Intelligent control method and system for domestic AI server

Abstract

The invention relates to the technical field of computers, and discloses a domestic AI server intelligent control method and system, and the method comprises the steps: setting a special interface at each slot position of a PCIe expansion board, and integrating an in-place detection pin and an I2C bus; directly connecting the in-place detection pin and the I2C bus to a mainboard BMC (Baseboard Management Controller) chip through a special cable; the BMC generates a fan speed regulation instruction based on the slot real-time data; and the BMC provides a GPU parameter input interface and stores the GPU parameter input interface to the classification database. According to the invention, through the synergistic effect of an original hardware direct connection architecture, dynamic mapping control and an elastic adaptation mechanism, the energy efficiency performance of the equipment is obviously improved; according to the method, the traditional lengthy deployment period is compressed to an instant level, and the operation and maintenance complexity of the system is unprecedentedly reduced on the premise of ensuring the extreme computing power output, so that the domestic server realizes breakthrough leap in the aspects of safety, reliability and management agility.

Description

Technical Field

[0001] This application relates to the field of computer technology, such as a method and system for intelligent management and control of domestically produced AI servers. Background Technology

[0002] With the explosive growth of artificial intelligence applications, AI servers, as core computing infrastructure, are experiencing continuous increases in integration scale and computing density. Compared to traditional servers, modern AI servers are generally equipped with multiple high-performance GPU cards, significantly increasing system complexity and power consumption, and placing almost stringent requirements on thermal management. Meanwhile, driven by both changes in the international situation and information security needs, AI servers using domestically produced processors and GPUs have become an industry necessity.

[0003] However, the development of domestic AI servers faces severe challenges: on the one hand, due to limitations in domestic chip manufacturing processes, their power consumption is significantly higher than that of mainstream international products with the same performance, resulting in prominent noise issues during equipment operation. In particular, under non-full-capacity operating conditions, traditional heat dissipation solutions struggle to balance cooling efficiency and noise control. On the other hand, the domestic GPU ecosystem is in a rapid development phase, with the mixed use of chips from different manufacturers and generations becoming increasingly common. However, existing management technologies suffer from severe inadequacy in recognizing and compatibility with heterogeneous GPUs, and the adaptation cycle for new models is too long, greatly restricting deployment flexibility.

[0004] Current server management solutions in the industry have significant limitations: In terms of heat dissipation control, they generally adopt fan speed adjustment strategies based on overall temperature feedback, which cannot accurately respond to localized high-heat areas, resulting in low energy efficiency and excessive noise. In terms of hardware management, they rely on pre-built firmware libraries to identify GPU models, which is difficult to adapt to the rapid iteration of domestic chips, and the problem of lagging support for new hardware is prominent. Especially for PCIe expansion architecture, existing designs focus more on link expansion capabilities and lack in-depth support for slot-level status monitoring, causing the heat dissipation strategy to be out of touch with the actual situation.

[0005] The aforementioned problems are intertwined, forming a key bottleneck restricting the deployment of domestically produced AI servers: the contradiction between high-power chips and crude heat dissipation strategies forces devices to struggle to balance excessive noise and computing power loss; the lack of management in mixed-use scenarios leads to a significant reduction in system reliability; and the delay in adapting to new domestic hardware directly hinders the overall industry development process. The industry urgently needs a solution that deeply integrates hardware architecture innovation and intelligent control strategies to achieve a fundamental breakthrough in high efficiency, low noise, and wide compatibility for domestically produced AI servers.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0008] This disclosure provides a method and system for intelligent management and control of domestically produced AI servers, so as to achieve high efficiency, low noise and wide compatibility of domestically produced AI servers.

[0009] In some embodiments, the intelligent management and control method for domestically produced AI servers includes: Hardware architecture: Each slot on the PCIe expansion board has a dedicated interface that integrates in-situ detection pins and the I2C bus; Direct signal connection: The in-situ detection pin is directly connected to the I2C bus to the motherboard BMC chip via a dedicated cable; Dynamic control: BMC generates fan speed control commands based on real-time slot data; Flexible adaptation: BMC provides an interface for inputting GPU parameters and storing them in a classification database.

[0010] Optionally, the dedicated interface includes: The in-situ detection circuit includes a first pin, a second pin, and a third pin. The first pin is connected to the power supply to provide the operating voltage for the detection circuit. The second pin is connected to the ground line to form a current loop reference. The third pin outputs high and low level signals. It outputs a high level when the graphics processor is inserted into the slot and a low level when it is not inserted. The dual-channel I2C bus includes a fourth pin and a fifth pin. The fourth pin serves as a serial data signal line, transmitting the temperature and power consumption data of the graphics processor; the fifth pin serves as a serial clock signal line, synchronizing the data communication timing.

[0011] Optionally, the dedicated cable is a silver-plated shielded cable with a length ≤15cm and a signal delay <1ms.

[0012] Optionally, the dynamic control includes: Establish a mapping relationship between GPU slots and fan groups; The target rotation speed is dynamically calculated based on the power consumption curve of a domestically produced GPU. The calculation formula is as follows: Target speed = reference speed + (real-time temperature - reference temperature) × adjustment coefficient.

[0013] Optionally, the adjustment coefficient is 5% / ℃, the reference temperature is 60℃, and the reference speed is 30%.

[0014] Optionally, the flexible adaptation includes: The BMCWeb interface allows input fields for: GPU model, TDP power consumption, and maximum temperature. Automatically categorize cards into training cards or inference cards and bind them to preset strategy templates.

[0015] In some embodiments, the domestically developed AI server intelligent management and control system includes: Signal acquisition module: Obtains GPU temperature and on-state status via direct connection to the I2C bus through PCIe slot; Dynamic mapping module: Binds fan groups to the physical location of slots and generates a slot-fan control matrix; Elastic database: Stores user-entered GPU parameters and associated cooling strategies.

[0016] Optionally, the elastic database is an embedded SQLite database, and the index fields include: GPU model, category, TDP power consumption, and policy ID.

[0017] Optionally, the dynamic mapping module performs the following process: a) Detect the GPU slot's in-slot status; b) Match the air duct area to which the slot belongs; c) Bind the controllable fan group for this area; d) Adjust fan speed based on real-time temperature.

[0018] In some embodiments, the storage medium stores program instructions, which, when executed, perform the dynamic control and flexible adaptation steps of the aforementioned intelligent management and control method for domestic AI servers.

[0019] The domestic AI server intelligent management and control method and system provided in this disclosure can achieve the following technical effects: This invention's intelligent management and control method for domestically produced AI servers fundamentally improves the overall performance of these servers through a direct hardware connection architecture and a collaborative control mechanism. At the hardware level, by utilizing dedicated slot-level interfaces and extremely short-path signal direct connections, instantaneous and accurate capture of device status is achieved, significantly eliminating control lag issues caused by signal delays in traditional solutions. At the management level, based on dynamically generated slot-fan mapping relationships and customized speed adjustment strategies tailored to the characteristics of domestically produced chips, the noise level of the equipment is significantly reduced while ensuring the continuous and stable operation of high-power chips. At the compatibility level, an open parameter adaptation mechanism unprecedentedly improves the management reliability of heterogeneous GPU mixed-insertion scenarios. These three aspects work together to form a closed loop, enabling domestically produced AI servers to achieve unprecedented levels of quiet operation and flexible deployment while ensuring maximum computing power output.

[0020] The domestically developed AI server intelligent management and control system breaks through traditional management bottlenecks with its modular architecture. The signal acquisition module, with its direct-connection design, significantly improves the real-time performance and accuracy of domestic GPU status monitoring; the dynamic mapping module, through intelligent binding of physical location and heat dissipation resources, achieves precise wind control of localized temperature zones, fundamentally solving the high noise problem of traditional overall heat dissipation solutions; and the elastic database module completely breaks down the compatibility barriers of new hardware models, allowing users to instantly expand management capabilities. The deep synergy of these three components not only significantly reduces operational complexity but also enables domestically developed servers to maintain a leading management adaptability in the rapidly evolving chip ecosystem.

[0021] The program embedded in the storage medium transforms innovative control logic into efficiently executable digital instructions, endowing domestically produced servers with the core capability of self-optimization. The dynamic control flow implemented by the program enables the device to autonomously adjust its heat dissipation strategy based on real-time operating conditions, significantly improving energy efficiency. Furthermore, the programmatic execution of the flexible adaptation function fundamentally overturns the traditional firmware upgrade model, enabling real-time management of new hardware. This breakthrough at the software level allows domestically produced AI servers to demonstrate outstanding autonomous decision-making capabilities and unprecedented deployment agility in complex and ever-changing operating environments.

[0022] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0023] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1 This is a schematic diagram of the hardware architecture of a domestic AI server intelligent management and control method provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the BMC software flow of a domestic AI server intelligent management control method provided in an embodiment of this disclosure. Detailed Implementation

[0024] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0025] The terms "first," "second," etc., used in the embodiments of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0026] In this disclosure, the terms "upper," "lower," "inner," "middle," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for better describing the embodiments of this disclosure and their implementations, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to require them to be constructed and operated in a specific orientation. Furthermore, some of the aforementioned terms may be used to indicate other meanings besides orientation or positional relationship; for example, the term "upper" may in some cases indicate a dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this disclosure according to the specific circumstances.

[0027] Furthermore, the terms "set up," "connect," and "fix" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this disclosure according to the specific circumstances.

[0028] Unless otherwise stated, the term "multiple" means two or more.

[0029] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0030] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0031] It should be noted that, unless otherwise specified, the embodiments and features described in the present disclosure can be combined with each other.

[0032] This invention aims to address three key issues facing domestically produced AI servers: First, domestically produced GPUs suffer from significantly higher power consumption due to manufacturing process limitations. Traditional fan speed control solutions are forced to run at full speed under half-load or mixed-use conditions, resulting in excessive noise and low energy efficiency. Second, heterogeneous domestically produced GPUs (such as training cards and inference cards mixed together) have rapid model iterations and large parameter differences. Existing BMC firmware cannot dynamically recognize new hardware, leading to the failure of heat dissipation strategies and a decrease in system stability. Third, new domestically produced GPUs require a long adaptation period from market launch to BMC firmware support, severely restricting deployment efficiency.

[0033] To address the aforementioned issues, this invention proposes a three-level collaborative technical solution: At the hardware level, a dedicated physical interface is designed in the PCIe expansion board slot, integrating device presence detection pins and I2C bus channels, and directly connected to the motherboard BMC chip via shielded cables to achieve millisecond-level accurate acquisition of slot-level status; At the control level, a real-time dynamic mapping relationship is established between the GPU physical slot, heat dissipation airflow, and fan group, and a frequency conversion control algorithm is generated based on the measured power consumption curve of domestic GPUs, adjusting the speed of local fans as needed according to real-time temperature, breaking the traditional overall speed control paradigm; At the adaptation level, a user-defined parameter interface is opened based on the BMC management interface, supporting manual input of key parameters for GPU models not pre-installed, and a categorized storage database is built to automatically bind heat dissipation strategy templates, enabling plug-and-play functionality for new models.

[0034] The beneficial effects of this invention are significantly reflected in three breakthroughs: First, through slot-level precise heat dissipation control, it significantly reduces equipment noise while ensuring the stable operation of domestically produced high-power chips, especially under half-load conditions; second, the flexible adaptation mechanism completely eliminates blind spots in heterogeneous GPU identification, ensuring reliable management and heat dissipation optimization in scenarios where domestically produced GPUs from different manufacturers and generations are mixed; third, users can input parameters of new domestically produced GPUs in real time, compressing the adaptation cycle to near zero and greatly improving deployment flexibility. In addition, the BMC integrated monitoring interface provides slot-level status visualization, significantly improving operational efficiency.

[0035] Combination Figure 1-2 As shown in the embodiments of this disclosure, a method for intelligent management and control of domestically produced AI servers is provided, including aspects such as hardware architecture, direct signal connection, dynamic control, and flexible adaptation.

[0036] Hardware architecture: Each slot on the PCIe expansion board has a dedicated interface that integrates in-situ detection pins and the I2C bus; Direct signal connection: The in-situ detection pin is directly connected to the I2C bus to the motherboard BMC chip via a dedicated cable; Dynamic control: BMC generates fan speed control commands based on real-time slot data; Flexible adaptation: BMC provides an interface for inputting GPU parameters and storing them in a classification database.

[0037] Specifically, each slot on the PCIe expansion board is equipped with a dedicated composite interface, integrating a 3-pin presence detection circuit (PIN1: +3.3V power supply, PIN2: ground, PIN3: status signal) to sense the GPU insertion / removal status in real time; and a dual-channel I2C bus is configured (PIN4: SDA, PIN5: SCL), which is directly connected to the motherboard BMC chip through a silver-plated shielded cable, with the cable length controlled within 15cm to ensure signal delay <1ms; the motherboard uses a Phytium D2000 processor + domestic BMC-CI100 chipset, whose GPIO expander parses the slot signals and caches temperature / power consumption data through onboard memory.

[0038] That is, the dedicated interface includes in-situ detection circuitry and a dual-channel I2C bus.

[0039] The in-situ detection circuit includes a first pin, a second pin, and a third pin. The first pin is connected to the power supply to provide the operating voltage for the detection circuit. The second pin is connected to the ground line to form a current loop reference. The third pin outputs high and low level signals. It outputs a high level when the graphics processor is inserted into the slot and a low level when it is not inserted.

[0040] The dual-channel I2C bus includes a fourth pin and a fifth pin. The fourth pin serves as a serial data signal line, transmitting temperature and power consumption data of the graphics processor; the fifth pin serves as a serial clock signal line, synchronizing the data communication timing.

[0041] Optionally, the BMC firmware executes the following core logic: mapping relationship generation: bind the corresponding fan group (Fan1-3) according to the physical coordinates of the GPU slot (e.g., Slot1-3 is located in the left air duct area) and establish a dynamic control matrix.

[0042] Optionally, the dynamic control includes: Establish a mapping relationship between GPU slots and fan groups; The target rotation speed is dynamically calculated based on the power consumption curve of a domestically produced GPU. The calculation formula is as follows: Target speed = reference speed + (real-time temperature - reference temperature) × adjustment coefficient.

[0043] Optionally, the adjustment coefficient is 5% / ℃, the reference temperature is 60℃, and the reference speed is 30%.

[0044] Optionally, the flexible adaptation includes: The BMCWeb interface allows input fields for: GPU model, TDP power consumption, and maximum temperature. Automatically categorize cards into training cards or inference cards and bind them to preset strategy templates.

[0045] In some embodiments, the domestically developed AI server intelligent management and control system includes: Signal acquisition module: Obtains GPU temperature and on-state status via direct connection to the I2C bus through PCIe slot; Dynamic mapping module: Binds fan groups to the physical location of slots and generates a slot-fan control matrix; Elastic database: Stores user-entered GPU parameters and associated cooling strategies.

[0046] Optionally, the elastic database is an embedded SQLite database, and the index fields include: GPU model, category, TDP power consumption, and policy ID.

[0047] Optionally, the dynamic mapping module performs the following process: a) Detect the GPU slot's in-slot status; b) Match the air duct area to which the slot belongs; c) Bind the controllable fan group for this area; d) Adjust fan speed based on real-time temperature.

[0048] As a preferred option, an abnormal fuse mechanism is introduced: when the temperature of a single slot exceeds 105°C for 5 minutes, the GPU frequency is automatically reduced and the heat dissipation is accelerated.

[0049] As an example, such as Figure 1 As shown, the intelligent management and control method for domestic AI servers consists of a motherboard, a PCIe expansion board, and a chassis. The motherboard mainly integrates a domestic CPU, BMC, memory, PCIe interface, and other peripheral circuits and expansion interfaces, providing the server with functions such as task scheduling management and out-of-band BMC management. The PCIe expansion board integrates two high-performance PCIe switch chips for expanding PCIe x16 slots. In addition to connecting to the CPU, the PCIe switch uplink can also connect to another PCIe switch downlink on the PCIe expansion board. The PCIe switch supports P2P communication. The expansion board supports three common topology modes: Balance, Common, and Cascade, offering flexible expansion options to meet different application needs.

[0050] According to the AI ​​server chassis structure and overall architecture, the CPU motherboard is located in the bottom 1U space of the chassis, and the PCIE expansion board and AI / GPU card are located in the upper 3U space of the chassis. The PCIE on the motherboard is connected to the uplink interface of the PCIE Switch on the PCIE expansion board via a SlimiSAS cable. The I2C bus and PCIE clock of the BMC on the motherboard are connected to the I2C interface on the PCIE expansion board via a dedicated line for the BMC to monitor and manage the status of the AI / GPU.

[0051] As another example, according to Figure 2As shown, after the server powers on and the BMC completes initialization and file system startup, the BMC detects the presence status of the AI / GPU card via I2C and the GPIO Expander chip. When the AI / GPU card is present, it matches the channel where the AI / GPU card is located with the overall airflow and fans to perform appropriate fan control. When the server is detected to be powered on, the BMC reads the AI / GPU card ID and other hardware and software information via I2C, and intelligently controls the fans based on real-time monitored temperature information, AI / GPU card power consumption information, and corresponding channel fan information. The detected AI / GPU card information can be displayed on the corresponding interface for easy remote management and maintenance.

[0052] To ensure comprehensive firmware development and adaptation, and to prevent BMC management anomalies when expanding with special AI / GPU cards, a pre-defined expansion card information input interface is provided during BMC design and development. Users can input card model, relevant parameters, and other information as instructed in the window. By clicking the input button, BMC will automatically classify and store the GPU card information for easy retrieval or use later.

[0053] As an example, the system's collaborative workflow is as follows: 1. Initialization phase: - After powering on, the BMC scans the PCIe slots and reads the VID / PID of the installed GPU via I2C; - Match the database; if an unregistered model is detected, an on-screen alarm will be triggered to prompt for input.

[0054] 2. Runtime control: - Collect the temperature of the tank every 2 seconds and dynamically calculate the target fan speed; - When a user adds GPU parameters, the corresponding slot monitoring will be automatically activated.

[0055] 3. Troubleshooting: - When the temperature exceeds the limit, mark the location of the slot and highlight the alarm on the web interface; -Supports remote manual overlay speed adjustment strategy.

[0056] In summary, domestically produced AI servers use domestically produced CPUs and GPUs. Due to various limitations of domestically produced chips, the power consumption of domestic CPUs and GPUs is relatively high. General-purpose AI servers can support up to 8 or even 10 AI / GPU cards. Different AI / GPU cards can be matched or mixed according to different application scenarios. The full-configuration design of the whole machine has high power consumption and strong system heat dissipation capacity. Considering the actual energy efficiency ratio of the equipment and effectively reducing the noise of the whole machine, it is necessary to monitor the temperature of key chips and the internal temperature of the chassis in real time according to the power consumption of CPU and GPU, and intelligently adjust the fan speed to effectively control the noise of the equipment and improve the energy efficiency ratio of the equipment.

[0057] The intelligent management method for domestically produced AI servers utilizes domestically produced CPUs and BMC chips. In the design of the AI ​​server's PCIe expansion board, the PCIe card expansion slots are reserved for PCIe device presence detection signals and I2C buses. These signals are connected to the BMC chip on the motherboard via board-level circuitry and dedicated signal lines. The BMC software is adapted to commonly used AI / GPU cards, categorizing and storing the ID information, power consumption, and other parameters of different PCIe cards. GPU-layer fan control strategies are formulated based on power consumption curves. Simultaneously, a web interface is provided for inputting AI / GPU card information, which is then categorized and stored after input.

[0058] After powering on, the BMC obtains the ID information of the configured AI / GPU card via I2C. Based on the interface display or user requirements, it calls other parameters and power consumption information, monitors the temperature information of the server motherboard CPU and AI / GPU in real time, and dynamically adjusts the fan speed on the corresponding channel according to the correspondence between the AI / GPU card slot and the overall fan. Especially when the AI / GPU card is not fully configured, it can effectively reduce noise and improve the energy efficiency ratio of the device compared with the traditional overall fan control method.

[0059] This disclosure provides a computer-readable storage medium storing computer-executable instructions, which are configured to execute the dynamic control and flexible adaptation steps in the above-described intelligent management and control method for domestically produced AI servers.

[0060] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0061] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code; it can also be a transient storage medium.

[0062] The foregoing description and accompanying drawings fully illustrate embodiments of the present disclosure to enable those skilled in the art to practice them. Other embodiments may include structural and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included or substituted for parts and features of other embodiments. Embodiments of the present disclosure are not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from its scope. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for intelligent management and control of domestically produced AI servers, characterized in that, include: Hardware architecture: Each slot on the PCIe expansion board has a dedicated interface that integrates in-situ detection pins and the I2C bus; Direct signal connection: The in-situ detection pin is directly connected to the I2C bus to the motherboard BMC chip via a dedicated cable; Dynamic control: BMC generates fan speed control commands based on real-time slot data; Flexible adaptation: BMC provides an interface for inputting GPU parameters and storing them in a classification database.

2. The method according to claim 1, characterized in that, The dedicated interface includes: The in-situ detection circuit includes a first pin, a second pin, and a third pin. The first pin is connected to the power supply to provide the operating voltage for the detection circuit. The second pin is connected to the ground line to form a current loop reference. The third pin outputs high and low level signals. It outputs a high level when the graphics processor is inserted into the slot and a low level when it is not inserted. The dual-channel I2C bus includes a fourth pin and a fifth pin. The fourth pin serves as a serial data signal line, transmitting the temperature and power consumption data of the graphics processor; the fifth pin serves as a serial clock signal line, synchronizing the data communication timing.

3. The method according to claim 2, characterized in that, The dedicated cable is a silver-plated shielded cable with a length of ≤15cm and a signal delay of <1ms.

4. The method according to claim 1, 2 or 3, characterized in that, The dynamic control includes: Establish a mapping relationship between GPU slots and fan groups; The target rotation speed is dynamically calculated based on the power consumption curve of a domestically produced GPU. The calculation formula is as follows: Target speed = reference speed + (real-time temperature - reference temperature) × adjustment coefficient.

5. The method according to claim 4, characterized in that, The adjustment coefficient is 5% / ℃, the reference temperature is 60℃, and the reference speed is 30%.

6. The method according to claim 5, characterized in that, The elastic adaptation includes: The BMCWeb interface allows input fields for: GPU model, TDP power consumption, and maximum temperature. Automatically categorize cards into training cards or inference cards and bind them to preset strategy templates.

7. A domestically developed intelligent management and control system for AI servers, characterized in that, include: Signal acquisition module: Obtains GPU temperature and on-state status via direct connection to the I2C bus through PCIe slot; Dynamic mapping module: Binds fan groups to the physical location of slots and generates a slot-fan control matrix; Elastic database: Stores user-entered GPU parameters and associated cooling strategies.

8. The system according to claim 7, characterized in that, The elastic database is an embedded SQLite database, and the index fields include: GPU model, category, TDP power consumption, and policy ID.

9. The system according to claim 7, characterized in that, The dynamic mapping module performs the following process: a) Detect the GPU slot's in-slot status; b) Match the air duct area to which the slot belongs; c) Bind the controllable fan group for this area; d) Adjust fan speed based on real-time temperature.

10. A storage medium storing program instructions, characterized in that, When the program instructions are executed, they perform the dynamic control and flexible adaptation steps of the intelligent management and control method for domestic AI servers as described in any one of claims 1 to 6.

Images