TW201906520A - Server system - Google Patents

Abstract

A server system includes a frame, a host server, a GPU board, a first fan module and a power supply unit. The frame has a host server accommodating space and a GPU module accommodating space. The GPU module accommodating space is located on the top of the host server accommodating space. The host server is removable disposed in the host server accommodating space. The GPU board is located in the GPU module accommodating space, and is located on the top of the host server. The first fan module is located in the GPU module accommodating space, is located on a side of the GPU board and on the top of the host server. The power supply unit is located in the GPU module accommodating space, is located on the top of another side of the GPU board, and is configured to provide electrical power to the first fan module and the GPU board.

Description

Server architecture

The present invention relates to a server architecture, and more particularly to a server architecture including an image processor.

Artificial Intelligence (AI) refers to letting computers learn from a large amount of data to simulate the wisdom of human beings. Among them, deep learning is the fastest growing field of artificial intelligence, and the purpose is to make the computer closer to the ability of humans to observe complex situations, learning, reaction and other independent thinking. Because the Graphic Processing Unit (GPU) is better at parallel workloads than the single-threaded Central Processing Unit (CPU), and the GPU is designed to perform complex mathematical and geometric calculations, Therefore, the GPU is more suitable than the CPU for the needs of a large number of matrix operations in deep learning, which helps accelerate the ability of deep learning. It is true that today's deep learning solutions rely on GPUs.

As a result, server-related vendors are actively developing server models that can be equipped with GPUs to meet market demands.

In view of this, the present invention provides a server architecture suitable for carrying an image processor.

A server architecture according to the present invention includes a server rack, a host server, a graphics processor card, a first fan module, and a power supply unit. The server rack has a host housing space and a graphics processing module space. The graphics processing module space is located above the host housing space. The host server can be detached from the host accommodating space. The graphics processor card is placed in the graphics processing module space and located on the host server. The first fan module is housed in the graphics processing module space, located on one side of the graphics processor card and above the host server. The power supply unit is housed in the graphics processing module space and is located above the other side of the graphics processor card to provide power to the first fan module and the graphics processor card.

In the server architecture disclosed by the present invention, the server rack divides the space into a host housing space and a graphics processing module space, and the planning is simple and clear. Moreover, the graphics processing module space is located above the host housing space, so that the graphics processor and the host server do not interfere with each other, which not only facilitates the upgrade and maintenance operations, but also configures the upper and lower layers. The graphics processing module space achieves maximum width utilization, which in turn allows for the maximum number of graphics processors to enhance the overall depth learning capability.

In addition, the host housing space accommodates the host server, with the power supply unit and the first fan module, which helps the server frame to form an independently operable servo system.

The above description of the disclosure and the following embodiments are intended to illustrate and explain the spirit and principles of the invention, and to provide further explanation of the scope of the invention.

The detailed features and advantages of the present invention are described in detail in the embodiments of the present invention. The related objects and advantages of the present invention will be readily understood by those skilled in the art. The following examples are intended to describe the present invention in further detail, but are not intended to limit the scope of the invention.

In addition, the embodiments of the present invention are disclosed in the following drawings, and for the sake of clarity, the details of the invention are described in the following description. However, it should be understood that these practical details are not intended to limit the invention. In addition, for the sake of simplicity of the drawings, some conventional structures and components will be illustrated in a simplified schematic manner, and even some of the drawings omits the structure of cables (cables, or cables). In order to keep the picture clean and tidy, declare it first.

Furthermore, unless otherwise defined, all terms used herein, including technical and scientific terms, have their ordinary meaning, and their meaning is understood by those skilled in the art. Furthermore, the definition of the above vocabulary should be interpreted in the present specification as having the same meaning as the technical field related to the present invention. Unless specifically defined, these terms are not to be construed as too idealistic or formal.

1 to 3, FIG. 1 is a perspective exploded view of a server architecture according to an embodiment of the present invention, and FIG. 2 is an exploded perspective view of the upper cover removed according to the server architecture of FIG. A side cross-sectional view of the server architecture of 1.

This embodiment provides a server architecture 1 including a server rack 10, a host server 20, a graphics processor board 30, a first fan module 40, and a power supply unit 50.

The server rack 10 includes an upper cover 100, a bottom plate 110, two side plates 120, a partition plate 130, two cross beams 140, and a back plate 150. Among them, it should be stated that the upper cover 100 is omitted in FIG. 2 in order to keep the image in the drawing appropriately sized for viewing.

Specifically, the two side plates 120 are respectively disposed on opposite sides of the bottom plate 110. The partition 130 is connected between the two side plates 120 and at a distance from the bottom plate 110. The two beams 140 are connected to one side of the two side plates 120 away from the bottom plate 110, so as to strengthen the structural strength of the whole server frame 10. However, the beam is optional, and the present invention is not limited thereto, and the present invention is not limited by the number of beams. limit. The back plate 150 is disposed at a rear end of the bottom plate 110 and connected between the two side plates 120. The upper cover 100 may be disposed on one side of the two side plates 120 away from the bottom plate 110.

It should be noted that, in this embodiment, the partition 120 separates the space between the side panel 120 and the bottom panel 110 from a host housing space S1 and a graphics processing module space S2. More specifically, the main body accommodating space S1 is formed between the two side plates 120 and the lower surface of the bottom plate 110, and the graphic processing module space S2 is formed between the two side plates 120 and the upper surface of the bottom plate 110, and thus is in a relative position. The graphics processing module space S2 is located above the host housing space S1.

Moreover, in this embodiment and other embodiments, the server rack 10 is a 4U rack, wherein the height h1 of the host housing space S1 is 1U, and the height h2 of the graphics processing module space S2 is 3U.

The host server 20 can be detached from the host accommodating space S1. Therefore, it can be understood that the host server 20 is a server with a height of 1U. Specifically, the host server 20 includes a carrier 200, a motherboard 210, a plurality of computing modules 220, a plurality of hard disks 230, a second fan module 240, an electrical port 250, and a plurality of first Input and output connector 260. The motherboard 210 is disposed on the carrier 200, and the hard disk 230 and the second fan module 240 are disposed on the carrier 200 and connected to the motherboard 210. The computing module 220, the electrical connector 250, and the first input/output connector 260 are disposed on the motherboard 210. Thereby, the host server 20 can be made a self host of the server architecture 1 itself, but the invention is not limited thereto. For example, in other embodiments, host server 20 may also be a other host of other server architectures. In addition, the present invention is not limited to the number of the computing module 220, the hard disk 230, and the first input/output connector 260 in the host server 20, and the number of the components may be increased or decreased according to actual needs.

In this embodiment, the server structure 1 further includes a slide rail group 60 disposed in the main body accommodating space S1 and disposed on the two side panels 120 of the server rack 10. The two sides of the carrier disc 200 are provided with corresponding slide rail groups. The inner rail structure allows the carrier tray 200 to be slidably disposed between the rail sets 60 such that the host server 20 is detachably disposed on the server rack 10.

The graphics processor board 30 is received in the graphics processing module space S2 and disposed on the partition 130. Further, the graphics processor board 30 has a plurality of graphics processor sockets 310 and a plurality of expansion card slots 320. Graphics processor socket 310 can be used to plug in a graphics processor (also known as a graphics processing card or GPU) 311. The graphics processor 311 described herein is an electronic component for performing image computing operations. In this embodiment, the graphics processor card 30 has eight graphics processor sockets 310 for eight graphics processors 311 to be plugged in, which helps to improve the depth of the server architecture 1 learning. It is to be noted, however, that the present invention is not limited to the number of graphics processor sockets 310. The expansion card slot 320 can be used to connect the expansion card 321 . The expansion card 321 described herein is, for example, a separate network interface card (NIC) or a connector supporting the Mini SAS external connection (SFF-8644). Such as expansion cards. The expansion card 321 has a second input/output connector 3211 thereon. However, it is to be noted that the present invention is not limited to the number of expansion card slots 320 or the types of expansion cards that can be supported. It will be appreciated that the server rack 10 provides 3 U of space to accommodate various plug-in electronics (such as graphics processor 311 and stand-alone network expansion card 321) on graphics processor card 30.

The first fan module 40 is received in the graphics processing module space S2 and disposed on the partition 130 away from one side of the backplane 150. Further, the first fan module 40 includes five sub-fans 410 that can be used to dissipate heat from the electronic components inside the graphics processing module space S2. It is to be noted, however, that the present invention is not limited to the number of sub-fans 410. For example, in other embodiments, the number of sub-fans 410 may be 5 or less or 5 or more.

The power supply unit 50 is housed in the graphics processing module space S2 and located on one side of the side panel 120 near the backplane 150. Specifically, in the embodiment, the server frame 10 further includes a pair of support frames 160 respectively disposed on the side plates 120. The support frame 160 is kept at a distance from the partition plate 130, and is located at one of the side plates 120 near the back plate 150. On the side, the power supply unit 50 is disposed on the support frame 160 to be positioned above the partition 130.

In addition, in the embodiment, the power supply unit 50 can be electrically connected to the foregoing electronic components (such as the graphics processor board 30 and the first fan module 40) in the server architecture 1 via internal wires (not shown). To provide the power required to operate the electronic components.

4, which is the same as FIG. 3, which is a side cross-sectional view of the server architecture according to FIG. 1, but FIG. 4 mainly illustrates the internal configuration of the server architecture 1 by the configuration of the aforementioned components in the server rack 10. space plan.

As shown in FIG. 4, the area Z1 in which the first fan module 40 is disposed is located on the front side of the server rack 10; the area Z2 in which the graphics processor board 30 is disposed is located inside the area Z1; and the area Z3 in which the graphics processor 311 is disposed is located. The area Z4 is located above the area Z1 and is located inside the area Z1; the area Z4 where the independent network expansion card 321 is disposed is located above the area Z2, and the area Z3 is located on the side of the server rack 10 away from the side of the area Z1; the power supply unit 50 is disposed The zone Z5 is located on the rear side of the server rack 10 and above the zone Z4. The aforementioned areas Z1 to Z5 are located in the graphics processing module space S2. The host server 20 is disposed in the area Z6 (that is, the host housing space S1), and is located below the area Z1 and the area Z2, that is, at the bottom of the server rack 10.

From this point of view, the internal planning of the entire server architecture 1 is simple and clear, the components of each area will not interfere with each other in upgrading, replacement or maintenance, which will facilitate the subsequent performance expansion and upgrade operations.

In addition, in this embodiment, the components such as the host server 20, the first fan module 40, and the power supply unit 50 are assembled to the server rack 10 in a screw-less, tool-less manner. Helps improve the efficiency of repair and replacement operations.

Next, please refer to FIG. 5, which is a rear view of the server architecture according to FIG. 1.

In the present embodiment, the power supply unit 50 and the second input/output connector 3211 are both located in the area of the graphics processing module space S2 near the rear side of the server rack 10, and the host server is provided by the foregoing simple area planning. The electrical port 250 of 20 and the first input-output connector 260 are located in the host housing space S1, but are also located in the area near the rear side of the server rack 10. Therefore, the host server 20 can be connected to the upper power supply unit 50 via an external trace (not shown) from the rear end of the entire server architecture 1, or can be connected to another external line (not shown). The second input/output connector 3211 of the expansion card 321 is connected for signal exchange, but the invention is not limited thereto. The host server 20 can be the computing host of the server architecture 1 itself or the computing host of other server architectures. Therefore, in other embodiments, the host server 20 and the graphics processor 311 in the same server architecture 1 can be used. Only maintain coexistence relationships without communication connections. In addition, the server architecture 1 is also adapted to connect an external power source (not shown) so that the host server 20 can obtain electrical energy by electrically connecting the external power source.

According to the server architecture of the above embodiment, the server rack separates the space into the host accommodating space and the graphics processing module space, and the planning is simple and clear. Moreover, the graphics processing module space is located above the host housing space, so that the graphics processor and the host server do not interfere with each other, which not only facilitates the upgrade and maintenance operations, but also configures the upper and lower layers. The graphics processing module space achieves maximum width utilization, which in turn allows for the maximum number of graphics processors to enhance the overall depth learning capability.

In addition, the host housing space accommodates the host server, with the power supply unit and the first fan module, which helps the server frame to form an independently operable servo system.

In addition, components such as the host server, the first fan module, and the power supply unit are assembled in the servo rack without screws and in a tool-less manner, which helps to improve the efficiency of maintenance and replacement operations.

Although the present invention has been disclosed above in the foregoing embodiments, it is not intended to limit the invention. It is within the scope of the invention to be modified and modified without departing from the spirit and scope of the invention. Please refer to the attached patent application for the scope of protection defined by the present invention.

1‧‧‧Server Architecture

10‧‧‧Server Rack

20‧‧‧Host server

30‧‧‧Graphical processor board

40‧‧‧First fan module

50‧‧‧Power supply unit

60‧‧‧slide group

100‧‧‧Upper cover

110‧‧‧floor

120‧‧‧ side panels

130‧‧‧Baffle

140‧‧‧ beams

150‧‧‧ Backplane

160‧‧‧Support frame

200‧‧‧ Carrying tray

210‧‧‧ motherboard

220‧‧‧ Computing Module

230‧‧‧ Hard disk

240‧‧‧Second fan module

250‧‧‧Electrical connection埠

260‧‧‧First input and output connector

310‧‧‧Graphic processor socket

311‧‧‧graphic processor

320‧‧‧Expansion card slot

321‧‧‧Independent network expansion card

410‧‧‧Child fan

3211‧‧‧Second input and output connector

H1, h2‧‧‧ height

S1‧‧‧Host accommodating space

S2‧‧‧Graphic Processing Module Space

Z1~Z6‧‧‧ area

FIG. 1 is a perspective exploded view of a server architecture according to an embodiment of the invention. 2 is an exploded perspective view of the upper cover removed according to the server architecture of FIG. 1. 3 is a side cross-sectional view of the server architecture in accordance with FIG. 1. 4 is a side cross-sectional view of the server architecture in accordance with FIG. 1. Figure 5 is a rear elevational view of the server architecture in accordance with Figure 1.

Claims (10)

A server architecture includes: a server rack having a host accommodating space and a graphics processing module space, the graphics processing module space is located above the host accommodating space; and a host server is detachable The device is placed in the host processing space; a graphics processor card is placed in the graphics processing module space and located on the host server; a first fan module is received in the graphics processing module a space on one side of the graphics processor card and above the host server; and a power supply unit housed in the graphics processing module space and located on the other side of the graphics processor card Above, the power is supplied to the first fan module and the graphics processor card. The server architecture of claim 1, wherein the graphics processor card has a plurality of graphics processor sockets and at least one expansion card slot, the graphics processor sockets being closer to the expansion card slots The first fan module is located below the power supply unit. The server architecture of claim 1, wherein the host housing space has at least one first input/output connector, the graphics processing module space has at least one second input/output connector, the first input output connector The second input and output connector is electrically connected. The server architecture of claim 1, wherein the server rack comprises a bottom plate, two side plates and a partition, the two side plates are respectively disposed on opposite sides of the bottom plate, and the partition plate is connected to the two side plates. And maintaining a distance from the bottom plate, the partition partitions the main board accommodating space and the graphic processing module space between the side panel and the bottom board. The server architecture of claim 4, wherein the server rack further comprises at least one beam connected between the two side panels and located in the graphics processing module space. The server architecture of claim 4, wherein the server rack further comprises an upper cover disposed on a side of the two side plates away from the bottom plate. The server architecture of claim 1, wherein the host server comprises a motherboard, at least one computing module, at least one hard disk, a second fan module, and an electrical port, the at least one computing module The at least one hard disk, the second fan module and the electrical connection port are connected to the motherboard, wherein the electrical connection port is located below the power supply unit. The server architecture of claim 7, further comprising a slide group disposed on the server rack, the host server further comprising a carrier, the motherboard, the at least one computing module, the at least A hard disk, the second fan module and the electrical connection port are disposed on the carrier disk, and the carrier disk is detachably disposed on the server frame via the track group. The server architecture of claim 7 further includes an external trace located outside the server rack, and the electrical connection of the host server is electrically connected to the power supply unit via the external trace. The electronic device of claim 1, further comprising an internal trace, the graphics processor card being electrically connected to the power supply unit via the internal trace.