CN108762818A - A kind of optimization design server and maintaining method - Google Patents

Abstract

The invention discloses an optimally designed server and a maintenance method. The optimizedly designed server is arranged in layers according to the functional modules of the traditional server system and the GPU server system, and the number of layers is multi-layered to realize the uninterrupted operation of the whole machine. It can be stretched and stretched freely, and one side of each layer except the top layer is equipped with a tank chain for pulling the traditional server system or GPU server system, which ensures the coherence of the various layers and avoids the When the components are lost, a fixed buckle is provided on the other side of each layer, which strengthens the structural firmness of the server after the optimized design.

Description

An optimized design server and maintenance method

technical field

The invention relates to the field of artificial intelligence servers, in particular to an optimized design server and a maintenance method.

Background technique

Artificial intelligence is playing an increasingly important role in the online services of the current world. Within companies such as Google, Facebook, Microsoft, and Baidu, GPUs play a huge role in the field of "deep learning". GPU is a large-scale SIMD collection with a large number of vector computing systems. When performing stream computing with repetitive functions, it can process a large amount of trivial information in parallel. It only needs to set the function data and keep entering it. The power of the CPU is that it is excellent in the face of control-intensive operations, but for data-intensive operations, such as what color should be displayed by each pixel on the screen, it is relatively limited by the operating mechanism of the CPU and the pressure will be great. .

With the vigorous development of artificial intelligence, GPUs are widely used in large numbers. For large-scale data centers, a large number of pooled servers have begun to be deployed. This type of traditional server structure does not support the thermal operation and maintenance of GPUs. Once the server needs to be repaired or To upgrade the GPU, it is necessary to power off the server as a whole for related operations, which is not as convenient as replacing the hard disk, fan, and power supply. This will cause the entire service to stop operating and bring great challenges to the entire system. Therefore, design and development are urgently needed A server that can be hot-operated and easily upgraded with a GPU meets market demand and meets the entire service.

Contents of the invention

In order to solve the above problems, the present invention provides an optimally designed server. The optimally designed server GPU supports thermal operation and maintenance, and can realize the maintenance of a single component. The maintenance method is convenient and quick.

Based on this, technical scheme of the present invention is:

An optimally designed server, including a traditional server system, a GPU server system, tank chains, wires and fixing buckles, the traditional server system and the GPU server system are arranged in layers according to functional modules, and the number of layers is multi-layered, except for the top layer Tank chains for pulling traditional server systems or GPU server systems are provided on the first side of each of the layers, the wires are inserted into the tank chains, and fixed buckles are provided on the second sides of each of the layers.

Further, the traditional server system includes a CPU processor module, a storage module, a network module, a power supply and a fan.

Further, the GPU server system is provided with multiple GPU processor modules.

Further, the number of GPU processor modules is 6.

Further, the traditional server system and the GPU server system are independently arranged in different layers.

Further, the wires are power supply wires and data service wires.

In addition, the present invention also provides a maintenance method for an optimally designed server. Using the above-mentioned optimally designed server, when the server is upgraded or replaced by operation and maintenance,

a. Enter the traditional server system and disable the corresponding GPU server system;

b. Release the fixing buckle and pull out the GPU server system that needs to be replaced or operated and maintained, and push back the replaced GPU server system after successful replacement;

c. Install the fixing buckle, re-enter the traditional server system to re-update the driver and identify the corresponding GPU server system, then the server will enter the normal business mode.

Implementing the embodiment of the present invention has the following beneficial effects:

1. In the optimized server of the present invention, the traditional server system and the GPU server system are layered according to functional modules, and the number of layers is multi-layered, so that the whole machine can be freely stretched and stretched under the condition of uninterrupted operation, and the top layer is removed. One side of each outer layer is equipped with a tank chain for pulling the traditional server system or GPU server system, which ensures the coherence of each layer, avoids the loss of parts when pulling the layer, and The other side of each layer is provided with a fixed buckle, which strengthens the structural firmness of the server after the optimized design.

2. The GPU server system of the present invention is equipped with a plurality of GPU processor modules, preferably 6 GPU processors. When performing artificial intelligence, deep learning, and neural network deduction, the work efficiency can be greatly improved, and its energy efficiency can be comparable to multiple traditional server.

3. The traditional server system and the GPU server system are independently installed in different layers, which is convenient for later maintenance; in addition, the wires are used for power supply wires and data service lines, and are interspersed between the tank chain and the traditional server system or GPU The server system is connected to realize the access of the server CPU to the GPU resources.

4. The maintenance method of the server of the present invention is clear and reasonable, and the server is operated and maintained or iteratively upgraded to realize uninterrupted operation of the server and avoid data loss.

Description of drawings

FIG. 1 is a schematic diagram of the overall structure of the optimization design server described in this embodiment.

FIG. 2 is a flow chart of the maintenance method by using the optimization design server described in this embodiment.

Explanation of reference signs:

Among them, 1. Traditional server system, 2. GPU server system, 3. Tank chain, 4. Fixed buckle.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings and embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in Fig. 1, a kind of optimized design server described in the embodiment of the present invention includes a traditional server system 1, a GPU server system 2, a tank chain 3, the electric wire and a fixed buckle 4, a traditional server system 1 and a GPU server system 2. According to the layered setting of functional modules, and the number of layers is multi-layered, the first side of each layer except the top layer is provided with a tank chain 3 for pulling the traditional server system 1 or GPU server system 2, and the wires Inserted in the tank chain 3, a fixed buckle 4 is provided on the second side of each layer. In the optimally designed server of the embodiment of the present invention, the traditional server system 1 and the GPU server system 2 are layered according to functional modules, and the number of layers is multi-layered, and each layer is specifically a traditional server system 1 or a GPU server system 2 None of them are specific, and they can be freely stretched and stretched under the condition of uninterrupted operation of the whole machine, and one side of each layer except the top layer is equipped with a tank chain 3 for pulling the traditional server system 1 or GPU server system 2, ensuring The arrangement continuity between each layer is improved, and the loss of components is avoided when the layer is pulled, and the other side of each layer is provided with a fixed buckle 4, which strengthens the structural firmness of the server after the optimized design.

Wherein, traditional server system 1 comprises CPU processor module, storage module, network module, power supply and fan, GPU server system 2 is provided with a plurality of GPU processor modules, preferably 6 GPU processors, for artificial intelligence, depth During learning and neural network deduction, work efficiency can be greatly improved, and its energy efficiency can be comparable to that of multiple traditional servers. Both the traditional server system 1 and the GPU server system 2 are independently set in different layers, and the independent setting is convenient for later maintenance; in addition, the wires are power supply wires and data service lines, and are interspersed between the tank chain 3 and the traditional server system 1 or The GPU server system 2 is connected to realize the access of the server CPU to the GPU resource.

In addition, a tank chain 3 is used on one side of each layer except the top layer, and the tank chain 3 is suitable for use in reciprocating motions, and can pull and protect the built-in electric wires; each tank chain 3 The link is composed of left and right chain plates and upper and lower cover plates. Each link can be opened and the chain links can rotate freely, which is convenient for installation and maintenance. Greatly improves the convenience of wiring; the tank chain 3 has high pressure and tensile load, good toughness, high elasticity and wear resistance, flame retardant, stable performance at high and low temperatures, can be used outdoors, and runs at a stable speed .

As shown in FIG. 2, the present invention also provides a maintenance method for an optimally designed server. Using the above-mentioned optimally designed server, when the server is upgraded or replaced by operation and maintenance,

a. Enter the traditional server system 1 and disable the corresponding GPU server system 2;

b. Release the fixing buckle 4 to pull out the GPU server system 2 that needs to be replaced or maintained, and push back the replaced GPU server system 2 after successful replacement;

c. Install the fixing buckle, re-enter the traditional server system 1 to re-update the driver and identify the corresponding GPU server system 2, then the server enters the normal business mode.

The maintenance method of the server of the present invention is clear and reasonable, and the server is operated and maintained or iteratively upgraded to realize uninterrupted operation of the server and avoid data loss.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and substitutions can also be made, and these improvements and substitutions are also possible. It should be regarded as the protection scope of the present invention.

Claims (7)

1. an optimized design server, is characterized in that, comprises traditional server system, GPU server system, tank chain, electric wire and fixed buckle, described traditional server system and GPU server system are arranged according to functional module layering, and layered quantity is Multi-layer, except the top layer, the first side of each layer is provided with a tank chain for pulling the traditional server system or GPU server system, the wires are inserted in the tank chain, and the second side of each layer is provided with With fixed buckle. 2. The optimized design server according to claim 1, wherein said traditional server system comprises a CPU processor module, a storage module, a network module, a power supply and a fan. 3. The optimization design server according to claim 1, wherein a plurality of GPU processor modules are arranged in the GPU server system. 4. The optimized design server according to claim 3, wherein the number of said GPU processor modules is six. 5. The optimized design server according to claim 1, wherein the traditional server system and the GPU server system are independently arranged in different layers. 6. The optimized design server according to claim 1, wherein the electric wires are power supply electric wires and data service wirings. 7. A maintenance method for an optimally designed server, using the optimally designed server according to any one of claims 1 to 6, characterized in that, when the server is upgraded or replaced by operation and maintenance, a. Enter the traditional server system and disable the corresponding GPU server system; b. Unfasten the buckle and pull out the GPU server system that needs to be upgraded or maintained, and push it back to the replaced GPU server system after successful replacement; c. Install the fixing buckle, re-enter the traditional server system to re-update the driver and identify the corresponding GPU server system, then the server will enter the normal business mode.