AI host system and power consumption dynamic monitoring and optimizing method thereof

Abstract

The invention discloses an AI host system and a power consumption dynamic monitoring and optimizing method thereof. According to the system, dense deployment of as many as ten or more AI hosts is achieved through a high-integration mechanical structure, a single-power-source multi-path power supply and state memory mechanism is adopted, temperature and operation state data are collected in real time in combination with a dynamic monitoring unit, and time-sharing power-on, zero standby power consumption and intelligent recovery are achieved. According to the invention, the energy efficiency and reliability of the system are remarkably improved, and the method is suitable for edge calculation and multi-camera cooperative monitoring scenes.

Description

Technical Field

[0001] This invention relates to the fields of computer hardware systems and energy management technology, and in particular to an AI host system suitable for high-density, high-load AI inference scenarios, and its method for dynamic monitoring and optimization of power consumption. Background Technology

[0002] In fields such as security monitoring and industrial inspection that require multiple cameras to work collaboratively, multiple AI hosts are typically deployed for real-time video stream analysis. Existing technical solutions often employ distributed deployment or simple stacked integration, which has the following drawbacks: Low power supply efficiency: Traditional solutions use multiple independent power supplies or an unoptimized single power supply, which cannot cope with the drastic fluctuations in the load of AI inference tasks. There are significant "ghost loads" (standby power consumption), resulting in low power supply efficiency and high overall PUE.

[0003] Insufficient reliability: When multiple hosts are started simultaneously or the load increases suddenly, the power supply voltage is prone to fluctuations, which may cause the host to crash or lose data. At the same time, frequent current surges will shorten the lifespan of the hardware.

[0004] Low operation and maintenance recovery efficiency: After an unexpected power outage and subsequent recovery, the system cannot automatically return to its pre-outage operating state, requiring manual intervention, which affects the continuity of critical AI inference services.

[0005] Therefore, there is an urgent need for an AI host system solution that can achieve precise power supply management, dynamically adapt to load changes, and has high reliability and high energy efficiency. Summary of the Invention

[0006] To address the technical problems existing in the prior art, this invention provides a highly integrated, energy-efficient, and stable AI host system and its power consumption dynamic monitoring and optimization method.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an AI host system, comprising: Chassis; and integrated within the chassis: Multiple AI hosts are used to perform AI inference tasks; The highly integrated mechanical structure includes a custom bracket for fixing the AI ​​host and a cable storage slot located on the back or side of the custom bracket. A single-power-source multi-channel power supply system, comprising an AC / DC power module, an intelligent power distribution unit, and a main control board; The dynamic monitoring unit is used to monitor the temperature and operating status of each host in real time; The centralized control unit is located inside the main control board and is used to achieve power consumption optimization and state memory.

[0008] Preferably, the plurality of AI hosts are 10 or more AI hosts with x86 / ARM architecture.

[0009] Preferably, the customized bracket is made of modular aluminum alloy or steel material, supports stacking or rack mounting, and includes a thermal expansion coefficient matching and vibration isolation structure.

[0010] Preferably, the cable storage slot has built-in cable clips, a labeling system, and elastic cushioning material, supporting color coding and centralized outlet cabling.

[0011] Preferably, the intelligent power distribution unit has a state memory function, which records the last switching state of each AI host through a non-volatile memory, so as to achieve time-sharing power-on and zero standby power consumption.

[0012] Preferably, the dynamic monitoring unit uploads temperature and operating status data to the centralized control unit via the I2C bus to achieve real-time load and power consumption correlation analysis.

[0013] Secondly, this invention provides a method for dynamic monitoring and optimization of power consumption in an AI host system, the method being applied to the aforementioned system, the method comprising: The AC / DC power module outputs multiple independent and isolated DC power supplies to the intelligent power distribution unit. The dynamic monitoring unit monitors the temperature and operating status of each AI host in real time and uploads the monitoring data via the I2C bus. The centralized control unit executes the following power supply strategy based on the state memory stored in the non-volatile memory of the main control board: Time-sharing power-on strategy: When the system starts up, the intelligent power distribution unit controls multiple AI hosts that are recorded as "power on" to be powered on sequentially at preset time intervals. Power outage recovery strategy: After the system experiences a power outage and is powered on again, it automatically reads the state memory and controls the intelligent power distribution unit to restore each AI host to its respective on / off state before the power outage; Zero standby power consumption strategy: During system operation, when the centralized control unit determines that an AI host has entered a "shutdown" state, it controls the intelligent power distribution unit to cut off the power supply to the host.

[0014] Preferably, the state memory includes continuously recording and updating the last known on / off state of each AI host in the cluster.

[0015] Thirdly, the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-described method.

[0016] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-described method. Compared with the prior art, the beneficial effects of the present invention are as follows: Extreme energy efficiency: By precisely cutting off the power supply to the powered-off host, "ghost load" is completely eliminated, significantly reducing the overall PUE when the cluster is under partial load.

[0017] High reliability: Time-sharing power-on avoids current surges during closing, and dynamic monitoring prevents overload and overheating, enhancing system stability and hardware lifespan.

[0018] Intelligent recovery: After a power outage, the system can automatically return to its previous working state, greatly reducing manual intervention and ensuring the continuity of critical AI inference tasks 24 / 7.

[0019] High integration and easy maintenance: The highly integrated mechanical structure design optimizes the spatial layout and simplifies the wiring and maintenance process. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall structure of the AI ​​host system of the present invention.

[0023] Figure 2 This is a schematic diagram of the internal structure of the AI ​​host system of the present invention. Figure 1 .

[0024] Figure 3 This is a schematic diagram of the internal structure of the AI ​​host system of the present invention. Figure 2 .

[0025] Figure 4 This is a flowchart of the power consumption dynamic monitoring and optimization method of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0028] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0029] Example 1 This embodiment details the specific composition of an AI host system. The system constructs a highly integrated hardware platform, whose core concept lies in solving the energy efficiency and reliability issues in high-density AI host deployments within existing multi-camera monitoring systems through a three-tiered collaborative design of "mechanical structure - power supply system - monitoring and control."

[0030] like Figure 1-3 As shown, the AI ​​host system includes a chassis 100, which serves as the physical carrier and supporting structure of the system. The chassis 100 integrates the following core functional modules: 1. Highly integrated mechanical structure This structure is the physical basis for the system to achieve high-density deployment, stable operation and convenient maintenance. It mainly consists of a customized bracket 101 and a cable storage channel (not shown in the figure), which is used to densely deploy 10 or more x86 / ARM architecture AI hosts 103.

[0031] The custom bracket 101 features a modular design and is made of aluminum alloy or steel. Each bracket unit precisely matches the dimensions and mounting holes of the x86 or ARM architecture AI host 103, supporting stacking or standard rack mounting. The bracket design takes into account thermal expansion coefficient matching and vibration isolation, allows for quick disassembly and fixing with screws, and supports lateral expansion, allowing for dense deployment of up to 10 or more AI hosts 103 within a single rack.

[0032] The cable management channel, located on the back or side of the custom bracket 101, is a key design element for optimizing space and cabling. This channel features a segmented cover to conceal power and data cables. Cable clips and a labeling system with color coding are installed inside the channel for easy cable identification and maintenance. The cable channel is aligned with the host interface to reduce bending radius and prevent signal interference. Furthermore, elastic cushioning material is integrated inside the channel to prevent cable abrasion. All cables connect to the network switch or power supply through a centralized outlet within the channel, greatly simplifying cabling complexity.

[0033] 2. Single-power-source multi-channel power supply system This system is the core of achieving intelligent power consumption management and high reliability. It consists of an AC / DC power module 104, an intelligent power distribution unit (not shown in the figure), and a main control board 106 working together.

[0034] The AC / DC power module 104 serves as the sole power source for the system. It is a high-power, high-efficiency power module used to convert mains power into multiple independent and isolated DC power outputs (e.g., 12V, 5V).

[0035] Each output port of the intelligent power distribution unit is controlled by the main control board 106 and has independent on / off control capability.

[0036] The main control board 106 is the control center of the system. It integrates a microprocessor and non-volatile memory.

[0037] The centralized control unit, serving as the intelligent decision-making core of the system, is integrated into the main control board 106. It executes power optimization strategies, including time-sharing power-on, zero standby power consumption, and power-off recovery, based on real-time data reported by the dynamic monitoring unit and state memories in non-volatile memory. The centralized control unit's functions are implemented through firmware running on the microprocessor. It issues precise instructions to the intelligent power distribution unit by accessing state memories in non-volatile memory (i.e., the last known on / off state ("power on" or "power off") of each AI host 103, which is continuously recorded and updated.

[0038] The specific power supply link allocation includes: after the switching power supply is started, the output power is first transmitted to the intelligent power distribution module, which divides the power into two independent power supply branches: one is a power supply branch adapted to the working voltage of the main control board to ensure the stable operation of the main control board; the other is a power supply branch adapted to the working voltage of the AI ​​host, which provides power to 10 or more AI hosts at the same time.

[0039] 3. Dynamic monitoring and heat dissipation unit The dynamic monitoring unit (not shown in the figure) consists of a microcontroller and digital temperature sensors distributed in the AI ​​host 103 area and near the power module. It is used to monitor the temperature and operating status (such as load) of each host in real time. The monitoring data is uploaded to the main control board 106 via the I2C bus.

[0040] The heat dissipation unit 110, such as a fan, operates according to the instructions of the main control board 106. The main control board 106 dynamically adjusts the speed of the heat dissipation unit 110 (such as a fan) through a PWM signal based on the temperature data reported by the dynamic monitoring unit, thereby achieving precise heat dissipation.

[0041] 4. Core Computing and External Interfaces Multiple AI hosts 103 (e.g., 10 or more) are fixed on a custom bracket 101 to perform inference tasks such as AI image recognition. Each AI host 103 can be independently connected to an external monitoring camera to form a data acquisition unit with a corresponding "host-camera" configuration.

[0042] A network switch 111 is integrated into the chassis 100 to enable data exchange between all AI hosts 103 and the external network. The rear panel of the chassis 100 has multiple network cable ports 112 for connecting external cameras and the network.

[0043] In this embodiment, the localized centralized control interface is preferably an external high-definition capacitive touchscreen 113, which is connected to the main control board 106. The touchscreen 113 is used to display parameters such as the number, operating status, system voltage, and temperature of all AI hosts 103, and to receive user touch commands (such as one-click restart or individual power on / off of a host), achieving localized centralized control with millisecond-level response.

[0044] 5. Auxiliary Function Module The main power switch 114 and the switch switch 115 are located on the front panel of the chassis and are used to control the power supply of the entire machine and the network switch, respectively; the filter switch 116 is located on the front panel of the chassis and is used to connect the power filtering function to provide the system with clean and stable mains power input.

[0045] The main unit cover 117 is used to enclose the chassis 100, serving to prevent dust, provide protection, and provide safety isolation.

[0046] In practical applications, the AI ​​host system described above in this embodiment adopts a centralized hardware integration design, integrating multiple AI hosts (including host computing modules) inside the chassis. Each host is independently connected to a monitoring camera, forming a corresponding data acquisition unit of "host + camera". This data acquisition unit is integrated with the main control board and the single power supply multi-path power supply system in the same chassis, abandoning the traditional distributed deployment or simple stacking scheme, greatly reducing external wiring, lowering installation and maintenance costs, and avoiding the risk of local unit failure due to differences in the site environment, thus ensuring the continuity of monitoring.

[0047] In practical applications, the single-power multi-path power supply system of the AI ​​host system described in this embodiment uses a high-power switching power supply, i.e., an AC / DC power module, as the sole power source. The output voltage is divided into two paths through an intelligent power distribution unit. One path powers the main control board, and the other path powers multiple AI hosts. This replaces the traditional multi-independent power supply solution, reduces the space occupied by the chassis, and avoids interference from voltage fluctuations of different power supplies on data interaction, ensuring the stability of real-time transmission and processing of monitoring data.

[0048] In practical applications, the main control board, as the core control unit, establishes electrical connections with the AI ​​host (power control link), touch screen (data interaction and command transmission link), and temperature sensor (data acquisition link) to form a closed-loop working system of "power supply support, control center, and peripheral response" to ensure that each module works together to achieve monitoring data acquisition and equipment operation and maintenance management.

[0049] In practical applications, the AI ​​host system described in this embodiment adds a dynamic monitoring unit between the AC / DC power module and the intelligent power distribution unit to monitor the power voltage fluctuations of multiple AI hosts, thereby avoiding AI host crashes or data loss caused by power supply voltage fluctuations, improving the stability of the system under high load, and realizing dynamic power consumption adaptation optimization of the system.

[0050] In summary, this embodiment constructs a high-performance AI host system integrating power supply, control, data acquisition, and monitoring through the deep integration and collaboration of the aforementioned modules. The highly integrated mechanical structure forms the physical foundation, the single-power-multiple-path power supply system is the core of energy and intelligent management, the dynamic monitoring unit and the centralized control unit together constitute the system's "nerves" and "brain," and the AI ​​host, network switch, and other components are the core components for executing computational tasks. This architecture effectively solves the shortcomings of traditional solutions in terms of integration, energy efficiency, and reliability.

[0051] Example 2 This invention provides a method for dynamic monitoring and optimization of power consumption in an AI host system, applied to the system described in Embodiment 1. The core of this method lies in transforming the traditional passive power supply mode into a state-aware, proactive intelligent management and control mode through deep collaboration between software logic and hardware systems, thereby systematically solving the energy efficiency and reliability issues of high-density AI computing clusters. The method flow is as follows: Figure 4 As shown, the specific steps include: S201, Power Supply Output: After the system is powered on, the AC / DC power module starts working, outputting multiple independent and isolated DC power supplies to the intelligent power distribution unit. This step establishes the system's basic energy supply, providing stable and clean multi-channel DC power for all subsequent intelligent power distribution. Its "single power supply" design fundamentally avoids the voltage fluctuation interference problem caused by individual differences in multi-power supply schemes.

[0052] S202. Status Monitoring and Data Upload: The dynamic monitoring unit monitors the temperature and operating status of each AI host in real time and uploads the monitoring data via the I2C bus. Specifically, when the dynamic monitoring unit starts working, its integrated microcontroller collects temperature data of key areas of each AI host in real time through digital temperature sensors and obtains the host's operating status data (such as power consumption and load rate) through the monitoring circuit. All this real-time data is continuously uploaded to the centralized control unit in the main control board via the I2C bus. This step constitutes the system's "sensory nerves," providing the centralized control unit with the necessary environmental and load data support for decision-making, realizing precise control from "blind control" to "data-driven" control.

[0053] S203. Intelligent Strategy Execution: Through the centralized control unit, combining the real-time data uploaded by the dynamic monitoring unit and the state memory stored in the non-volatile memory, the following three core power supply strategies are executed: Time-sharing power-on strategy: During system startup, the centralized control unit reads the status memory and controls the intelligent power distribution unit to sequentially power on all AI hosts recorded as "power on" at preset time intervals (e.g., 1-2 seconds), rather than simultaneously. This strategy is a proactive surge current suppression method. By staggering the startup peak current of each host, it effectively avoids the huge current surge to the AC / DC power modules and data center power distribution caused by all hosts being powered on simultaneously, significantly improving system startup reliability and hardware lifespan.

[0054] Power Outage Recovery Strategy: When the system experiences an unexpected rack-level power outage and is subsequently powered back on, the centralized control unit first automatically reads the on / off state of each AI host recorded in the non-volatile memory just before the power outage. Then, it controls the intelligent power distribution unit to precisely restore each AI host to its pre-outage on / off state. This strategy enables automatic disaster recovery of business operations. It ensures that critical AI inference tasks (such as 24 / 7 monitoring) can be automatically resumed to the maximum extent possible after power outage recovery, greatly reducing manual intervention by maintenance personnel and guaranteeing business continuity and service robustness.

[0055] Zero standby power consumption strategy: During system operation, when the centralized control unit determines that an AI host has entered a "shutdown" state through operating system signals or power consumption criteria, it immediately updates the state memory and controls the intelligent power distribution unit to completely cut off the power supply to that host. This strategy precisely eliminates "ghost loads." In traditional systems, even when a host is powered off, it still consumes standby power. This strategy achieves true "zero standby power consumption" through physical power disconnection, significantly reducing the average PUE of the entire system during partial load operation of the cluster, achieving ultimate energy efficiency.

[0056] S204. State Memory Maintenance: Throughout the entire system operation cycle, the centralized control unit continuously and in real-time updates the state records in the non-volatile memory. Whether operated by a user via a touchscreen or automatically powered on / off according to the strategy, any change in the AI ​​host state is immediately recorded. This step ensures the real-time nature and accuracy of the "state memory," providing a unique and reliable data source for all the aforementioned intelligent strategies, and is the cornerstone for the correct execution of the entire method.

[0057] Compared to existing technologies, the power consumption dynamic monitoring and optimization method for the AI ​​host system in this embodiment has the following advantages: Compared to existing technologies where all hosts are powered on simultaneously, resulting in large inrush currents, this embodiment avoids current surges by using time-sharing power-on based on state memory during system startup. This protects the power supply system and hardware components, extends equipment lifespan, and achieves high system reliability.

[0058] Compared to existing technologies where the host still consumes standby power ("ghost load"), this embodiment completely cuts off power supply through a zero standby power consumption strategy, significantly reducing the overall system PUE, saving operating costs, and achieving the ultimate energy efficiency of green computing.

[0059] Compared to existing technologies that require manual intervention to restore services after an unexpected power outage, involving confirmation and activation of each machine and resulting in long service interruptions, this embodiment uses a power outage recovery strategy to automatically restore the system to its pre-outage state, greatly reducing manual intervention, ensuring the continuity of critical services 24 / 7, and achieving intelligent recovery and high availability.

[0060] Compared to existing technologies where daily management and control are decentralized and rely on manual or simple remote commands, this embodiment uses a centralized control unit to make fully automatic and intelligent decisions based on real-time data and status memory, achieving precise management and control with millisecond-level response, simplifying the operation and maintenance process, and improving operation and maintenance efficiency.

[0061] The specific implementation methods of the above methods and steps are the same as the specific implementation methods of the functions of each module in Embodiment 1 above, and will not be repeated here.

[0062] Example 3 The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods.

[0063] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0064] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0065] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An AI host system, characterized in that, include: Chassis; And those integrated into the chassis: Multiple AI hosts are used to perform AI inference tasks; The highly integrated mechanical structure includes a custom bracket for fixing the AI ​​host and a cable storage slot located on the back or side of the custom bracket. A single-power-source multi-channel power supply system, comprising an AC / DC power module, an intelligent power distribution unit, and a main control board; The dynamic monitoring unit is used to monitor the temperature and operating status of each host in real time; The centralized control unit is located inside the main control board and is used to achieve power consumption optimization and state memory.

2. The system according to claim 1, characterized in that, The multiple AI hosts refer to 10 or more AI hosts based on x86 / ARM architecture.

3. The system according to claim 1, characterized in that, The customized bracket is made of modular aluminum alloy or steel and supports stacking or rack mounting, including thermal expansion coefficient matching and vibration isolation structure.

4. The system according to claim 1, characterized in that, The cable storage tray has built-in cable clips, a labeling system, and elastic cushioning material, and supports color coding and centralized outlet cabling.

5. The system according to claim 1, characterized in that, The intelligent power distribution unit has a state memory function, which records the last switching state of each AI host through a non-volatile memory, realizing time-sharing power-on and zero standby power consumption.

6. The system according to claim 1, characterized in that, The dynamic monitoring unit uploads temperature and operating status data to the centralized control unit via the I2C bus to achieve real-time load and power consumption correlation analysis.

7. A method for dynamic monitoring and optimization of power consumption in an AI host system, characterized in that, The method is applied to the system as described in any one of claims 1-6, and the method includes: The AC / DC power module outputs multiple independent and isolated DC power supplies to the intelligent power distribution unit. The dynamic monitoring unit monitors the temperature and operating status of each AI host in real time and uploads the monitoring data via the I2C bus. The centralized control unit executes the following power supply strategy based on the state memory stored in the non-volatile memory of the main control board: Time-sharing power-on strategy: When the system starts up, the intelligent power distribution unit controls multiple AI hosts that are recorded as "power on" to be powered on sequentially at preset time intervals; Power outage recovery strategy: After the system experiences a power outage and is powered on again, it automatically reads the state memory and controls the intelligent power distribution unit to restore each AI host to its respective on / off state before the power outage; Zero standby power consumption strategy: During system operation, when the centralized control unit determines that an AI host has entered the "shutdown" state, it controls the intelligent power distribution unit to cut off the power supply to the host.

8. The method according to claim 7, characterized in that, The state memory includes continuously recording and updating the last known on / off state of each AI host in the cluster.

9. An electronic device comprising a processor, a memory, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in claim 7 or 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in claim 7 or 8.

Images