A method for determining GPU power consumption and a computing device

By utilizing the power consumption value and power conversion factor of the hot-swap controller, the power consumption value of the GPU is estimated, solving the problem that the GPU power consumption cannot be queried immediately after the operating system starts, and realizing real-time power consumption monitoring.

CN118193312BActive Publication Date: 2026-11-17XFUSION DIGITAL TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410205522.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-23
Publication Date
2026-11-17
Estimated Expiration
2044-02-23

AI Technical Summary

Technical Problem

In the existing technology, GPU power consumption acquisition depends on the loading of the operating system's driver, which results in a 1-2 minute delay after the operating system starts up, making it impossible to query GPU power consumption immediately, and invalid values ​​cannot monitor GPU power consumption.

Method used

By combining the power consumption value of the hot-swappable controller connected to the GPU with the power conversion factor and the power consumption of other devices, the power consumption value of the GPU can be estimated, enabling monitoring of GPU power consumption during the operating system startup process.

Benefits of technology

During the operating system startup process, the power consumption of the GPU can be accurately estimated, solving the lag problem and realizing real-time power consumption monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118193312B_ABST
    Figure CN118193312B_ABST
Patent Text Reader

Abstract

This application provides a method, apparatus, and system for determining GPU power consumption. In an embodiment, the method is applied to a computing device including a GPU and a GPU substrate management controller. The GPU substrate management controller monitors a first power consumption value of a hot-swap controller connected to the GPU and a second power consumption value of the GPU. The method includes: accessing the GPU substrate management controller to obtain the first and second power consumption values ​​at a given time point; and determining a target power consumption value of the GPU at the given time point based on the first power consumption value when the second power consumption value at that time point is invalid. This allows for estimation of the GPU's power consumption value during OS startup if the GPU's power consumption value is invalid, thereby enabling querying of GPU power consumption during OS startup.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of server technology, and in particular to a method for determining GPU power consumption and a computing device. Background Technology

[0002] A GPU server consists of a GPU substrate, which integrates multiple GPUs. These GPUs have high power consumption, so it is urgent to monitor and manage their power consumption to ensure the safe operation of the GPU server.

[0003] The GPU substrate comes with a built-in Hardware Management Console (HMC), which can monitor and manage the GPU substrate. Currently, the GPU substrate has implemented GPU power consumption monitoring using a GPU power consumption sensor, and can provide an external Reddit interface to query the GPU power consumption collected by the GPU power consumption sensor.

[0004] However, GPU power consumption measurement depends on the loading of the GPU driver by the operating system (OS). If this condition is not met, the GPU power sensor will display an invalid value (NA). Furthermore, the OS needs time to load the GPU driver, resulting in a 1-2 minute lag between OS startup and the display of a valid value by the GPU power sensor. Therefore, GPU power consumption cannot be queried immediately after OS startup. Summary of the Invention

[0005] This application provides a method and computing device for determining GPU power consumption. When the GPU power consumption value is invalid during OS startup, the GPU power consumption value can be estimated by using the power consumption value of the hot-swap controller connected to the GPU, thereby allowing the GPU power consumption to be queried during OS startup.

[0006] In a first aspect, embodiments of this application provide a GPU power consumption determination method applied to a computing device. The computing device includes a GPU and a GPU substrate management controller. The GPU substrate management controller monitors a first power consumption value of a hot-plug controller connected to the GPU and a second power consumption value of the GPU. The method includes: accessing the GPU substrate management controller to obtain the first power consumption value and the second power consumption value at a sampling time point; and when the second power consumption value at the sampling time point is invalid, determining a target power consumption value of the GPU at the sampling time point based on the first power consumption value at the sampling time point.

[0007] In this solution, if the GPU power consumption value is invalid during OS startup, the power consumption value of the GPU can be estimated by using the power consumption value of the hot-swappable controller connected to the GPU, thus allowing the GPU power consumption to be queried during OS startup.

[0008] In one possible implementation, the method further includes: when the second power consumption value at the acquisition time point is valid, using the second power consumption value at the acquisition time point as the target power consumption value of the GPU.

[0009] In one possible implementation, the target power consumption value of the GPU at the acquisition time point is determined based on the first power consumption value at the acquisition time point, including: when the n consecutively acquired second power consumption values ​​up to the acquisition time point are invalid, the target power consumption value of the GPU at the acquisition time point is obtained based on the first power consumption value at the acquisition time point; n is a positive integer greater than or equal to 2.

[0010] In one possible implementation, the target power consumption of the GPU at the acquisition time point is the first power consumption value at the acquisition time point.

[0011] In one possible implementation, the target power consumption of the GPU at the acquisition time point is obtained based on the first power consumption value at the acquisition time point, including: determining a first parameter value and a second parameter value, wherein the first parameter value is the conversion coefficient of power transfer from the hot-swap controller to the GPU; the second parameter value is the power consumption of other devices besides the GPU among the devices connected to the hot-swap controller; and the target power consumption of the GPU at the acquisition time point is estimated based on the first parameter value, the second parameter value, and the first power consumption value at the acquisition time point.

[0012] In this solution, by taking into account the power conversion factor and the power consumption of devices other than the GPU connected by the hot-swappable controller, the power consumption of the GPU can be estimated relatively accurately.

[0013] In one possible implementation, the target power consumption value of the GPU at the acquisition time point is obtained based on the first power consumption value at the acquisition time point, including: when some of the n second power consumption values ​​collected consecutively up to the acquisition time point are valid, for invalid second power consumption values ​​among the n second power consumption values, the second power consumption value is modified based on the first power consumption value at the acquisition time point of the second power consumption value; n is a positive integer greater than or equal to 2; the n second power consumption values ​​after modification of the invalid second power consumption value are weighted and averaged to obtain the target power consumption value of the GPU at the acquisition time point.

[0014] In this solution, by modifying multiple invalid second power consumption values ​​and smoothing out the valid and modified second power consumption values, the power consumption of the GPU can be estimated more accurately.

[0015] In one possible implementation, the computing device includes a processor, which is at least used to run a driver for the GPU to obtain a power consumption value for the GPU. A second power consumption value for the GPU is obtained from the processor by the GPU substrate management controller. The second power consumption value for the GPU becomes effective after the processor runs the GPU driver.

[0016] In one possible implementation, the GPU power consumption determination method is performed by the Baseboard Management Controller (BMC) in the computing device.

[0017] Secondly, embodiments of this application provide a GPU power consumption determination device, which includes several modules. Each module is used to execute various steps in the GPU power consumption determination method provided in the first aspect of this application. The division of modules is not limited here. For the specific functions performed by each module of this GPU power consumption determination device and the beneficial effects achieved, please refer to the functions of each step in the GPU power consumption determination method provided in the first aspect of this application; further details will not be repeated here.

[0018] For example, a GPU power consumption determination device includes:

[0019] The access module is used to access the GPU substrate management controller to obtain the first power consumption value and the second power consumption value at the acquisition time point; the GPU substrate management controller is used to monitor the first power consumption value of the hot-swap controller connected to the GPU and the second power consumption value of the GPU.

[0020] The estimation module is used to determine the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the second power consumption value at the acquisition time point is invalid.

[0021] In this solution, if the GPU power consumption value is invalid during OS startup, the power consumption value of the GPU can be estimated by using the power consumption value of the hot-swappable controller connected to the GPU, thus allowing the GPU power consumption to be queried during OS startup.

[0022] In one possible implementation, the device further includes a power determination module, used to take the second power consumption value at the acquisition time as the target power consumption value of the GPU when the second power consumption value at the acquisition time is valid.

[0023] In one possible implementation, the estimation module is used to obtain the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the n consecutively acquired second power consumption values ​​up to the acquisition time point are invalid; n is a positive integer greater than or equal to 2.

[0024] In one possible implementation, the target power consumption of the GPU at the acquisition time point is the first power consumption value at the acquisition time point.

[0025] In one possible implementation, an estimation module is used to determine a first parameter value and a second parameter value. The first parameter value is the conversion coefficient of power transfer from the hot-swap controller to the GPU. The second parameter value is the power consumption of other devices besides the GPU among the devices connected to the hot-swap controller. Based on the first parameter value, the second parameter value, and the first power consumption value at the acquisition time point, the target power consumption value of the GPU at the acquisition time point is estimated.

[0026] In this solution, by taking into account the power conversion factor and the power consumption of devices other than the GPU connected by the hot-swappable controller, the power consumption of the GPU can be estimated relatively accurately.

[0027] In one possible implementation, the estimation module is used to modify the second power consumption value based on the first power consumption value at the acquisition time when the second power consumption value among the n second power consumption values ​​collected continuously up to the acquisition time is partially valid; n is a positive integer greater than or equal to 2; and the n second power consumption values ​​after modification of the invalid second power consumption value are weighted and averaged to obtain the target power consumption value of the GPU at the acquisition time.

[0028] In this solution, by modifying multiple invalid second power consumption values ​​and smoothing out the valid and modified second power consumption values, the power consumption of the GPU can be estimated more accurately.

[0029] In one possible implementation, a GPU substrate management controller is connected to a processor, which is at least used to run a GPU driver to obtain a power consumption value for the GPU. A second power consumption value for the GPU is obtained by the GPU substrate management controller from the processor. The second power consumption value for the GPU is valid after the processor runs the GPU driver.

[0030] Thirdly, this application provides a computing device, which includes: a baseboard management controller (BMC), a GPU, a GPU baseboard management controller, and a processor; the GPU baseboard management controller is used to monitor a first power consumption value of a hot-swap controller connected to the GPU and obtain a second power consumption value of the GPU from the processor.

[0031] BMC is used to access the GPU substrate management controller to obtain the first and second power consumption values ​​at the acquisition time point;

[0032] BMC is used to determine the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the second power consumption value at the acquisition time point is invalid.

[0033] In one possible implementation, the processor is at least used to run the GPU driver to obtain the GPU's power consumption value, and the second power consumption value of the GPU becomes effective after the processor runs the GPU driver.

[0034] In one possible implementation, the BMC is used to take the second power consumption value at the acquisition time as the target power consumption value of the GPU when the second power consumption value at the acquisition time is valid.

[0035] In one possible implementation, the BMC is used to obtain the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the n consecutively acquired second power consumption values ​​up to the acquisition time point are invalid; n is a positive integer greater than or equal to 2.

[0036] In one possible implementation, the target power consumption of the GPU at the acquisition time point is the first power consumption value at the acquisition time point.

[0037] In one possible implementation, the BMC is used to determine a first parameter value and a second parameter value. The first parameter value is the conversion factor of the power transferred from the hot-swap controller to the GPU. The second parameter value is the power consumption of other devices besides the GPU among the devices connected to the hot-swap controller. Based on the first parameter value, the second parameter value, and the first power consumption value at the acquisition time point, the target power consumption value of the GPU at the acquisition time point is estimated.

[0038] In this solution, by taking into account the power conversion factor and the power consumption of devices other than the GPU connected by the hot-swappable controller, the power consumption of the GPU can be estimated relatively accurately.

[0039] In one possible implementation, the BMC is used to modify the second power consumption value based on the first power consumption value at the acquisition time when the second power consumption value among the n second power consumption values ​​collected continuously up to the acquisition time is partially valid; n is a positive integer greater than or equal to 2; and the n second power consumption values ​​after modification of the invalid second power consumption value are weighted and averaged to obtain the target power consumption value of the GPU at the acquisition time.

[0040] In this solution, by modifying multiple invalid second power consumption values ​​and smoothing out the valid and modified second power consumption values, the power consumption of the GPU can be estimated more accurately.

[0041] Fifthly, embodiments of this application provide a GPU power consumption determination apparatus, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method provided in the first aspect.

[0042] In a sixth aspect, embodiments of this application provide a GPU power consumption determination apparatus, characterized in that the apparatus executes computer program instructions to perform the method provided in the first aspect. Exemplarily, the apparatus may be a chip such as a BMC chip, or a processor.

[0043] In one example, the device may include a processor that can be coupled to memory, read instructions from the memory, and execute the methods provided in the first aspect according to those instructions. The memory may be integrated into the chip or processor, or it may be independent of the chip or processor.

[0044] In a seventh aspect, embodiments of this application provide a computer storage medium storing instructions that, when executed on a computer, cause the computer to perform the method provided in the first aspect.

[0045] Eighthly, embodiments of this application provide a computer program product containing instructions that, when executed on a computer, cause the computer to perform the method provided in the first aspect. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the structure of the computing device provided in the embodiments of this application;

[0047] Figure 2 This is a schematic diagram showing the connection relationship between the GPU substrate management controller 130, the processor 111, and the BMC 114 provided in the embodiments of this application;

[0048] Figure 3 This is a flowchart illustrating the GPU power consumption determination method provided in the embodiments of this application. Figure 1 ;

[0049] Figure 4 This is a schematic diagram of a specific scheme for the GPU power consumption determination method provided in the embodiments of this application;

[0050] Figure 5 This is a flowchart illustrating the GPU power consumption determination method provided in the embodiments of this application. Figure 2 ;

[0051] Figure 6 This is a schematic diagram illustrating a specific application of the GPU power consumption determination method provided in the embodiments of this application;

[0052] Figure 7 This is a communication diagram of the GPU power consumption determination device provided in the embodiments of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be described below with reference to the accompanying drawings.

[0054] In the description of the embodiments of this application, the words "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a specific manner.

[0055] In the description of the embodiments in this application, the term "and / or" is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, B existing alone, and A and B existing simultaneously. Furthermore, unless otherwise stated, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple terminals refer to two or more terminals.

[0056] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0057] The following explanations cover some of the terms used in this embodiment. It should be noted that these explanations are for the convenience of those skilled in the art and do not constitute a limitation on the scope of protection claimed in this application.

[0058] The Baseboard Management Controller (BMC) is a dedicated controller for monitoring and managing servers. Its four main functions are: ① Device Information Management: Recording server information (model, manufacturer, date, production and technical information of each component, chassis information, motherboard information, etc.) and BMC information (server hostname, IP address, BMC firmware version, etc.); ② Server Status Monitoring and Management: Monitoring the health status of various server components (CPU, memory, hard drive, fans, chassis, etc.) such as temperature and voltage, and adjusting fan speeds in real time based on temperature data to prevent overheating and control overall power consumption; if any abnormality occurs in a single board component, the information is promptly reported to the upper-level network management system via various industry-standard protocols such as SNMP, SMTP, and Redfish; ③ Remote Control Management of the Server: Server power on / off, restart, maintenance, firmware updates, system installation, etc.; ④ Maintenance Management: Log management, user management, BIOS management, alarm management, etc. The BMC consists of two parts: the BMC chip and the BMC firmware. The BMC chip is a chip integrated on the motherboard or a device plugged into the motherboard via PCIe or similar means. This chip acts as the CPU for the BMC firmware, and it is equipped with its own RAM, Flash memory, and other peripheral devices. As soon as the server is plugged in, the BMC firmware starts running quickly. BMC firmware is the software program running on the BMC chip; it can be understood as software that runs as soon as the server is powered on. It does not depend on other hardware on the system (such as the CPU or memory), nor on the BIOS or OS. However, the BMC can interact with the BIOS and OS, enabling better platform management. The OS has system management software that can work in conjunction with the BMC to achieve better management results. Therefore, it can be seen that the BMC is a small operating system independent of the server system, responsible for remote server control, management, and monitoring.

[0059] Graphics Processing Unit (GPU): Also known as display core, visual processor, or display chip, it is a microprocessor specifically designed to perform image and graphics-related calculations on personal computers, workstations, game consoles, and some mobile devices (such as tablets and smartphones).

[0060] GPU Baseboard Management Controller: This controller provides monitoring, control, and firmware update capabilities. For example, it could be the Hardware Management Console (HMC) provided by the NVIDIA Hopper HGX 8 GPU baseboard. The HMC offers Redfish-based monitoring, control, and firmware update functions, with advanced maintainability and debuggability. It can be developed separately from the host BMC software.

[0061] A hot-swap controller (HSC) is a circuit or chip used to control and manage the connection and disconnection process of hot-swap devices. Hot-swapping refers to inserting or removing a device while the power is on, without turning off the power or affecting other operating devices. The main functions of an HSC include power management, current protection, hot-swap detection, and status indication. It can control the current when inserting or removing devices, ensuring safe insertion and removal and avoiding current surges, while also providing monitoring and signal indication of device status.

[0062] PCIe (Peripheral Component Interconnect Express) is a high-speed serial computer expansion bus standard. It supports 256 buses, each bus can support up to 32 devices, and each device can support up to 8 functions. Therefore, the BDF (Bus, Device, Function) constitutes the identification number of each device connected to the PCIe bus (for ease of description and distinction, it is called a PCIe device). Specifically, when using the PCIe protocol, many types of PCIe devices can be connected to the PCIe bus in a server, including hard drives, network cards, and controllers. The PCIe bus is a high-speed serial point-to-point dual-channel high-bandwidth transmission. Each connected PCIe device is allocated its own dedicated channel bandwidth and does not share the bus bandwidth. In actual board design, it is necessary to determine the bandwidth required by the device and configure it correctly in the BIOS.

[0063] A device driver is a special program that enables communication between a computer and a device. The primary function of a driver is to facilitate data transfer between the computer system and hardware devices. Only with the help of a driver can the two communicate and perform specific functions. A driver acts as a mediator between the operating system and the hardware, enabling bidirectional communication. It conveys the functionality of the hardware device to the operating system, while simultaneously transmitting standard operating system instructions to the hardware device.

[0064] Operating System (OS): Located on the server's motherboard, the OS contains GPU drivers that can obtain GPU information, such as power consumption. The motherboard communicates with the hard drives and BMC (Browser Controller Center) connected to the hard drive backplane via PCIe (Peripheral Component Interconnect Express) high-speed serial bus. Each motherboard can communicate with multiple hard drive backplanes.

[0065] Universal Serial Bus (USB) is a serial bus standard and a technical specification for input / output interfaces. It is widely used in information and communication products such as personal computers and mobile devices, and has been extended to other related fields such as photographic equipment, digital televisions (set-top boxes), and game consoles.

[0066] This application provides a computing device. This computing device can be a server or a computing node, with the computing node possessing complete server functionality.

[0067] For example, such as Figure 1 As shown, the computing device 100 may include a motherboard 110, a power supply 120, and a GPU substrate 130. The power supply 120 is electrically connected to the motherboard 110 and is used to supply power to the devices connected to the motherboard 110.

[0068] For example, the devices connected to the motherboard 110 include a processor 111, memory 112, a programmable logic device (PLD) 113, a baseboard management controller (BMC) 114, a PCIe slot 115, a fan 116, a hard drive 117, and a GPU substrate 130.

[0069] The processor 111 may include a central processing unit (CPU), as well as other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0070] The memory 112 can be represented as a memory module, inserted into a memory slot on the motherboard 110, and is used for external cache. Exemplarily, the memory 112 can be random access memory (RAM). By way of example, but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM).

[0071] Among them, the programmable logic device 113 can be a complex programmable logic device (CPLD, which is a digital integrated circuit in which the user constructs logic functions according to their own needs) or a field programmable gate array (FPGA).

[0072] Among them, PCIe slot 115 is suitable for expanding at least one of GPU cards, network cards, video capture cards, HBA (Host Bus Adapter) cards, RAID (redundant arrays of independent disks) cards, and SSD (solid state disk or solid state drive), and can also support the expansion of various adapter cards.

[0073] Among them, hard disk 117 can be either a hard disk drive (HDD) or a solid state drive (SSD). For example, the SSD can be a PCIe-based SSD, such as an NVMe (Non-Volatile Memory Host Controller Interface Specification) SSD for accessing non-volatile memory media attached via the PCIe bus. It should be understood that hard disk 123 is merely an example of non-volatile memory and does not constitute a specific limitation. In practical applications, non-volatile memory can be selected based on the actual situation.

[0074] Among them, such as Figure 2 As shown, the GPU substrate 130 includes several GPUs 131 and one GPU substrate management controller 132, wherein the GPU substrate management controller 132 is connected to the several GPUs 131. The GPUs are connected to the processor 111 via a bus such as a PCIe bus, and the GPU substrate management controller 132 is connected to the processor 111 via a bus (such as a PCIe bus). The GPU substrate management controller 132 can also be connected to the BMC 114 via a bus (such as a USB bus). The processor 111 can load the driver for the GPU 131. After loading the driver for the GPU 131, it can run a power monitoring service to monitor the power consumption of the GPU 131 via a bus (such as a PCIe bus). Correspondingly, the GPU substrate management controller 132 can obtain and record the power consumption value of the GPU 131 monitored by the processor 111 via the bus (such as a PCIe bus). When the processor 111 has not loaded the driver for the GPU 131, the power consumption value of the GPU 131 obtained by the GPU substrate management controller 132 is invalid. When the processor 111 loads the driver for the GPU 131, the power consumption value of the GPU 131 obtained by the GPU substrate management controller 132 is valid.

[0075] Among them, such as Figure 2The GPU substrate management controller 132 may include the HSC and GPU power sensor described above. The HSC is used to connect to GPUs 131. One GPU 131 is connected to one HSC, and the number of HSCs is the same as the number of GPUs 131. The GPU power sensor is a virtual sensor used to monitor the power consumption value of the GPUs 131 obtained by the GPU substrate management controller 132 from the processor 111. The HSC includes a load power monitor, which monitors the power consumption of the HSC. The power consumption of the HSC indicates the total power consumption of all devices connected to the HSC, and the devices connected to the HSC include at least GPUs 131. The GPU substrate management controller 114 can access the GPU substrate management controller 132 via a bus (e.g., a USB bus). The GPU substrate management controller 132 can feed back the power consumption of the HSC collected by the load power monitor and the power consumption value of the GPUs 131 collected by the GPU power sensor to the BMC 114.

[0076] It should be noted that, Figure 1 This is merely an example of computing device 100 and does not constitute a specific limitation. In practical applications, it can include more than Figure 1 More or fewer devices.

[0077] In related technologies, the power consumption of GPU131 depends on the OS loading the GPU131 driver to run the power control service. If the OS does not load the GPU131 driver, the power consumption value of GPU131 monitored by the GPU substrate management controller 132 is an invalid value (NaN). Furthermore, the OS requires a certain amount of time to load the GPU131 driver, resulting in a 1-2 minute lag between the completion of OS startup and the display of a valid value. Therefore, the BMC114 cannot immediately query a valid power consumption value for GPU131 after OS startup.

[0078] Based on this, this application proposes a method for determining GPU power consumption.

[0079] This method estimates the power consumption of the GPU131 by detecting the power consumption of the HSC and the layout of the devices connected to the HSC (including GPU131), and designs a GPU power consumption estimation model. When there is no valid power consumption value for GPU131, the estimated value from the GPU power consumption estimation model is used, thus enabling monitoring of GPU power consumption even when the GPU131 power consumption value is invalid. This is only a brief description of the method; for details, please refer to the description below.

[0080] Next, in conjunction with the computing device provided above, a method for determining GPU power consumption provided in the embodiments of this application will be described in detail.

[0081] Figure 3 This is a flowchart illustrating a GPU power consumption determination method provided in an embodiment of this application. This embodiment can be applied to electronic devices, specifically servers or general computers. Figure 3 As shown in the embodiments of this application, a GPU power consumption determination method includes at least the following steps:

[0082] Step 301: The computing device 100 accesses the GPU substrate management controller 132 to obtain the first power consumption value and the second power consumption value at the acquisition time point. The first power consumption value is the power consumption value of the HSC connected to the GPU 131, and the second power consumption value is the power consumption value of the GPU 131.

[0083] For example, the first power consumption value is obtained by the load power consumption monitor in the HSC, which is used to monitor the power consumption of the HSC.

[0084] For example, the second power consumption value is obtained by a virtual GPU power consumption monitor in the GPU substrate management controller 132, which monitors a valid value when the processor 111 runs the driver for the GPU 131.

[0085] In practice, the computing device 100 collects the power consumption value monitored by the load power consumption monitor and the power consumption value monitored by the GPU power consumption monitor at a preset time interval, such as 100 milliseconds, to obtain the first power consumption value and the second power consumption value at the current collection time point.

[0086] Step 302: When the second power consumption value is invalid, the computing device 100 determines the target power consumption value of the GPU 131 at the acquisition time point based on the first power consumption value at the acquisition time point.

[0087] If the computing device 100 determines that the second power consumption value has no specific value, it may consider the second power consumption value to be invalid. For example, if the computing device 100 determines that the second power consumption value is nano, it indicates that the second power consumption value is invalid. When the second power consumption value is invalid, it means that the power consumption of the GPU 131 cannot be monitored. In this case, it can be estimated using the first power consumption value of the upper-level HSC of the GPU 131.

[0088] When the second power consumption value at the acquisition time point is invalid, there are two ways to obtain the target power consumption value of GPU131 at the acquisition time point based on the first power consumption value at the acquisition time point.

[0089] Implementation Method 1: Determine the first parameter value and the second parameter value. The first parameter value is the conversion coefficient of the power input to the HSC to the GPU 131. The second parameter value is the power consumption of other devices besides the GPU 131 in the devices connected to the HSC. Based on the first parameter value, the second parameter value and the first power consumption value at the acquisition time point, estimate the target power consumption value of the GPU 131 at the acquisition time point.

[0090] Implementation method 2: Use the first power consumption value at the acquisition time point as the target power consumption value of GPU131 at the acquisition time point.

[0091] In implementation method 1, HSC, as the upper-level device of GPU131, manages the power and signals of GPU131. Given the first power consumption value of HSC, the power consumption value of GPU can be estimated according to the device layout design model connected to HSC. Here, it is assumed that HSC Power represents the first power consumption value of HSC, GPU Power represents the second power consumption value of GPU131, k represents the first coefficient, the value of the first coefficient is the first parameter value, b represents the second coefficient, the value of the second coefficient is the second parameter value, and the power consumption estimation model of GPU131 is HSC Power = GPU Power * k + b. For example, assuming that GPU substrate 130 has 8 GPUs131, GPUPower represents the sum of the power consumption values ​​of all GPUs131 on GPU substrate 130, then the relationship between HSC Power and GPU Power can be summarized as formula (1): GPU Power = [(HSC1 Power – b) + … + (HSC8 Power – b)] / k.

[0092] Regarding k and b, considering scenarios where the GPU131 driver may never be installed, there will be no actual feedback on k and b. In this case, computing device 100 needs to preset initial values ​​for k and b. When there is actual feedback on the second power consumption value of GPU131, that is, when the second power consumption value of GPU131 is a valid value, k and b can be solved according to HSC Power = GPU Power * k + b to update k and b. Subsequently, the updated k and b are used to estimate the power consumption of GPU131.

[0093] In particular, in some actual testing scenarios, since the power consumption of GPU131 is much higher than that of other devices in the HSC-connected device group besides GPU131, the value of b is about 0. In this case, the relationship between GPU Power and HSC Power can be simplified to HSC Power = GPU Power * k.

[0094] Specifically, when the value of k is approximately 1, the relationship between GPU Power and HSC Power can be simplified to HSC Power = GPUPower. For example, assuming that the GPU substrate 130 has 8 GPUs 131, GPU Power represents the sum of the power consumption values ​​of all GPUs 131 on the GPU substrate 130. Then, the relationship between HSC Power and GPU Power can be summarized as formula (2): GPU Power = HSC1 Power + ... + HSC8 Power. Correspondingly, in this scenario, the target power consumption value of GPU 131 can be determined using implementation method 2.

[0095] In some possible implementations, when the second power consumption value is invalid, the computing device 100 obtains the target power consumption value of the GPU 131 at the acquisition time based on the first power consumption value at the acquisition time when the n (positive integers greater than or equal to 2, for example, 10) second power consumption values ​​collected consecutively up to the acquisition time are invalid.

[0096] The n consecutive second power consumption values ​​collected up to the current acquisition time can be understood as the n most recent second power consumption values ​​collected at preset time intervals. For example, assuming the second power consumption values ​​are collected continuously at 100-millisecond intervals, the second power consumption values ​​at acquisition time points t0, t1, ..., ti are obtained, where ti represents the current acquisition time point. Assuming n = 10, the 10 consecutive second power consumption values ​​collected up to the current acquisition time point are the second power consumption values ​​t(i-9), t(i-8), ..., t(i-1), and ti.

[0097] In some other possible implementations, when the second power consumption value is invalid, the computing device 100 modifies the invalid second power consumption value among the n (positive integers greater than or equal to 2) second power consumption values ​​that have been continuously collected up to the collection time point when they are valid. Then, it performs a weighted average of the n second power consumption values ​​after the invalid second power consumption value is modified to obtain the target power consumption value of the GPU 131 at the collection time point.

[0098] It should be noted that in this embodiment of the application, for each acquisition, a first power consumption value and a second power consumption value need to be acquired; therefore, at the acquisition time of the second power consumption value, the first power consumption value will also be acquired; correspondingly, the first power consumption value at the acquisition time of the second power consumption value can be understood as the first power consumption value acquired at the acquisition time of the second power consumption value.

[0099] When the computing device 100 determines that the second power consumption value has a specific value, it can consider the second power consumption value to be valid.

[0100] The specific implementation method for modifying the invalid second power value based on the first power value at the time point of the second power value acquisition is as follows: calculate the estimated value of the GPU based on the first power value at the time point of the second power value acquisition (for details, please refer to the description of the aforementioned implementation method 1 and implementation method 2), and use the estimated value of the GPU as the modified second power value.

[0101] In one example, the weighting coefficient in the weighted average can be 1. The specific implementation of the weighted average of the n second power consumption values ​​after the invalid second power consumption value is modified to obtain the target power consumption value of the GPU at the acquisition time is as follows: The computing device 100 averages the n second power consumption values ​​after the invalid second power consumption value is modified to obtain the target power consumption value of the GPU 131 at the acquisition time.

[0102] For example, such as Figure 4 As shown, when the computing device 100 detects that the OS is powered on, it collects the first power consumption value monitored by the load power consumption monitor (HSC Power value) and the second power consumption value monitored by the GPU power consumption monitor (GPU Power value) at preset time intervals. It then determines whether the most recently collected n GPU Power values ​​are invalid. If so, the current GPU Power value is an estimate based on the current HSC Power value, for example, GPU Power = HSC Power, or GPU Power = HSC Power / k, or GPU Power = (HSC Power - b) / k. If not, it determines whether the most recently collected n GPU Power values ​​are valid. If so, the current GPU Power value is the last collected GPU Power value. If not, it determines whether the current GPU Power value is valid. If valid, the current GPU Power value is the GPU Power value collected at the current moment. If invalid, it sequentially determines whether each of the n GPU Power values ​​is valid. If valid, the GPU Power value is retained; otherwise, the GPU Power value is modified to the value collected at the corresponding time (and GPU Power value). The estimated value of the HSCPower value is obtained at the same time as the Power value acquisition time. For example, GPU Power = HSC Power, or GPU Power = HSC Power / k, or GPU Power = (HSC Power - b) / k. Then, the average of the n GPU Power values ​​after modifying the invalid GPU Power value is used to obtain the GPU Power value at the current time.

[0103] Step 303: When the second power consumption value of the computing device 100 at the acquisition time point is valid, the second power consumption value at the acquisition time point shall be used as the target power consumption value of the GPU 131 at the acquisition time point.

[0104] When the computing device 100 determines that the second power consumption value has a specific value, it can consider the second power consumption value to be valid. For example, when the computing device 100 determines that the second power consumption value is valid, it means that the OS has completed power-on and loaded the driver for the GPU 131. At this time, the second power consumption value can represent the actual power consumption of the GPU 131, and the second power consumption value can be directly used as the target power consumption value of the GPU 131.

[0105] In this solution, when the GPU power consumption is invalid, the power consumption of the GPU is estimated based on the power consumption of the HSC connected to the GPU, so that the power consumption of the GPU can be monitored when the GPU power consumption is invalid.

[0106] Figure 3 The embodiments shown are merely basic embodiments of the method described in this application. Other preferred embodiments of the method can be obtained by making certain optimizations and extensions based on them.

[0107] In one embodiment, based on the foregoing embodiments, this embodiment provides a more specific description and a certain degree of optimization. This embodiment uses the interaction between the BMC and the GPU substrate management controller 132 as an example for description. For example... Figure 5 As shown, the method in this embodiment includes:

[0108] Step 501: BMC114 accesses GPU substrate management controller 132 to obtain the first power consumption value and the second power consumption value at the acquisition time point. The first power consumption value is the power consumption value of the HSC connected to GPU131, and the second power consumption value is the power consumption value of GPU131.

[0109] In some possible scenarios, when the GPU substrate management controller 132 is an HMC, the BMC114 can access the REDFISH interface to access the HMC and obtain the first and second power consumption values ​​at the acquisition time point.

[0110] For details, please refer to the description of step 301 above, which will not be repeated here.

[0111] Step 502: When the second power consumption value is invalid, BMC114 determines the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point.

[0112] For details, please refer to the description of step 302 above, which will not be repeated here.

[0113] Step 503: When the second power consumption value at the acquisition time point is valid, BMC114 uses the second power consumption value at the acquisition time point as the target power consumption value of the GPU at the acquisition time point.

[0114] For details, please refer to the description of step 303 above, which will not be repeated here.

[0115] Based on the GPU power consumption determination method provided above, a specific application of the GPU power consumption determination method will be explained.

[0116] In this embodiment of the application, it is assumed that the estimated value of GPU Power is gpu_simu=(HSC1 Power+…+HSC8Power) / K, and the effective value of GPU Power is gpu_samp, which represents the read value of the Redfish interface of HMC.

[0117] In this embodiment of the application, when calculating the GPU Power value, different processing procedures are required depending on the number of valid values ​​collected. There are three processing procedures, as follows:

[0118] (1) If n consecutive data collections are all invalid values, the estimated value gpu_simu is used;

[0119] 2) If n consecutive data samples are collected, and all are valid values, then the collected value gpu_samp is used:

[0120] 3) Collect data continuously for n times. If the GPU Power value collected in the last collection is invalid, and there are both valid and invalid values ​​in the n collections, the invalid value is replaced with the estimated value gpu_simu. Then, the GPU Power values ​​collected in the n collections after replacing the invalid value with gpu_simu are averaged to achieve a smooth transition between the estimated value and the true value.

[0121] Figure 6 This is a flowchart illustrating a specific application of the GPU power consumption determination method provided in this application. Figure 6 As shown, specifically, when the computing device 100 is powered on, the GPU baseboard management controller 132 detects that the HSC power value is valid and the GPU power value is nan; correspondingly, the BMC 114 collects the HSC power value as nan and the GPU power value as nan.

[0122] Next, the server's OS is powered on, meaning the OS starts running. At this time, the HSC power value collected by the BMC114 is a valid value, and the GPU power value is the current gpu_simu value.

[0123] Then, the OS enables the GPU131 driver to remain resident. The GPU substrate management controller 132 waits for 1-2 minutes and detects that the GPU power value is valid. At this time, some of the n GPU power values ​​recently collected by the BMC114 are NaN (invalid), so the current GPU power value is the average of the n most recently collected GPU power values ​​(composed of gpu_simu and gpu_samp). After a period of time, the n most recently collected GPU power values ​​by the BMC114 are valid, so the current GPU power value is the current gpu_samp. In addition, k can be updated to (HSC1 Power + ... + HSC8 Power) / gpu_samp.

[0124] Next, the OS disables the GPU131 driver from being resident. The GPU baseboard management controller 132 detects that the HSC power value is valid and the GPU power value is nan (invalid). Correspondingly, some of the n GPU power values ​​recently collected by the BMC114 are nan (invalid). If the current GPU power value is valid, it is not modified. If the current GPU power value is invalid, the current GPU power value is modified to the average of the n most recently collected GPU power values ​​(composed of gpu_simu and gpu_samp). After a period of time, the n most recently collected GPU power values ​​by the BMC114 are na n, and the current GPU power value is the current gpu_simu.

[0125] Based on the same concept as the method embodiments of this application, this application also provides a GPU power consumption determination device. The GPU power consumption determination device includes several modules, each module being used to execute various steps in the GPU power consumption determination method provided in this application embodiment. The division of modules is not limited here. Those skilled in the art will clearly understand that in practical applications, the various steps in the GPU power consumption determination method provided in this application embodiment can be assigned to different modules as needed, that is, the internal structure of the device can be divided into different modules to complete all or part of the functions described above. The modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the modules in the above device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0126] For example, the GPU power consumption determination device is used to execute a GPU power consumption determination method provided in the embodiments of this application. Figure 7 This is a schematic diagram of the GPU power consumption determination device provided in an embodiment of this application. Figure 7 As shown, the GPU power consumption determination device provided in this application embodiment includes:

[0127] Access module 701 is used to access the GPU substrate management controller to obtain the first power consumption value and the second power consumption value at the acquisition time point; the GPU substrate management controller is used to monitor the first power consumption value of the hot-swap controller connected to the GPU and the second power consumption value of the GPU.

[0128] The estimation module 702 is used to determine the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the second power consumption value at the acquisition time point is invalid.

[0129] In this solution, if the GPU power consumption value is invalid during OS startup, the power consumption value of the GPU can be estimated by using the power consumption value of the hot-swappable controller connected to the GPU, thus allowing the GPU power consumption to be queried during OS startup.

[0130] In one possible implementation, the device further includes a power determination module, used to take the second power consumption value at the acquisition time as the target power consumption value of the GPU when the second power consumption value at the acquisition time is valid.

[0131] In one possible implementation, the estimation module 702 is used to obtain the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the n consecutively acquired second power consumption values ​​up to the acquisition time point are invalid; n is a positive integer greater than or equal to 2.

[0132] In one possible implementation, the target power consumption of the GPU at the acquisition time point is the first power consumption value at the acquisition time point.

[0133] In one possible implementation, the estimation module 702 is used to determine a first parameter value and a second parameter value, wherein the first parameter value is the conversion coefficient of power transfer from the hot-swap controller to the GPU; the second parameter value is the power consumption of other devices besides the GPU in the devices connected to the hot-swap controller; and the target power consumption value of the GPU at the acquisition time point is estimated based on the first parameter value, the second parameter value and the first power consumption value at the acquisition time point.

[0134] In this solution, by taking into account the power conversion factor and the power consumption of devices other than the GPU connected by the hot-swappable controller, the power consumption of the GPU can be estimated relatively accurately.

[0135] In one possible implementation, the estimation module 702 is used to modify the second power consumption value based on the first power consumption value at the acquisition time when the second power consumption value among the n second power consumption values ​​collected continuously up to the acquisition time is partially valid; n is a positive integer greater than or equal to 2; and to obtain the target power consumption value of the GPU at the acquisition time by performing a weighted average of the n second power consumption values ​​after modification of the invalid second power consumption value.

[0136] In this solution, by modifying multiple invalid second power consumption values ​​and smoothing out the valid and modified second power consumption values, the power consumption of the GPU can be estimated more accurately.

[0137] In one possible implementation, a GPU substrate management controller is connected to a processor, which is at least used to run a GPU driver to obtain a power consumption value for the GPU. A second power consumption value for the GPU is obtained by the GPU substrate management controller from the processor. The second power consumption value for the GPU is valid after the processor runs the GPU driver.

[0138] Based on the same concept as the embodiments of the method in this application, this application also provides an electronic device. This electronic device can be a computing device 100. The structure of the electronic device can be found in [reference needed]. Figure 1 The computing device 100 is included. Additionally, the electronic device may include components that are more... Figure 1 The computing devices in the middle are 100 more devices.

[0139] In practical applications, a memory such as hard disk 123 or RAM 112 can store computer programs. When the electronic device is running, BMC 114 can read the computer programs stored in the memory such as hard disk 123 or RAM 112 to implement the steps in the GPU power consumption determination method described above, for example... Figure 3 Steps 301 to 303 in the process.

[0140] For example, a computer program can be divided into one or more modules / units, which can be a series of computer program instruction segments capable of performing a specific function. One or more modules / units are stored in a memory such as a hard disk 123 and executed by the BMC 114 to complete this application. For example, the computer program can be divided into an access module 701 and an estimation module 702; the specific functions of each module are described above.

[0141] In addition to the methods, apparatuses, computing devices, and electronic devices described above, embodiments of this application may also provide a computer program product comprising computer program instructions. When executed by a processor, these computer program instructions cause the processor to perform the steps of a GPU power consumption determination method according to various embodiments of this application described in the "Method" section of this specification. The computer program product can be written in any combination of one or more programming languages ​​to execute computer program code for performing the operations of embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The computer program code can be in source code form, object code form, executable file, or some intermediate form. The computer program code can be executed entirely on a user's computing device, partially on a user's device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0142] Furthermore, embodiments of this application may also provide a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of a GPU power consumption determination method according to various embodiments of this application as described in the "Method" section above. The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, a computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0143] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0144] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0145] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0146] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0147] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0148] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

[0149] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application.

Claims

1. A method for determining GPU power consumption, characterized in that, Applied to a computing device, the computing device including a GPU and a GPU substrate management controller, the GPU substrate management controller being used to monitor a first power consumption value of a hot-plug controller connected to the GPU and a second power consumption value of the GPU, the method comprising: Access the GPU substrate management controller to obtain the first power consumption value and the second power consumption value at the acquisition time point; When the second power consumption value at the acquisition time point is invalid, the target power consumption value of the GPU at the acquisition time point is determined based on the first power consumption value at the acquisition time point; the determination of the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point includes: If n consecutive second power consumption values ​​collected up to the acquisition time point are invalid, the target power consumption value of the GPU at the acquisition time point is obtained based on the first power consumption value at the acquisition time point; where n is a positive integer greater than or equal to 2; obtaining the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point includes: Determine a first parameter value and a second parameter value, wherein the first parameter value is the conversion coefficient of power transferred from the hot-swap controller to the GPU; and the second parameter value is the power consumption of other devices besides the GPU among the devices connected to the hot-swap controller. Based on the first parameter value, the second parameter value, and the first power consumption value at the acquisition time point, the target power consumption value of the GPU at the acquisition time point is estimated.

2. The method according to claim 1, characterized in that, The method further includes: When the second power consumption value at the acquisition time point is valid, the second power consumption value at the acquisition time point is used as the target power consumption value of the GPU.

3. The method according to claim 1, characterized in that, The target power consumption value of the GPU at the acquisition time point is the first power consumption value at the acquisition time point.

4. The method according to claim 1, characterized in that, The process of obtaining the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point includes: If some of the n consecutively collected second power consumption values ​​up to the acquisition time point are valid, for the invalid second power consumption values ​​among the n second power consumption values, the second power consumption value is modified based on the first power consumption value at the acquisition time point of the second power consumption value; where n is a positive integer greater than or equal to 2. The target power consumption value of the GPU at the acquisition time point is obtained by taking a weighted average of the n modified second power consumption values ​​after invalidating the second power consumption value.

5. The method according to any one of claims 1 to 4, characterized in that, The computing device includes a processor, which is at least configured to run a driver for a GPU to obtain a power consumption value for the GPU. A second power consumption value for the GPU is a power consumption value obtained by the GPU substrate management controller from the processor. The second power consumption value for the GPU becomes valid after the processor runs the driver for the GPU.

6. A computing device, characterized in that, The computing device includes: a substrate management controller (BMC), a GPU, a GPU substrate management controller, and a processor; the GPU substrate management controller is used to monitor a first power consumption value of the hot-plug controller connected to the GPU and obtain a second power consumption value of the GPU from the processor. The BMC is used to access the GPU substrate management controller to obtain the first power consumption value and the second power consumption value at the acquisition time point. The BMC is used to determine the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point when the second power consumption value at the acquisition time point is invalid; the determination of the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point includes: If n consecutive second power consumption values ​​collected up to the acquisition time point are invalid, the target power consumption value of the GPU at the acquisition time point is obtained based on the first power consumption value at the acquisition time point; where n is a positive integer greater than or equal to 2; obtaining the target power consumption value of the GPU at the acquisition time point based on the first power consumption value at the acquisition time point includes: Determine a first parameter value and a second parameter value, wherein the first parameter value is the conversion coefficient of power transferred from the hot-swap controller to the GPU; and the second parameter value is the power consumption of other devices besides the GPU among the devices connected to the hot-swap controller. Based on the first parameter value, the second parameter value, and the first power consumption value at the acquisition time point, the target power consumption value of the GPU at the acquisition time point is estimated.

7. The computing device according to claim 6, characterized in that, The processor is used to run the GPU driver to obtain the power consumption value of the GPU, and the second power consumption value of the GPU becomes effective after the processor runs the GPU driver.

8. The computing device according to claim 6 or 7, characterized in that, The BMC is used to take the second power consumption value at the acquisition time point as the target power consumption value of the GPU when the second power consumption value at the acquisition time point is valid.

Citation Information

Patent Citations

  • A heat dissipation method for intelligently regulating the rotational speed of a fan according to the power consumption of the equipment

    CN109298764A

  • Power consumption adjustment method, device and equipment of graphics processor, and storage medium

    CN116610477A