face corresponding to the end connected to the server box 1, which are connected to the connecting holes 13 through adapter screws. The server box 1 is provided with equidistant slide grooves 9 at the position corresponding to the connection of the GPU block 4. The GPU block 4 is provided with a guide slider 14 that is slidably connected to the slide groove 9 inside the server box 1, and fan groups are provided at both ends of the GPU block 4. The server box 1 is provided with eight equidistantly distributed connecting devices 6 that are connected to the mating surface of the GPU block 4, and the server box 1 is provided with a partition 8 group of symmetrically and equidistantly distributed card blocks 7 that are engaged with the GPU block 4.

[0029] Furthermore, the server box 1 is a rectangular box as a whole, and the end of the server box 1 corresponding to the protective shell 2 is an open structure. Three vertical partitions 8 are provided inside the server box 1, and a slide groove 9 is provided between two adjacent vertical partitions 8 and is connected to the GPU block 4 in an axially symmetrical manner, and a rectangular bin is provided at both ends of the two partitions 8 to be connected to the fan group. The partitions 8 inside the server box 1 are provided with equidistantly distributed ventilation holes 10 on both sides parallel to each other. The end of the server box 1 corresponding to the side away from the protective shell 2 is provided with two rows of equidistantly distributed matching holes 12, and a hollow block 11 is provided inside the matching hole 12 to be connected to the clamping block 7. The ventilation holes 10 opened on the partition 8 cooperate with the fan group to greatly increase the temperature inside the server and improve computing efficiency.

[0030] Furthermore, the connecting device 6 is composed of three parts, namely a data connecting rod 22, a spring member 23 and a matching block 24, and the connecting device 6 is welded to the end face of the server box 1 facing away from the protective shell 2, the spring member 23 is concentric with the data connecting rod 22, and the spring member 23 is welded to the matching block 24 at one end of the protective shell 2, and the spring member 23 is always in a compressed state to ensure the stability of the fixed connection to the GPU block 4.

[0031] Furthermore, the main body of the GPU block 4 is a rectangular plate surface, and the GPU block 4 is provided with two mating bevel blocks 15 axially symmetrically distributed at one end corresponding to the connection device 6, and a data connection hole 17 for mating with the data connection rod 22 is opened at the center of one end of the mating bevel block 15 facing away from the GPU block 4 body. The mating bevel block 15 is in an inverted trapezoidal shape, and a sliding bevel block 16 is provided at the end of the mating bevel block 15 facing away from the connection device 6. The mating bevel block 15 and the sliding bevel block 16 are horizontally symmetrical and closed when idle.

[0032] Furthermore, the sliding bevel 16 is trapezoidal as a whole, and the sliding bevel 16 is consistent with the external shape of the data connection hole 17, and a mating groove 18 for sliding with the GPU block 4 is opened at the center position of the sliding bevel 16. The end of the sliding bevel 16 corresponding to the mating bevel 15 is a large end face, and the sides of the sliding bevel 16 and the mating bevel 15 are both inclined bevel structures, which are convenient for mating with the card block 7.

[0033] Furthermore, the fan group consists of two parts, fan 1 3 and fan 2 5, and the fan groups are distributed at both ends of the GPU block 4. Fan 1 3 and fan 2 5 are both perpendicular to the GPU block 4. The wind force of the fan group is outward wind, which, combined with the ventilation holes 10 opened on the partition 8, can improve the heat dissipation efficiency.

[0034] Furthermore, the clamping block 7 is composed of three parts, namely a connecting block 19, a matching bevel block 2 21 and three equidistantly distributed spring groups 20 welded together. The middle part of the clamping block 7 is the spring group 20, the end of the clamping block 7 corresponding to the welded hollow block 11 is the connecting block 19, and the clamping block 7 corresponding to the matching bevel block 15 and the sliding bevel block 16 of the GPU block 4 is the matching bevel block 2 21. The spring group 20 is in a compressed state when working and in a relaxed state when idle, and the GPU block 4 is effectively clamped by using the compression rebound effect.

[0035] Furthermore, the end face of the mating bevel block 21 facing away from the spring group 20 is an inclined surface, and the angle of the inclined surface is consistent with the side inclined surface of the mating bevel block 15 and the sliding bevel block 16. The pointed corners of the inclined surfaces of the mating bevel block 21, the mating bevel block 15 and the sliding bevel block 16 are all arc chamfered at the positions where they cooperate with each other. The arc chamfers of the three facilitate their interaction with each other, thereby improving the smoothness of installation and removal of the GPU block 4.

[0036] Working principle: Install the multi-card GPU server correctly according to the diagram, and insert multiple GPU blocks 4 into the corresponding positions inside the server box 1 through the cooperation of the guide slider 14 and the slide groove 9. When the matching bevel block 15 at the bottom of the GPU block 4 contacts the surface of the card block 7 inside the server box 1, the card block 7 is compressed and recovered due to the extrusion effect, causing the GPU block 4 to continue to slide down until the matching bevel block 15 exceeds the card block 7. The card block 7 rebounds quickly, and the matching bevel block 15 contacts the matching block 24 of the connecting device 6. The connecting device 6 begins to compress. Let go, and the rebound effect of the connecting device 6 causes the GPU block 4 to be connected to the inside of the server box 1 , and the data between the GPU block 4 and the connecting device 6 is connected and transmitted through the data connecting rod 22 and the data connecting hole 17, so that four GPU blocks 4 with equal distances are inserted into the server box 1. When the GPU block 4 needs to be taken out, it is only necessary to continue to press the GPU block 4 downward so that the bottom of the GPU block 4 and the matching inclined block 15 are laterally symmetrical and freely sliding data connecting hole 17 are also inserted into it. At this time, the connecting device 6 is compressed, and the rebound effect of the connecting device 6 causes the closed matching inclined block 15 and the data connecting hole 17 to be squeezed upward, and the clamping block 7 is compressed and recovered again until the GPU block 4 is taken out, thus completing the installation and removal work.

[0037] Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-GPU server, comprising a server box (1), a protective housing (2), a GPU block (4), a connecting device (6), a card block (7), and a fan assembly, characterized in that: The server box (1) is provided with connection holes (13) at four corners of the end surface corresponding to the end connected to the protective shell (2), and the protective shell (2) is provided with connection holes that are connected to the connection holes (13) through adapter screws at the end surface corresponding to the end connected to the server box (1). The server box (1) is provided with equidistant slide grooves (9) at positions corresponding to the positions connected to the GPU block (4). The GPU block (4) is provided with guide sliders (14) that are slidably connected to the slide grooves (9) at the interior of the server box (1), and fan groups are provided at both ends of the GPU block (4). The server box (1) is provided with eight equidistantly distributed connection devices (6) that are connected to the matching surface of the GPU block (4), and the server box (1) is provided with a partition (8) group of symmetrically and equidistantly distributed clamping blocks (7) that are clamped to the GPU block (4).

2. The multi-GPU server according to claim 1, wherein: The server box (1) is a rectangular box as a whole. The end of the server box (1) corresponding to the protective shell (2) is an open structure. Three vertical partitions (8) are provided inside the server box (1). A slide groove (9) is provided between two adjacent vertical partitions (8) and is connected to the GPU block (4) in an axially symmetrical manner. A rectangular bin is provided at both ends of the two partitions (8) and is connected to the fan group. The partitions (8) inside the server box (1) are provided with ventilation holes (10) distributed at equal intervals on both sides thereof. Two rows of matching holes (12) distributed at equal intervals are provided inside the server box (1) corresponding to the end away from the protective shell (2). A hollow block (11) is provided inside the matching hole (12) and is connected to the clamping block (7).

3. The multi-GPU server according to claim 1, wherein: The connecting device (6) is composed of three parts, namely a data connecting rod (22), a spring member (23) and a matching block (24), and the connecting device (6) is welded to an end face of the server box (1) facing away from the protective shell (2), the spring member (23) is concentric with the data connecting rod (22), and the spring member (23) is welded to the matching block (24) at one end of the protective shell (2).

4. The multi-GPU server according to claim 2, wherein: The main body of the GPU block (4) is a rectangular plate surface. The GPU block (4) is provided with two axially symmetrically distributed matching bevel blocks (15) at one end corresponding to the connection device (6). A data connection hole (17) for matching and connecting with a data connection rod (22) is provided at the center of one end of the matching bevel block (15) facing away from the GPU block (4) body. The matching bevel block (15) is in an inverted trapezoidal shape. A sliding bevel block (16) is provided at one end of the matching bevel block (15) facing away from the connection device (6).

5. The multi-GPU server according to claim 4, wherein: The sliding bevel (16) is trapezoidal in shape as a whole. The sliding bevel (16) is consistent with the external shape of the data connection hole (17). A matching sliding groove (18) for sliding with the GPU block (4) is provided at the center of the sliding bevel (16). The end of the sliding bevel (16) corresponding to the matching bevel block (15) is a large end surface.

6. The multi-GPU server according to claim 1, wherein: The fan group consists of two parts, fan one (3) and fan two (5), and the fan group is distributed at both ends of the GPU block (4), and both fan one (3) and fan two (5) are perpendicular to the GPU block (4).

7. The multi-GPU server according to claim 2, wherein: The clamping block (7) is composed of three parts, namely a connecting block (19), a second matching bevel block (21), and three spring groups (20) distributed at equal distances and correspondingly welded. The middle part of the clamping block (7) is the spring group (20), one end of the clamping block (7) corresponding to the welded hollow block (11) is the connecting block (19), and the clamping block (7) corresponding to the matching bevel block (15) and the sliding bevel block (16) of the GPU block (4) is the matching bevel block (21).

8. The multi-GPU server according to claim 7, wherein: The end face of the second mating bevel block (21) facing away from the spring assembly (20) is an inclined surface, and the angle of the inclined surface is consistent with the side inclined surfaces of the first mating bevel block (15) and the sliding bevel block (16). The positions where the pointed corners of the inclined surfaces of the second mating bevel block (21), the first mating bevel block (15) and the sliding bevel block (16) are mated with each other are all arc chamfers.

Images