CN210109806U

CN210109806U - GPU interconnection structure

Info

Publication number: CN210109806U
Application number: CN201921003997.0U
Authority: CN
Inventors: 王红明; 崔博; 薛广营
Original assignee: Suzhou Wave Intelligent Technology Co Ltd
Current assignee: Suzhou Wave Intelligent Technology Co Ltd
Priority date: 2019-06-29
Filing date: 2019-06-29
Publication date: 2020-02-21
Anticipated expiration: 2029-06-29

Abstract

The utility model provides a GPU interconnection structure, including the first module shell and the second module shell of superpose, be equipped with first GPU integrated circuit board in the first module shell, be equipped with second GPU integrated circuit board in the second module shell, first GPU integrated circuit board and second GPU integrated circuit board are connected through high-speed connector, are connected with the apron on the second module shell. The utility model discloses an ingenious design to GPU interconnection structure makes its integrated circuit board width that has solved many GPU interconnections too big, can't erect the problem of inserting under limited quick-witted case high condition.

Description

GPU interconnection structure

Technical Field

The invention relates to the field of telecommunication equipment, in particular to a GPU (graphics processing unit) interconnection structure.

Background

With the rapid improvement of computer level in recent years, the functions of face recognition and automatic driving of automobiles are rapidly developed, the applications are based on a deep learning algorithm, and most of the hardware platforms supporting the algorithm at present are based on the strong parallel computing capability of a GPU (graphics processing Unit). Generally, the high-end explicit cocaine 3D operation throughput is huge, and usually requires an X16 channel, and such server platforms have special requirements on PCIE resources. Therefore, in order to meet the requirement of ultra-strong computing power, the current server design usually adopts multiple CPUs, and multiple standard GPU cards are integrated inside the chassis.

In a traditional 4-GPU chip interconnection structure scheme, usually 4 GPUs are placed on the same PCB board card, and interconnection signals are realized through routing inside the PCB. Basically, the scheme can only be applied to a transverse card insertion machine box, and for a machine box with the height of 4U and below which a vertical card is required to be inserted, the width of the card is too large due to the interconnection of 4 GPUs, and the vertical card insertion cannot be carried out under the condition of limited machine box height. This limits the 4GPU interconnection. Still other 4GPU interconnection schemes employ a bridge board or a bridge cable, which is costly and difficult to maintain.

Disclosure of Invention

The invention provides a GPU (graphics processing unit) interconnection structure, which solves the problem that in the prior art, a plurality of GPU interconnection board cards are too wide to be vertically inserted under the condition of limited chassis height.

A GPU interconnection structure comprises a first module shell and a second module shell which are overlapped, wherein a first GPU board card is arranged in the first module shell, a second GPU board card is arranged in the second module shell, the first GPU board card and the second GPU board card are connected through a high-speed connector, and a cover plate is connected onto the second module shell. A plurality of GPU chips are respectively arranged on the two board cards, so that the width of each GPU board card is reduced, and vertical insertion can be realized under the condition of limited chassis height. The first GPU board card and the second GPU board card are connected through the high-speed connector, and therefore interconnection among GPU chips is achieved.

Furthermore, a GPU chip and a first high-speed connector are arranged on the front face of the first GPU board card, a GPU chip is arranged on the front face of the second GPU board card, and a second high-speed connector used for being connected with the first high-speed connector is arranged on the back face of the second GPU board card. The first GPU board card and the second GPU board card are connected through the first high-speed connector and the second high-speed connector, and therefore interconnection among GPU chips is achieved.

Further, first GPU integrated circuit board and second GPU integrated circuit board parallel arrangement, install two GPU chips on the front of first GPU integrated circuit board side by side, be equipped with four female first high-speed connectors on the front of first GPU integrated circuit board side by side, two GPU chips are installed on the front of second GPU integrated circuit board side by side, the back of second GPU integrated circuit board side by side be equipped with four be used for respectively with four female first high-speed connector butt joint's public first high-speed connector. The two GPU chips are arranged on the first GPU board card and the second GPU board card respectively, the width of the GPU board card is reduced, the GPU board card can be vertically plugged in a lower case, the space occupied by the two GPU board cards is reduced due to the fact that the first GPU board card and the second GPU board card are arranged in parallel, the first GPU board card is connected with the second GPU board card through the four female high-speed connectors and the four male high-speed connectors, and the GPU chips are interconnected.

Further, the second module shell comprises a second bottom plate parallel to the second GPU board card, two opposite sides of the second bottom plate are fixedly connected with second side plates perpendicular to the second bottom plate respectively, one side of the second bottom plate is provided with a second front baffle perpendicular to the second bottom plate and the second side plates, and the second bottom plate is provided with a hole for the male high-speed connector to pass through. The second bottom plate, the second side plate and the second front baffle form a semi-closed structure, the second GPU board card is positioned and protected, the second bottom plate is provided with a hole, the male high-speed connector can penetrate through the hole, and the male high-speed connector and the female high-speed connector are butted. The surface of the second module shell, which is opposite to the second front baffle, is in an open state, so that the second GPU board card is conveniently connected with the mainboard.

Further, the first module shell comprises a first bottom plate parallel to the first GPU board card, two opposite sides of the first bottom plate are fixedly connected with first side plates perpendicular to the first bottom plate respectively, and a first front baffle perpendicular to the first bottom plate and the first side plates is arranged on one side of the first bottom plate. The first bottom plate, the first side plate and the first front baffle form a semi-closed structure, and the first GPU board card is positioned and protected. The first module shell is in an open state on the surface opposite to the first front baffle, so that the first GPU board card is conveniently connected with the mainboard.

Furthermore, the second bottom plate is fixedly connected with a first mounting bar which is located in the same plane as the second side plate, the first mounting bar and the second side plate are located on two sides of the second bottom plate respectively, and the first side plate is connected with the first mounting bar through screws. First mounting bar and second bottom plate fixed connection, first curb plate pass through screwed connection with first mounting bar, have realized the demountable installation of second bottom plate and first curb plate.

Further, the cover plate is parallel to the second bottom plate, second mounting strips parallel to the second side plate are fixedly connected to two opposite sides of the cover plate, and the second side plate is connected with the second mounting strips through screws. The second mounting bar is fixedly connected with the cover plate, and the second side plate is connected with the second mounting bar through screws to realize the detachable mounting of the second side plate and the cover plate. The cover plate further improves the sealing performance of the second module shell and strengthens the protection of the second GPU board card.

Furthermore, through holes are densely distributed in the first front baffle and the second front baffle, operating handles are arranged on the first front baffle and the second front baffle, and pulling aids are arranged on the first bottom plate and the second bottom plate. The through holes densely distributed on the first front baffle and the second front baffle are favorable for airflow to enter the first module shell and the second module shell, and heat dissipation is facilitated. The operating handle arranged on the first front baffle and the second front baffle is convenient for the push-pull operation of the first module shell and the second module shell. The pulling-assisting device arranged on the first bottom plate and the second bottom plate is convenient for pulling and inserting the first module shell and the second module shell.

The circuit structure of first GPU integrated circuit board and second GPU integrated circuit board and the relation of connection of GPU chip and first GPU integrated circuit board and second GPU integrated circuit board all adopt prior art. The utility model discloses the front of first GPU integrated circuit board be the one side that is close first bottom plate on the first GPU integrated circuit board, the back of first GPU integrated circuit board be on the first GPU integrated circuit board with the positive relative one side of first GPU integrated circuit board. The utility model discloses the front of second GPU integrated circuit board be the one side that is close the second bottom plate on the second GPU integrated circuit board, the back of second GPU integrated circuit board be on the second GPU integrated circuit board with the positive relative one side of second GPU integrated circuit board.

According to the technical scheme, the invention has the following advantages:

the first GPU board card and the second GPU board card are connected through the high-speed connector, and therefore interconnection among GPU chips is achieved. A plurality of GPU chips are respectively arranged on the two board cards, so that the width of each GPU board card is reduced, and vertical insertion can be realized under the condition of limited chassis height. The space occupied by each GPU board card in a single direction is reduced, and the compactness of the structure in the case is improved.

Drawings

In order to more clearly illustrate the technical solution of the present invention, the drawings used in the description will be briefly introduced, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.

Fig. 1 is a schematic view of the installation relationship between the first module housing and the second module housing of the present invention.

Fig. 2 is a schematic view of the installation relationship of the components of the present invention.

Fig. 3 is a schematic plan view of the installation relationship of the components of the present invention.

Fig. 4 is a schematic view of the screw connection relationship between the first module housing and the second module housing of the present invention.

Fig. 5 is a schematic view of the installation relationship between the present invention and the case.

1. First GPU integrated circuit board, 2, second GPU integrated circuit board, 3, apron, 4, GPU chip, 5, female first high-speed connector, 6, public first high-speed connector, 7, second bottom plate, 8, second curb plate, baffle before 9, the second, 10, trompil, 11, first bottom plate, 12, first curb plate, 13, baffle before first, 14, help and pull out the ware, 15, first mounting bar, 16, second mounting bar, 17, operating handle, 18, first module shell, 19, second module shell.

Detailed Description

In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the present embodiment, and it is apparent that the embodiments described below are only a part of embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the scope of protection of this patent.

As shown in fig. 1-5, the present invention provides a GPU interconnection structure, which includes a first module housing and a second module housing that are stacked, wherein a first GPU board card is disposed in the first module housing, a second GPU board card is disposed in the second module housing, the first GPU board card and the second GPU board card are connected through a high-speed connector, and a cover plate is connected to the second module housing. A plurality of GPU chips are respectively arranged on the two board cards, so that the width of each GPU board card is reduced, and vertical insertion can be realized under the condition of limited chassis height. The first GPU board card and the second GPU board card are connected through the high-speed connector, and therefore interconnection among GPU chips is achieved. The first module shell comprises a first bottom plate parallel to the first GPU board card, two opposite sides of the first bottom plate are fixedly connected with a first side plate perpendicular to the first bottom plate respectively, and one side of the first bottom plate is provided with a first front baffle perpendicular to the first bottom plate and the first side plate. The first bottom plate, the first side plate and the first front baffle form a semi-closed structure, and the first GPU board card is positioned and protected. The first module shell is in an open state on the surface opposite to the first front baffle, so that the first GPU board card is conveniently connected with the mainboard. The second module shell comprises a second bottom plate parallel to the second GPU board card, two opposite sides of the second bottom plate are fixedly connected with second side plates perpendicular to the second bottom plate respectively, one side of the second bottom plate is provided with a second front baffle perpendicular to the second bottom plate and the second side plates, and the second bottom plate is provided with a hole for the male high-speed connector to pass through. The second bottom plate, the second side plate and the second front baffle form a semi-closed structure, the second GPU board card is positioned and protected, the second bottom plate is provided with a hole, the male high-speed connector can penetrate through the hole, and the male high-speed connector and the female high-speed connector are butted. The surface of the second module shell, which is opposite to the second front baffle, is in an open state, so that the second GPU board card is conveniently connected with the mainboard.

Fixedly connected with on the second bottom plate with the second curb plate is located the first mounting bar of coplanar, first mounting bar with the second curb plate is located respectively the both sides of second bottom plate, first curb plate passes through screwed connection with first mounting bar. First mounting bar and second bottom plate fixed connection, first curb plate pass through screwed connection with first mounting bar, have realized the demountable installation of second bottom plate and first curb plate. The cover plate is parallel to the second bottom plate, second mounting strips parallel to the second side plate are fixedly connected to two opposite sides of the cover plate, and the second side plate is connected with the second mounting strips through screws. The second mounting bar is fixedly connected with the cover plate, and the second side plate is connected with the second mounting bar through screws to realize the detachable mounting of the second side plate and the cover plate. The cover plate further improves the sealing performance of the second module shell and strengthens the protection of the second GPU board card.

Through holes are densely distributed in the first front baffle and the second front baffle, operation handles are arranged on the first front baffle and the second front baffle, and pulling-up assisting devices are arranged on the first bottom plate and the second bottom plate. The through holes densely distributed on the first front baffle and the second front baffle are favorable for airflow to enter the first module shell and the second module shell, and heat dissipation is facilitated. The operating handle arranged on the first front baffle and the second front baffle is convenient for the push-pull operation of the first module shell and the second module shell. The pulling-assisting device arranged on the first bottom plate and the second bottom plate is convenient for pulling and inserting the first module shell and the second module shell.

The front side of the first GPU board card is provided with a GPU chip and a first high-speed connector, the front side of the second GPU board card is provided with a GPU chip, and the back side of the second GPU board card is provided with a second high-speed connector used for being connected with the first high-speed connector. The first GPU board card and the second GPU board card are connected through the first high-speed connector and the second high-speed connector, and therefore interconnection among GPU chips is achieved. First GPU integrated circuit board and second GPU integrated circuit board parallel arrangement, install two GPU chips on the front of first GPU integrated circuit board side by side, be equipped with four female first high-speed connectors on the front of first GPU integrated circuit board side by side, two GPU chips are installed on the front of second GPU integrated circuit board side by side, the back of second GPU integrated circuit board side by side be equipped with four be used for respectively with four female first high-speed connectors's public first high-speed connector butt joint. The two GPU chips are arranged on the first GPU board card and the second GPU board card respectively, the width of the GPU board card is reduced, the GPU board card can be vertically plugged in a lower case, the space occupied by the two GPU board cards is reduced due to the fact that the first GPU board card and the second GPU board card are arranged in parallel, the first GPU board card is connected with the second GPU board card through the four female high-speed connectors and the four male high-speed connectors, and the GPU chips are interconnected.

The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The utility model provides a GPU interconnection structure, its characterized in that includes first module shell (18) and second module shell (19) of superpose, is equipped with first GPU integrated circuit board (1) in first module shell (18), is equipped with second GPU integrated circuit board (2) in the second module shell, and first GPU integrated circuit board (1) and second GPU integrated circuit board (2) are connected through high-speed connector, are connected with apron (3) on the second module shell.

2. A GPU interconnection structure according to claim 1, wherein a GPU chip (4) and a first high-speed connector are provided on the front side of the first GPU board card (1), a GPU chip (4) is provided on the front side of the second GPU board card (2), and a second high-speed connector for connecting with the first high-speed connector is provided on the back side of the second GPU board card (2).

3. The GPU interconnection structure of claim 2, wherein the first GPU board card (1) and the second GPU board card (2) are arranged in parallel, two GPU chips (4) are arranged on the front face of the first GPU board card (1) in parallel, four female high-speed connectors (5) are arranged on the front face of the first GPU board card (1) in parallel, two GPU chips (4) are arranged on the front face of the second GPU board card (2) in parallel, and four male high-speed connectors (6) used for being respectively in butt joint with the four female high-speed connectors (5) are arranged on the back face of the second GPU board card (2) in parallel.

4. A GPU interconnection structure according to claim 3, wherein the second module housing (19) comprises a second bottom plate (7) parallel to the second GPU board card (2), two opposite sides of the second bottom plate (7) are fixedly connected with a second side plate (8) perpendicular to the second bottom plate (7) respectively, one side of the second bottom plate (7) is provided with a second front baffle (9) perpendicular to both the second bottom plate (7) and the second side plate (8), and the second bottom plate (7) is provided with an opening (10) for the male high-speed connector (6) to pass through.

5. A GPU interconnection structure according to claim 4, wherein the first module housing (18) comprises a first bottom plate (11) parallel to the first GPU card (1), two opposite sides of the first bottom plate (11) are fixedly connected with first side plates (12) perpendicular to the first bottom plate (11), and a first front baffle (13) perpendicular to both the first bottom plate (11) and the first side plates (12) is arranged on one side of the first bottom plate (11).

6. A GPU interconnection structure according to claim 5, wherein a first mounting bar (15) which is located in the same plane as the second side plate (8) is fixedly connected to the second bottom plate (7), the first mounting bar (15) and the second side plate (8) are respectively located on two sides of the second bottom plate (7), and the first side plate (12) is connected with the first mounting bar (15) through screws.

7. A GPU interconnection structure according to claim 4, wherein the cover plate (3) is parallel to the second base plate (7), second mounting bars (16) parallel to the second side plate (8) are fixedly connected to two opposite sides of the cover plate (3), and the second side plate (8) is connected with the second mounting bars (16) through screws.

8. A GPU interconnection structure according to claim 5, wherein through holes are densely distributed in the first front baffle (13) and the second front baffle (9), operating handles (17) are arranged on the first front baffle (13) and the second front baffle (9), and pull-out aids (14) are arranged on the first bottom plate (11) and the second bottom plate (7).