An artificial intelligence computing power center business equipment group

Abstract

This invention discloses an AI computing center service equipment assembly, belonging to the technical field of AI computing equipment. It addresses at least one of the problems in existing technologies where excessively long connection lines between computing devices lead to insufficient bandwidth, high transmission latency, and significant transmission channel loss. The AI ​​computing center service equipment assembly of this invention comprises multiple modules arranged in multiple layers along a vertical direction. Within each layer, multiple modules are arranged in a matrix, and each module includes multiple individual computing devices arranged in at least one row. This invention can be applied to artificial intelligence.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence computing devices, and particularly relates to a business device group of an artificial intelligence computing center. BACKGROUND

[0002] As a new computing power infrastructure adapting to the development of artificial intelligence, the intelligent computing center takes AI chips as the core, integrates architecture computing systems as the platform, and takes big data as the resource support to build a one-stop public infrastructure including functions such as resource scheduling, algorithm development, model training, deployment, testing and release, which can accurately match the computing power demand of artificial intelligence technology in the two core application links of training and inference. Compared with traditional data centers, the intelligent computing center has significantly different technical characteristics and performance requirements: the power density of the intelligent computing center has shown a trend of normal development of 20-50KW, the GPU servers carried by it have higher requirements for bandwidth and delay, and it needs to realize greater transmission bandwidth and lower transmission delay. At the same time, the high-density cooling and heat dissipation problem caused by high power density has become a key technical problem that must be solved in the construction of the intelligent computing center.

[0003] However, the design concept and technical architecture of the traditional data center have been difficult to meet the above new computing power demand. The traditional data center focuses more on the storage, management and basic processing of large-scale data, and the core hardware mostly uses CPU chips. Its essence is a place for centralized management and maintenance of large-scale servers and storage devices, mainly used for processing various data such as enterprise data, user data and application program data, and the core demand is concentrated on ensuring the security, reliability and high-speed access of data, and the requirements for powerful computing power and low-delay transmission of computing power are relatively low. In terms of hardware layout, the computing devices of the traditional data center are mostly arranged in a single layer and dispersed in each room module. Due to the long physical distance between different room modules, the connection lines between the computing devices are too long, which causes problems such as insufficient bandwidth, high transmission delay, large transmission channel loss and the like, and cannot adapt to the core demand of the intelligent computing center for high bandwidth, low delay and high power density, and also cannot solve the high-density cooling and heat dissipation challenge faced by the intelligent computing center. SUMMARY

[0004] In view of the above analysis, the embodiments of the present application aim to provide a business device group of an artificial intelligence computing center, which is used to solve at least one of the problems of insufficient bandwidth, high transmission delay and large transmission channel loss caused by the too long connection lines between the computing devices in the prior art.

[0005] The purpose of the present application is mainly realized through the following technical solutions.

[0006] This invention provides an equipment assembly for an artificial intelligence computing center, comprising multiple modules arranged in multiple layers along the vertical direction;

[0007] In each layer, multiple modules are arranged in a matrix, and each module includes multiple individual computing devices, which are arranged in at least one row.

[0008] Furthermore, in each module, the individual computing devices in adjacent rows are arranged face-to-face or back-to-back.

[0009] Furthermore, hot and cold aisles are provided between individual computing devices in adjacent rows.

[0010] Furthermore, hot and cold aisles include cold aisles and hot aisles;

[0011] A cold aisle is provided between two rows of single computing devices arranged face-to-face, and a hot aisle is provided between two rows of single computing devices arranged back-to-back.

[0012] Furthermore, the width of the cold aisle is 1200mm to 2400mm, and the width of the hot aisle is 1200mm to 2400mm.

[0013] Furthermore, the equipment setup for the AI ​​computing center also includes a support grid frame, which is installed below the hot and cold aisles.

[0014] Furthermore, each computing device is a cube-shaped cabinet, which houses servers, storage devices, network devices, and / or cabling equipment.

[0015] Furthermore, the gap between two adjacent modules in the vertical direction is 800mm to 1000mm.

[0016] Furthermore, the equipment setup for the AI ​​computing center also includes a support framework, on which the modules are placed.

[0017] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0018] A) The AI ​​computing center equipment assembly provided by this invention arranges multiple modules in multiple layers along the vertical direction, with each computing device forming a cube-shaped layout. This stacked arrangement, consisting of multiple individual computing devices forming modules, and then assembling these modules into a computing center, effectively shortens the physical distance between computing devices, thereby reducing the length of connecting lines and significantly mitigating problems such as insufficient bandwidth, high transmission latency, and significant transmission channel loss. This meets the core requirements of AI computing centers for high bandwidth, low latency, and high power density, and is suitable for AI computing center scenarios with relatively high computing device power and relatively high requirements for bandwidth, transmission latency, and transmission loss.

[0019] B) The AI ​​computing center business equipment assembly provided by this invention allows for flexible adjustment of the number of modules and the number of individual computing devices in each module according to actual needs, making the connection between individual computing devices more regular, further optimizing the signal transmission path, and reducing signal interference and loss.

[0020] c) The AI ​​computing center equipment configuration provided by this invention features cold aisles between two rows of face-to-face single computing devices and hot aisles between two rows of back-to-back single computing devices. This allows for precise temperature control based on the heat distribution characteristics of different areas. The cold aisles effectively remove heat generated by the face-to-face single computing devices, preventing performance degradation due to overheating. The hot aisles provide a suitable temperature environment in the back-to-back arrangement, reducing operational anomalies in low-temperature environments. This targeted design of cold and hot aisles based on equipment layout further improves energy efficiency, reduces cooling and heating costs, effectively addresses the high-density cooling and heat dissipation challenges faced by AI computing centers, ensures stable operation of AI computing center equipment, and enhances overall computing performance and reliability.

[0021] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0022] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0023] Figure 1This is a schematic diagram of the structure of the artificial intelligence computing center business equipment assembly provided in Embodiment 1 of the present invention;

[0024] Figure 2 This is a front view of the artificial intelligence computing center business equipment assembly provided in Embodiment 1 of the present invention;

[0025] Figure 3 This is a top view of the artificial intelligence computing center business equipment assembly provided in Embodiment 1 of the present invention;

[0026] Figure 4 This is a side view of the artificial intelligence computing center business equipment assembly provided in Embodiment 1 of the present invention.

[0027] Figure label:

[0028] 1-Single computing device; 2-Cold aisle; 3-Hot aisle; 4-Supporting frame; 5-Supporting grid frame. Detailed Implementation

[0029] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0030] This invention provides a business equipment setup for an artificial intelligence computing center, see [link / reference]. Figures 1 to 4 It includes multiple modules, which are arranged in multiple layers along the vertical direction, for example, 2 to 4 layers. In each layer, multiple modules are arranged in a matrix (for example, 2×4 arrangement). Each module includes multiple single computing devices 1, for example, 3 to 8, and the multiple single computing devices 1 are arranged in at least one row.

[0031] In each module, the single computing devices 1 in two adjacent rows are arranged face-to-face or back-to-back, and there are hot and cold aisles (e.g., cold aisle 2 or hot aisle 3) between the single computing devices 1 in two adjacent rows.

[0032] The cold and hot aisles include cold aisle 2 and hot aisle 3. Cold aisle 2 is provided between two rows of single computing devices 1 arranged face to face, and hot aisle 3 is provided between two rows of single computing devices 1 arranged back to back.

[0033] It should be noted that, depending on the actual intelligent computing function requirements, the equipment setup of the artificial intelligence computing center in this embodiment can range from a few individual computing devices 1 to a combination of thousands of individual computing devices 1. The effect is particularly obvious for computing centers with a large number of individual computing devices 1 (more than 100 computing devices required), long building units (building units with a length of more than 100 meters), and high floor height (more than 10 meters).

[0034] Compared with existing technologies, the AI ​​computing center equipment setup provided by this invention takes into account the limitations of traditional data center server room layout and the long lines between computing devices. By arranging multiple modules vertically in multiple layers, with each computing device 1 forming a cube-shaped layout, and using a stacked arrangement, multiple individual computing devices 1 form modules, and then assembling these modules to form a computing center, this module-based arrangement can effectively shorten the physical distance between computing devices, thereby reducing the length of connecting lines and significantly reducing problems such as insufficient bandwidth, high transmission latency, and high transmission channel loss. This can meet the core requirements of AI computing centers for high bandwidth, low latency, and high power density, and is suitable for AI computing center scenarios with relatively high computing device power and relatively high requirements for bandwidth, transmission latency, and transmission loss.

[0035] Meanwhile, the number of modules and the number of individual computing devices 1 in each module can be flexibly adjusted according to actual needs, making the connection between individual computing devices 1 more regular, further optimizing the signal transmission path, and reducing signal interference and loss.

[0036] Furthermore, a cold aisle 2 is provided between two rows of face-to-face computing devices 1, and a hot aisle 3 is provided between two rows of back-to-back computing devices 1. This allows for precise temperature control based on the heat distribution characteristics of different areas. The cold aisle 2 can promptly remove the heat generated by the face-to-face computing devices 1, preventing performance degradation due to overheating. The hot aisle 3 provides a suitable temperature environment in the back-to-back areas, reducing equipment malfunctions in low-temperature environments. This design, which specifically sets up cold and hot aisles based on the equipment layout, further improves energy efficiency, reduces cooling and heating costs, effectively addresses the high-density cooling and heat dissipation challenges faced by AI computing centers, ensures the stable operation of AI computing center equipment, and enhances overall computing performance and reliability.

[0037] For example, a single computing device 1 is a cubic cabinet, which is equipped with servers, storage devices, network devices and / or cabling devices for collecting, processing, computing, storing, transmitting and / or retrieving electronic information. The cubic cabinet has a length of 1200mm to 1400mm, a width of 600mm to 800mm and a height of 2000mm to 2500mm.

[0038] Correspondingly, the width of cold aisle 2 is 1200mm to 2400mm, and the width of hot aisle 3 is 1200mm to 2400mm.

[0039] To address the issue of inadequate cabling space, the vertical gap between adjacent modules in the aforementioned AI computing center is 800mm to 1000mm, approximately one-third the height of a single computing device. This ensures sufficient cabling space without wasting space due to excessively large gaps. Furthermore, this size of gap facilitates the installation, debugging, and maintenance of the equipment by staff.

[0040] Understandably, in order to ensure stable support for the modules, the aforementioned AI computing center equipment assembly also includes a support frame 4, on which the modules are placed.

[0041] To ensure sufficient personnel access and load-bearing capacity, the aforementioned AI computing center equipment also includes a supporting grid frame 5, which is located below the hot and cold aisles. (See attached image.) Figure 1 and Figure 3 In this way, while ensuring the proper functioning of the hot and cold aisles, the supporting grid frame 5 provides a safe and convenient passageway for staff, facilitating the inspection and maintenance of the hot and cold aisles and related equipment. Furthermore, the supporting grid frame 5 has sufficient load-bearing capacity to support equipment and personnel of a certain weight, ensuring the structural stability and safety of the entire computing center's business equipment setup.

[0042] The specific solution of the present invention is illustrated below with examples:

[0043] Example 1

[0044] This embodiment provides a business equipment setup for an artificial intelligence computing center. A standard single computing device is 1200mm long, 600mm wide, and 2200mm high, forming a cubic cabinet. An 800mm space is reserved between two adjacent modules in the vertical direction for wiring, installation, and maintenance. In other words, the total height of each module and the reserved space is 9m.

[0045] Each module is arranged with 2 groups of 4 rows of cabinets. The cabinets are arranged face-to-face and back-to-back. Cold aisles or hot aisles are provided between different rows of cabinets. The face-to-face arrangement forms a cold aisle, and the back-to-back arrangement forms a hot aisle, thereby isolating airflow and improving cooling efficiency. Considering a standard single calculation unit length of 1200mm, a cold aisle width of 1800mm, and a hot aisle width of 1800mm, the total length of each layer is 10.2m.

[0046] In each layer, considering a total of 10 cabinets are arranged in the horizontal width direction of the same row of modules, and taking into account that the standard width of a single unit is 600mm, the width of each row of modules is 6m.

[0047] Based on the above layout of the AI ​​computing center's business equipment, we can conclude that each module's dimensions are 10.2m × 6m × 9m. Considering the transmission distance requirements between individual computing devices, such as the requirement that the transmission distance be less than 100 meters if using OM4 multimode optical cable, and taking into account the requirements of the cabling system architecture for the AI ​​computing center's business equipment, the data transmission sharing requirements between individual computing devices in different modules, module placement, module assembly architecture, and space utilization, the 10.2m × 6m × 9m size of the single cube module is more reasonable.

[0048] In this embodiment, the AI ​​computing center business equipment assembly has 120 cabinets, with 12 computing devices per row of modules. This is more suitable for a grid-like arrangement. It should be noted that the number of cabinets in a single row should not be increased indiscriminately, because more individual computing devices will require longer connection lines between different individual computing devices, resulting in problems such as lower bandwidth, higher transmission latency, and higher transmission channel loss.

[0049] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A set of business equipment for an artificial intelligence computing center, characterized in that, It includes multiple modules, which are arranged in multiple layers along the vertical direction; In each layer, multiple modules are arranged in a matrix, and each module includes multiple individual computing devices, which are arranged in at least one row.

2. The setup of the artificial intelligence computing center business equipment according to claim 1, characterized in that, In each module, the individual computing devices in adjacent rows are arranged face-to-face or back-to-back.

3. The setup of the artificial intelligence computing center business equipment according to claim 2, characterized in that, Hot and cold aisles are provided between individual computing devices in adjacent rows.

4. The setup of the artificial intelligence computing center business equipment according to claim 3, characterized in that, The hot and cold passages include cold passages and hot passages; A cold aisle is provided between two rows of single computing devices arranged face-to-face, and a hot aisle is provided between two rows of single computing devices arranged back-to-back.

5. The setup of the artificial intelligence computing center business equipment according to claim 4, characterized in that, The width of the cold aisle is 1200mm to 2400mm, and the width of the hot aisle is 1200mm to 2400mm.

6. The setup of the artificial intelligence computing center business equipment according to claim 2, characterized in that, The AI ​​computing center equipment assembly also includes a support grid frame, which is installed below the hot and cold aisles.

7. The setup of the artificial intelligence computing center business equipment according to claim 1, characterized in that, The single computing device is a cube cabinet, which contains servers, storage devices, network devices and / or cabling devices.

8. The setup of the artificial intelligence computing center business equipment according to claim 1, characterized in that, The gap between two adjacent modules in the vertical direction is 800mm to 1000mm.

9. The setup of the artificial intelligence computing center business equipment according to claim 1, characterized in that, The AI ​​computing center equipment assembly also includes a support frame, on which the module is placed.

Images