CN209103234U - A kind of host for supporting multiple GPU by 2U cabinet - Google Patents
A kind of host for supporting multiple GPU by 2U cabinet Download PDFInfo
- Publication number
- CN209103234U CN209103234U CN201920061726.4U CN201920061726U CN209103234U CN 209103234 U CN209103234 U CN 209103234U CN 201920061726 U CN201920061726 U CN 201920061726U CN 209103234 U CN209103234 U CN 209103234U
- Authority
- CN
- China
- Prior art keywords
- gpu
- cabinet
- expansion card
- fixed
- mainboard
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000007787 solid Substances 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000000034 method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 230000017525 heat dissipation Effects 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 229910052782 aluminium Inorganic materials 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
Landscapes
- Cooling Or The Like Of Electrical Apparatus (AREA)
Abstract
The utility model discloses a kind of hosts that multiple GPU are supported by 2U cabinet, comprising: 2U cabinet；Multiple GPU, multiple GPU are placed side by side on the side of 2U cabinet；Multiple expansion cards, the quantity of expansion card and the quantity of GPU are identical and connect respectively with multiple GPU；And mainboard, mainboard are connected to multiple expansion cards and are provided with multiple CPU, multiple CPU are separately connected corresponding expansion card by PCIex24 slot on mainboard or high-speed cable connector to be connected respectively to multiple GPU.Host disclosed by the utility model including multiple GPU can make full use of the confined space of 2U cabinet, rationally realize the interconnection of CPU and GPU, greatly improve the computing capability of server.
Description
Technical field
The utility model relates to cabinet fields, more specifically, particularly relating to a kind of master for supporting multiple GPU by 2U cabinet
Machine.
Background technique
GPU (Graphics Processing Unit, graphics processor) has high parallel organization, in processing figure number
Possess efficiency more higher than CPU according to complicated algorithm aspect, and GPU and CPU collocation use can be significantly improved into server
Computing capability.With the development of big data, requirement of the user to server computational power is continuously improved, therefore design can take
The multipath server for carrying multiple GPU is imperative.
But GPU volume is big, power consumption is high, and the interconnection between CPU is complicated.Therefore, the multichannel of multiple GPU is carried
Server design to make full use of server machine box space, make rational planning for CPU and GPU interconnection cabling, setting effectively heat dissipation strategy
There is high requirement.And 2U standard PC case space is limited, it is difficult to while accommodating multiple GPU.
Utility model content
In view of this, the purpose of the utility model embodiment is to propose a kind of master for supporting multiple GPU by 2U cabinet
Machine can make full use of the confined space of 2U cabinet, rationally realize the interconnection of CPU and GPU, and be provided with effective heat dissipation dress
It sets, greatly improves the computing capability of server.
Based on above-mentioned purpose, the utility model embodiment provides a kind of host that multiple GPU are supported by 2U cabinet, packet
It includes: 2U cabinet；Multiple GPU, the multiple GPU are placed side by side on the side of 2U cabinet；Multiple expansion cards, the number of the expansion card
It measures identical as the quantity of the GPU and is connect respectively with multiple GPU；And mainboard, the mainboard are connected to multiple expansion cards simultaneously
And multiple CPU are provided with, multiple CPU are separately connected corresponding by PCIex24 slot on mainboard or high-speed cable connector
Expansion card is to be connected respectively to multiple GPU.
In some embodiments, expansion card includes PCIex16 slot, for connecting GPU.
In some embodiments, part expansion card further includes PCIex8 slot, for connecting the PCIe device of extension.
In some embodiments, cabinet further includes the fixed bracket of GPU, and the fixed bracket of the GPU is fixed on cabinet wall.
In some embodiments, the first end of GPU is fixed on the side of cabinet, and second end is fixed on the GPU and fixes
On bracket.
In some embodiments, the first end of GPU is fixed by screws in the side of cabinet, and second end is solid by screw
It is scheduled on the fixed bracket of the GPU.
In some embodiments, mainboard includes multiple power-feed connectors, the multiple power-feed connector with multiple GPU pairs
It should connect.
In some embodiments, cabinet further includes radiator, and the radiator includes multiple fans.
In some embodiments, cabinet further includes blower tray, and the blower tray is fixed on cabinet wall.
In some embodiments, multiple fans are fixed on the blower tray.
The utility model has following advantageous effects: can make full use of the confined space of 2U cabinet, rationally realize
The interconnection of CPU and GPU, and it is provided with effective radiator, greatly improve the computing capability of server.
Detailed description of the invention
In order to illustrate the embodiment of the utility model or the technical proposal in the existing technology more clearly, below will be to embodiment
Or attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
It is some embodiments of the utility model, for those of ordinary skill in the art, in the premise not made the creative labor
Under, it can also be obtained according to these attached drawings other embodiments.
Fig. 1 is the structural schematic diagram of the embodiment of the host provided by the utility model that multiple GPU are supported by 2U cabinet；
Fig. 2 is the schematic diagram of the embodiment of the host provided by the utility model that multiple GPU are supported by 2U cabinet.
Specific embodiment
For the purpose of this utility model, technical solution and advantage is more clearly understood, below in conjunction with specific embodiment, and
Referring to attached drawing, the utility model embodiment is further described.
It should be noted that all statements for using " first " and " second " are for area in the utility model embodiment
Be divided to two non-equal entities of same names or non-equal parameter, it is seen that " first " " second " only for statement convenience,
It should not be construed as the restriction to the utility model embodiment, subsequent embodiment no longer illustrates this one by one.
U is a kind of unit for indicating server external dimensions, is the abbreviation of unit.The size of predetermined server be in order to
Make server keep size appropriate in order to be placed in the rack of irony or aluminum, there is the spiral shell of fixed server in rack
Hole, it is specified that be dimensioned such that server is corresponding with the screw hole in rack, with facilitate installation each server.Therefore, standard
The size of 2U cabinet it has been determined that and according to the installation method of existing GPU, two can only be installed in a general 2U standard PC case
A GPU.And in actual operation, it generally requires simultaneously with multiple GPU.
Based on above-mentioned purpose, the utility model embodiment proposes a kind of host that multiple GPU are supported by 2U cabinet, packet
It includes: 2U cabinet；Multiple GPU, the multiple GPU are placed side by side on the side of 2U cabinet；Multiple expansion cards, the number of the expansion card
It measures identical as the quantity of the GPU and is connect respectively with multiple GPU；And mainboard, the mainboard are connected to multiple expansion cards simultaneously
And multiple CPU are provided with, multiple CPU are separately connected corresponding by PCIex24 slot on mainboard or high-speed cable connector
Expansion card is to be connected respectively to multiple GPU.
Shown in fig. 1 is the structure of the embodiment of the host provided by the utility model that multiple GPU are supported by 2U cabinet
Schematic diagram.It is GPU1, GPU2, GPU3 and GPU4 respectively, above-mentioned 4 GPU are simultaneously discharged as shown in Figure 1, including 4 GPU in host
Set the rear side in 2U cabinet；Include a mainboard in 2U cabinet, include 4 CPU on mainboard, be respectively CPU0, CPU1, CPU2 and
The specific connection type of CPU3, GPU and CPU please refer to Fig. 2.
2U cabinet further includes the fixed bracket of GPU, and the fixed bracket of GPU is fixed on cabinet wall.For the stability of device,
The fixed bracket both sides GPU are each attached on cabinet wall, further, support bracket fastened three side GPU can be each attached to cabinet
On wall.
The first end of GPU is fixed on the side of cabinet, and second end is fixed on the fixed bracket of GPU.As shown in Figure 1, GPU
Both ends are separately fixed on the rear side and the fixed bracket of GPU of cabinet.According to preferred embodiment, the first end of GPU can pass through
Screw is fixed on the side of cabinet, and second end can be fixed by screws on the fixed bracket of the GPU.As shown in Figure 1, GPU
One end be fixed on cabinet rear wall by a fixed screw, the other end is fixed on the fixed bracket of GPU by two screws.
Host further includes radiator, and the radiator includes multiple fans.As shown in Figure 1, radiator includes 6
Fan.2U cabinet further includes blower tray, and blower tray is fixed on cabinet wall.For the stability of device, blower tray
Both sides are each attached on cabinet wall, and further, three sides of blower tray are each attached on cabinet wall.Multiple fans can be consolidated
It is scheduled on the blower tray.
Fig. 2 shows be the host provided by the utility model that multiple GPU are supported by 2U cabinet embodiment principle
Figure.As shown in Fig. 2, CPU is connected with GPU by four expansion cards, four expansion cards are respectively expansion card one, expansion card two, expand
Zhan Kasan and expansion card four.Specifically, GPU1 and CPU0 is connected by expansion card one, GPU2 and CPU1 are connected by expansion card two
It connects, GPU3 and CPU2 are connected by expansion card three, and GPU4 and CPU3 are connected by expansion card three.Four expansion cards include
PCIex16 slot can be used for connecting GPU.For example, expansion card one is connect by PCIex16 slot with GPU1.Part expansion card packet
PCIex8 slot is included, for connecting the PCIe device of extension.For example, the expansion card one, two and three in Fig. 2 is inserted including PCIex8
Slot can be used for the PCIe device of other x8/x4/x1 bandwidth of grafting.
CPU connects corresponding expansion card by PCIex24 slot or high-speed cable connector.For example, the expansion card in Fig. 2
One connects CPU0 by PCIex24 slot 11, and expansion card four is connected by the first high-speed cable connector 21 and the second high-speed cable
It connects device 22 and connects CPU1.
Mainboard includes multiple power-feed connectors, and the multiple power-feed connector is correspondingly connected with multiple GPU.For example, in Fig. 2
The first power-feed connector 31 connect with GPU1, for give GPU1 power supply.
It is illustrated by taking 4 GPU as an example below:
The second high-speed cable connector 22, PCIe x24 slot 11,12,13, CPU0-4 and 4 confessions are provided on mainboard
Electric connector 31,32,33 and 34.Second high-speed cable connector 22, PCIe x24 slot 11,12,13 are believed by PCIe respectively
Number with corresponding CPU communicate, 4 power-feed connectors respectively with corresponding GPU be connected, for corresponding GPU power.
Expansion card one, expansion card two and expansion card three are respectively equipped with a PCIe x16 slot and a PCIe x8 slot.
PCIe x16 slot is used to grafting GPU, and PCIe x8 slot can be with the PCIe device of other x8/x4/x1 bandwidth of grafting.Expansion card
Four are equipped with a PCIe x16 slot, are used to grafting GPU.
Expansion card one, expansion card two and expansion card three are equipped with PCIe x24 plug, and expansion card four is equipped with the first high-speed line
Cable connector 21.Expansion card one is plugged into the PCIe x24 slot 11 of mainboard by PCIe x24 plug thereon；Expansion card
Two are plugged into the PCIe x24 slot 12 of mainboard by PCIe x24 plug thereon；Expansion card three passes through PCIe thereon
X24 plug is plugged into the PCIe x24 slot 13 of mainboard；On the first high-speed cable connector 21 and mainboard on expansion card four
The second high-speed cable connector 22 be connected.
PCIe x16 plug on GPU1 is plugged into the PCIe x16 slot of expansion card one；PCIe x16 on GPU2 is inserted
Head is plugged into the PCIe x16 slot of expansion card two；PCIex16 plug on GPU3 is plugged into the PCIe x16 of expansion card three
In slot；PCIe x16 plug on GPU4 is plugged into the PCIe x16 slot of expansion card four.As a result, required for GPU
The signals such as PCIe, I2C, WAKE (wake-up signal), RST (reset signal) are via the PCIe x24 slot or the second high speed on mainboard
Wire and cable connector 22 is connected on expansion card, then via GPU plug conductive to GPU.
The utility model provides a kind of host that multiple GPU are supported by 2U cabinet, can make full use of 2U cabinet
The confined space rationally realizes the interconnection of CPU and GPU, and is provided with effective radiator, greatly improves the meter of server
Calculation ability.
It is exemplary embodiment disclosed by the utility model above, it should be noted that being limited without departing substantially from claim
The utility model embodiment scope of disclosure under the premise of, may be many modifications and modify.According to public affairs described herein
The function, step and/or movement for opening the claim to a method of embodiment are not required to the execution of any particular order.In addition, although originally
Element disclosed in utility model embodiment can be described or be required in the form of individual, but be unless explicitly limited odd number, can also be with
It is interpreted as multiple.
It should be understood that it is used in the present context, unless the context clearly supports exceptions, singular " one
It is a " it is intended to also include plural form.It is to be further understood that "and/or" used herein refers to including one or one
Any and all possible combinations of a above project listed in association.
It is for illustration only that above-mentioned the utility model embodiment discloses embodiment sequence number, does not represent the advantages or disadvantages of the embodiments.
It should be understood by those ordinary skilled in the art that: the discussion of any of the above embodiment is exemplary only, not
It is intended to imply that the utility model embodiment scope of disclosure (including claim) is limited to these examples；In the utility model reality
Under the thinking for applying example, it can also be combined between the technical characteristic in above embodiments or different embodiments, and exist such as
On the utility model embodiment different aspect many other variations, in order to it is concise they do not provided in details.Cause
This, all within the spirit and principle of the utility model embodiment, any omission, modification, equivalent replacement, improvement for being made etc.,
It should be included within the protection scope of the utility model embodiment.
Claims (10)
1. a kind of host for supporting multiple GPU by 2U cabinet characterized by comprising
2U cabinet；
Multiple GPU, the multiple GPU are placed side by side on the side of the 2U cabinet；
Multiple expansion cards, the quantity of the expansion card is identical as the quantity of the GPU and connect respectively with the multiple GPU；With
And
Mainboard, the mainboard are connected to the multiple expansion card and are provided with multiple CPU, and the multiple CPU passes through on mainboard
PCIex24 slot or high-speed cable connector be separately connected corresponding expansion card to be connected respectively to the multiple GPU.
2. host according to claim 1, which is characterized in that the expansion card includes PCIex16 slot, for connecting
GPU。
3. host according to claim 2, which is characterized in that part expansion card further includes PCIex8 slot, for connecting
The PCIe device of extension.
4. host according to claim 1, which is characterized in that further include the fixed bracket of GPU, the fixed bracket of the GPU is solid
It is scheduled on cabinet wall.
5. host according to claim 4, which is characterized in that the first end of the GPU is fixed on the side of cabinet, and second
End is fixed on the fixed bracket of the GPU.
6. host according to claim 5, which is characterized in that the first end of the GPU is fixed by screws in cabinet
Side, second end are fixed by screws on the fixed bracket of the GPU.
7. host according to claim 1, which is characterized in that the mainboard includes multiple power-feed connectors, the multiple
Power-feed connector is correspondingly connected with multiple GPU.
8. host according to claim 1, which is characterized in that further include radiator, the radiator includes multiple
Fan.
9. host according to claim 8, which is characterized in that further include blower tray, the blower tray is fixed on machine
On tank wall.
10. host according to claim 9, which is characterized in that the multiple fan is fixed on the blower tray.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201920061726.4U CN209103234U (en) | 2019-01-15 | 2019-01-15 | A kind of host for supporting multiple GPU by 2U cabinet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201920061726.4U CN209103234U (en) | 2019-01-15 | 2019-01-15 | A kind of host for supporting multiple GPU by 2U cabinet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN209103234U true CN209103234U (en) | 2019-07-12 |
Family
ID=67164049
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201920061726.4U Active CN209103234U (en) | 2019-01-15 | 2019-01-15 | A kind of host for supporting multiple GPU by 2U cabinet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN209103234U (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111273747A (en) * | 2020-02-29 | 2020-06-12 | 苏州浪潮智能科技有限公司 | System for 2U case rear window multi-PCIE external plug-in card |
-
2019
- 2019-01-15 CN CN201920061726.4U patent/CN209103234U/en active Active
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111273747A (en) * | 2020-02-29 | 2020-06-12 | 苏州浪潮智能科技有限公司 | System for 2U case rear window multi-PCIE external plug-in card |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8498121B2 (en) | Printed circuit assembly with determination of storage configuration based on installed paddle board | |
| CN211427190U (en) | Server circuit and mainboard based on Feiteng treater 2000+ | |
| JP2013541742A (en) | Dynamic multilink editing partitioning system and method | |
| CN213276460U (en) | Double-circuit server mainboard and server | |
| CN113835487B (en) | System and method for realizing memory pool expansion of high-density server | |
| US10445276B2 (en) | Server system having a hot plug motherboard | |
| US20170215296A1 (en) | Multi-bay apparatus | |
| CN209103234U (en) | A kind of host for supporting multiple GPU by 2U cabinet | |
| CN109033009A (en) | It is a kind of to support general and machine cabinet type server circuit board and system | |
| CN213276461U (en) | Double-circuit server mainboard and server | |
| CN108334172B (en) | Case for computer | |
| US10956353B1 (en) | Connector configurations for printed circuit boards | |
| CN209895219U (en) | Board card based on multifunctional Slimline connector | |
| GB2542909A (en) | Multi-server system interconnect | |
| CN215932518U (en) | Cloud computing ultra-fusion all-in-one machine equipment | |
| CN217157208U (en) | High-performance server mainboard and server | |
| CN107590097B (en) | Server IO equipment extension device | |
| CN216145157U (en) | Modular-design universal edge computing server | |
| CN216352292U (en) | Server mainboard and server | |
| CN209859073U (en) | Host configuration framework for intelligent cabinet | |
| CN208781131U (en) | A kind of cabinet | |
| CN209640782U (en) | A kind of PCIE expansion card of NVMe hard-disk interface | |
| CN209879377U (en) | Calculation board card | |
| CN208283857U (en) | A kind of cabinet | |
| CN208314678U (en) | Computer housing mainboard bracket |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant |