WO2022088800A1 - 一种服务器的电源控制方法、系统及装置 - Google Patents
一种服务器的电源控制方法、系统及装置 Download PDFInfo
- Publication number
- WO2022088800A1 WO2022088800A1 PCT/CN2021/109190 CN2021109190W WO2022088800A1 WO 2022088800 A1 WO2022088800 A1 WO 2022088800A1 CN 2021109190 W CN2021109190 W CN 2021109190W WO 2022088800 A1 WO2022088800 A1 WO 2022088800A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- gpu
- target
- power
- utilization rate
- level
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000011217 control strategy Methods 0.000 claims description 47
- 238000004590 computer program Methods 0.000 claims description 7
- 230000005764 inhibitory process Effects 0.000 claims description 5
- 230000001960 triggered effect Effects 0.000 claims description 3
- 230000001629 suppression Effects 0.000 abstract 1
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003990 capacitor Substances 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/30—Means for acting in the event of power-supply failure or interruption, e.g. power-supply fluctuations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
- G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Definitions
- the present invention relates to the field of servers, and in particular, to a power control method, system and device for a server.
- AI Artificial Intelligence, artificial intelligence
- GPU Graphics Processing Unit, graphics processor
- the GPU is the key to the performance of the AI server.
- EDPP Electronic design point peak current, peak current
- EDPP is usually 2 to 3 times the usual current. It is even more difficult to control. If the current control is not good, it will cause the main power supply of the system to directly shut down or restart.
- the control method of GPU EDPP is: adding a large capacitor in a PSU (Power supply unit, power supply unit) used to supply power to the GPU or on the main power board of the system to prevent the peak current of the GPU in a short time, And as shown in Figure 1, a management unit is added between the PSU (system power supply) and each GPU. If the utilization rate of the total system power supply is too high, the management unit will trigger the power brake signal of each GPU at the same time, and each GPU will be efficient The energy computing is changed to low-performance computing to reduce GPU power consumption and prevent the system from shutting down or restarting the total power supply, but this will also greatly reduce the computing performance of the entire AI server.
- PSU Power supply unit, power supply unit
- the purpose of the present invention is to provide a power control method, system and device for a server, which adopts the method of grading the utilization rate of the total power of the system, and suppresses the computing power of the GPU of the system when the utilization rate of the total power of the system is higher. , that is, the more the power consumption of the system GPU is reduced, so that the computing performance of the server can be guaranteed as much as possible while preventing the total power supply of the system from shutting down or restarting.
- the present invention provides a power control method for a server, including:
- the usage rate of the total system power is divided into levels in advance, and the GPU power control policies are set under the usage rates of the total system power at different levels; among them, the higher the level of the system total power usage, the higher the level.
- the GPU power control strategy set under the usage rate of the total system power is stronger to suppress the GPU computing power of the system;
- the usage rate of the total system power is divided into grades in advance, and the process of setting the GPU power control strategy under the usage rates of the total system power at different levels includes:
- the utilization rate of the total power of the system is divided into three levels in advance, and the low-level utilization rate, the medium-level utilization rate and the high-level utilization rate are obtained;
- a second GPU power control strategy for selecting a target trigger GPU from each GPU and triggering the power brake signal of the target trigger GPU according to a preset GPU trigger selection strategy is set for the middle-level utilization rate
- a third GPU power control strategy for selecting a target to shut down the GPU from the GPUs according to a preset GPU shutdown selection strategy and powering off the target to shut down the GPU is set for the high-level usage rate.
- the process of adjusting the load balancing distribution of each GPU in the system includes:
- Part of the computing workload of the GPU in the high workload state is evenly distributed to each GPU in the low workload state, so that the GPU originally in the high workload state is reduced to the low workload state.
- the process of selecting a target from each GPU to trigger the GPU and triggering the power brake signal of the target to trigger the GPU according to a preset GPU trigger selection strategy includes:
- the power brake signals of the GPUs in the high workload state are sequentially triggered until the actual usage level of the total power of the system drops to a low level.
- the process of selecting a target from each GPU to turn off the GPU and turning off the power of the target and turning off the GPU according to a preset GPU shutdown selection strategy includes:
- the power supply of the GPU in the low workload state is firstly turned off, and then the power supply of the GPU in the high workload state is turned off until the actual usage level of the total power supply of the system is reduced to a medium level.
- the power control method for the server further includes:
- the present invention also provides a power control system for a server, including:
- the preset module is used to classify the utilization rate of the total system power in advance, and set the GPU power control strategy under the utilization rate of the total system power of different levels; wherein, the utilization rate of the system total power of the higher level The higher the level, the stronger the inhibition of the GPU computing power of the system by the GPU power control policy set under the higher level of system total power usage;
- a determining module configured to obtain the actual utilization rate of the total power supply of the system, and determine a target utilization rate level corresponding to the actual utilization rate according to the classification result of the utilization rate of the total power supply of the system;
- the control module is configured to perform power control on the GPU in the system according to the GPU power control strategy corresponding to the target usage level.
- the preset module is specifically used for:
- the utilization rate of the total power of the system is divided into three levels in advance, and the low-level utilization rate, the medium-level utilization rate and the high-level utilization rate are obtained;
- a second GPU power control strategy for selecting a target trigger GPU from each GPU and triggering the power brake signal of the target trigger GPU according to a preset GPU trigger selection strategy is set for the middle-level utilization rate
- a third GPU power control strategy for selecting a target to shut down the GPU from the GPUs according to a preset GPU shutdown selection strategy and powering off the target to shut down the GPU is set for the high-level usage rate.
- the process of adjusting the load balancing distribution of each GPU in the system includes:
- Part of the computing workload of the GPU in the high workload state is evenly distributed to each GPU in the low workload state, so that the GPU originally in the high workload state is reduced to the low workload state.
- the present invention also provides a power control device for a server, including:
- a processor disposed between the total system power supply and each GPU in the system is used to implement the steps of any one of the above-mentioned server power control methods when executing the computer program.
- the present invention provides a power control method for a server.
- the usage rate of the total system power is divided into grades in advance, and GPU power control strategies are set one by one under the usage rates of the total system power at different levels;
- the higher the utilization rate of the total system power the stronger the degree of inhibition of the GPU computing power of the system by the GPU power control policy set at the higher level of the utilization rate of the total system power;
- the power usage level division result determines a target usage rate level corresponding to the actual usage rate; power control is performed on the GPU in the system according to the GPU power control strategy corresponding to the target usage rate level.
- the present application adopts the method of grading the usage rate of the total system power supply, and the computing power of the system GPU is suppressed more when the usage rate of the total system power supply is higher, that is, the power consumption of the system GPU is reduced more, thereby preventing the When the system power is shut down or restarted, the computing performance of the server should be guaranteed as much as possible.
- the present invention also provides a power control system and device for a server, which have the same beneficial effects as the above power control method.
- FIG. 1 is a schematic diagram of a power supply control of a server in the prior art
- FIG. 2 is a flowchart of a method for controlling a power supply of a server according to an embodiment of the present invention
- FIG. 3 is a schematic structural diagram of a power supply system for a GPU in a server according to an embodiment of the present invention.
- the core of the present invention is to provide a power control method, system and device for a server.
- the utilization rate of the total power of the system is graded, and the computing power of the GPU of the system is more restrained when the utilization rate of the total power of the system is higher. , that is, the more the power consumption of the system GPU is reduced, so that the computing performance of the server can be guaranteed as much as possible while preventing the total power supply of the system from shutting down or restarting.
- FIG. 2 is a flowchart of a method for controlling a power supply of a server according to an embodiment of the present invention.
- the power control method of the server includes:
- Step S1 Classify the usage rate of the total system power in advance, and set GPU power control policies one by one under the usage rates of the total system power at different levels.
- the present application classifies the utilization rate of the total system power in advance. It should be noted that the higher the utilization rate of the system total power, the higher the utilization rate of the system total power, that is, the lower the utilization rate of the system total power, such as , when the usage rate of the total system power is below 80%, there is no risk of shutting down or restarting the total system power. There is no need to classify the usage rate of the total system power below 80%.
- the usage rate of the power supply is divided into grades, such as setting 80%-90% as the first grade, 90%-100% as the second grade, and 100% and above as the third grade, the grade order is: first grade ⁇ second level ⁇ third level.
- the present application also sets GPU power control strategies one by one under the utilization rate of the total system power at different levels in advance, that is, under the utilization rate of the total system power at each level, a GPU power control strategy is set, which is understandable Yes, the higher the utilization rate of the total system power, the greater the risk of shutting down or restarting the system. Therefore, the GPU power control policy set at a higher level of the total system power utilization should suppress the GPU computing power of the system.
- the computing power of the system GPU is weaker, and the power consumption of the system GPU is reduced more to prevent the shutdown or restart of the total system power; the lower the usage rate of the total system power At this time, the stronger the computing power of the system GPU, the less power consumption of the system GPU is reduced, so as to ensure the computing performance of the server, thereby ensuring the computing performance of the server as much as possible while preventing the system from shutting down or restarting the total power supply.
- Step S2 Obtain the actual utilization rate of the total system power supply, and determine a target utilization rate level corresponding to the actual utilization rate according to the classification result of the utilization rate of the total system power supply.
- the present application obtains the actual utilization rate of the total power supply of the system, and then determines the utilization rate corresponding to the actual utilization rate of the total power supply of the system (called the target utilization rate) according to the classification result of the utilization rate of the total power supply of the system, so as to be Subsequently, the GPU power control strategy required under the actual utilization rate of the total system power is determined.
- Step S3 According to the GPU power control strategy corresponding to the target usage level, power control is performed on the GPU in the system.
- the present application determines the GPU power supply control strategy corresponding to the target usage rate level according to the corresponding relationship between the usage rate level and the GPU power supply control strategy, and then According to the GPU power control policy corresponding to the target usage level, power control is performed on the GPU in the system.
- the present invention provides a power control method for a server.
- the usage rate of the total system power is divided into grades in advance, and GPU power control strategies are set one by one under the usage rates of the total system power at different levels;
- the higher the utilization rate of the total system power the stronger the degree of inhibition of the GPU computing power of the system by the GPU power control policy set at the higher level of the utilization rate of the total system power;
- the power usage level division result determines a target usage rate level corresponding to the actual usage rate; power control is performed on the GPU in the system according to the GPU power control strategy corresponding to the target usage rate level. It can be seen that the present application adopts the method of grading the usage rate of the total system power supply.
- the computing power of the system GPU is more suppressed, that is, the power consumption of the system GPU is reduced more.
- the computing performance of the server should be guaranteed as much as possible.
- the usage rate of the total system power is divided into grades in advance, and the process of setting the GPU power control strategy under the usage rates of the total system power at different levels includes:
- the utilization rate of the total power of the system is divided into three levels in advance, and the low-level utilization rate, the medium-level utilization rate and the high-level utilization rate are obtained;
- a third GPU power control strategy for selecting a target to shut down the GPU from each GPU according to a preset GPU shutdown selection strategy and powering off the target to shut down the GPU is set for high-level usage.
- the preset of this application is set in advance, and only needs to be set once, and does not need to be reset unless it needs to be modified according to the actual situation.
- the present application divides the usage rate of the total power supply of the system into three levels: low, medium, and high, to obtain a low level usage rate, a medium level usage rate, and a high level usage rate.
- the first GPU power control strategy set for the low level utilization rate in this application is: adjusting The load balancing distribution of each GPU in the system, that is, part of the computing workload of the GPU in the high workload state is evenly distributed to each GPU in the low workload state, so that the GPU originally in the high workload state is reduced to the low workload state.
- the second GPU power control strategy set for the medium-level usage rate is: select a GPU (called a target-triggered GPU) that pre-changes high-efficiency computing to low-efficiency computing from each GPU according to the preset GPU trigger selection strategy, and then trigger the The target triggers the power brake signal of the GPU, thereby reducing the power consumption of the system GPU to prevent the system from being shut down or rebooted by the total power.
- the third GPU power control strategy set for the high-level usage rate is: according to the preset GPU shutdown selection strategy, a pre-power-off GPU (called a target shutdown GPU) is selected from each GPU, and then the target power-off GPU is turned off, thereby Reduced the power consumption of the system GPU to prevent a total system power shutdown or restart.
- a pre-power-off GPU called a target shutdown GPU
- the process of adjusting the load balancing distribution of each GPU in the system includes:
- Part of the computing workload of the GPU in the high workload state is evenly distributed to each GPU in the low workload state, so that the GPU originally in the high workload state is reduced to the low workload state.
- the present application configures integrated chips for acquiring GPU power supply parameters for each GPU in advance, so as to use the integrated chips to acquire current parameters of each GPU.
- the present application sets a current threshold such that: the current of the GPU in the high workload state > the current threshold > the low workload If the current parameter of the target GPU is greater than the preset current threshold, it means that the target GPU is in a high workload state; if not greater than the preset current threshold, the target GPU is in a high workload state;
- the current threshold indicates that the target GPU is in a low workload state, so that the workload state of the GPU can be judged according to the current size of the GPU.
- the specific process of adjusting the load balancing distribution of each GPU in the system is as follows: allocate part of the computing workload of the GPU in the high workload state to each GPU in the low workload state, so that the original high workload state The GPU is reduced to a low workload state, thereby reducing the power consumption of the system GPU to prevent the system's total power from being shut down or restarted without affecting the computing performance of the server.
- the process of selecting a target to trigger the GPU from each GPU and triggering the power brake signal of the target to trigger the GPU according to a preset GPU trigger selection strategy includes:
- the specific process of selecting a target from each GPU to trigger the GPU and triggering the power brake signal of the target to trigger the GPU is as follows: sequentially select the GPU in a high workload state from each GPU as the target to trigger the GPU, and trigger the current target to trigger the GPU. power brake signal until the actual usage level of the total system power is reduced to a low level.
- this GPU triggering method can prevent the system from shutting down or restarting the total power supply, while ensuring the maximum allowable computing performance of the server as much as possible.
- the process of selecting a target from each GPU to turn off the GPU and turning off the power of the target and turning off the GPU according to a preset GPU shutdown selection strategy includes:
- Priority is given to powering off GPUs in a low workload state, and then powering off GPUs in a high workload state, until the actual usage level of the total system power is reduced to a medium level.
- the specific process of selecting a target from each GPU to turn off the GPU and turning off the target to turn off the power of the GPU is as follows: from each GPU, the GPU in the low workload state is preferentially selected as the target to turn off the GPU in turn, and the GPU in the low workload state is turned off. After the selection is complete, select the GPU in the high workload state as the target to turn off the GPU in turn, and turn off the current target to turn off the power of the GPU until the actual usage level of the total system power is reduced to the medium level.
- this method of turning off the power of the GPU in a low workload state preferentially and turning off the power in sequence can prevent the total power of the system from shutting down or restarting, and at the same time ensure the maximum allowable computing performance of the server as much as possible.
- the power control method for the server further includes:
- the present application can also compare each power supply parameter of the target GPU obtained by using the integrated chip with its corresponding preset parameter safety threshold, for example, compare the current parameter of the target GPU with its corresponding preset current parameter safety threshold; the target The voltage parameter of the GPU is compared with its corresponding preset voltage parameter safety threshold; if any power supply parameter of the target GPU is greater than its corresponding preset parameter safety threshold, the integrated chip is used to directly turn off the power supply of the target GPU, thereby preventing the GPU chip from being damaged. burn.
- the power supply system of the GPU in the server includes a PSU and a power supply control device.
- the power supply control device includes a processor and an integrated chip (Hotswap IC) used to obtain GPU power supply parameters.
- the processor is used for executing memory storage.
- the computer program of the invention implements the steps of any one of the above-mentioned power control methods for the server.
- the present application also provides a power control system for a server, including:
- the preset module is used to classify the utilization rate of the total system power in advance, and set the GPU power control strategy under the utilization rate of the total system power of different levels; wherein, the utilization rate of the system total power of the higher level The higher the level, the stronger the inhibition of the GPU computing power of the system by the GPU power control policy set under the higher level of system total power usage;
- the determining module is used to obtain the actual utilization rate of the total power of the system, and determine the target utilization rate corresponding to the actual utilization rate according to the classification result of the utilization rate of the total power supply of the system;
- the control module is used to control the power supply of the GPU in the system according to the GPU power control strategy corresponding to the target usage level.
- the preset module is specifically used for:
- the utilization rate of the total power of the system is divided into three levels in advance, and the low-level utilization rate, the medium-level utilization rate and the high-level utilization rate are obtained;
- a third GPU power control strategy for selecting a target to shut down the GPU from each GPU according to a preset GPU shutdown selection strategy and powering off the target to shut down the GPU is set for high-level usage.
- the process of adjusting the load balancing distribution of each GPU in the system includes:
- Part of the computing workload of the GPU in the high workload state is evenly distributed to each GPU in the low workload state, so that the GPU originally in the high workload state is reduced to the low workload state.
- the present application also provides a power control device for a server, including:
- a processor disposed between the total system power supply and each GPU in the system is used to implement the steps of any one of the above-mentioned server power control methods when executing the computer program.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Power Sources (AREA)
Abstract
一种服务器的电源控制方法、系统及装置,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;获取系统总电源的实际使用率,并根据系统总电源的使用率等级划分结果确定与实际使用率对应的目标使用率等级;按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。可见,本申请在系统总电源的使用率越高的情况下,系统GPU的功耗降低越多,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
Description
本申请要求于2020年10月29日提交至中国专利局、申请号为202011181718.7、发明名称为“一种服务器的电源控制方法、系统及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本发明涉及服务器领域,特别是涉及一种服务器的电源控制方法、系统及装置。
AI(Artificial Intelligence,人工智能)服务器需要具备大量平行运算的能力,而GPU(Graphics Processing Unit,图形处理器)较适用于平行运算,被广泛应用于AI服务器。GPU是AI服务器的效能高低的关键,AI服务器拥有的GPU数量越多,AI服务器的效能越高,但GPU数量越多,所有GPU所需的电流就越多,电流控制也就越来越不容易。而GPU在短时间内要执行高效能计算时,可允许短时间电流上升,此短时间电流称为EDPP(Electrical design point peak current,峰值电流),EDPP通常是平时电流的2到3倍,电流更不好控制,若电流控制不佳,会造成系统总电源直接关机或重开机。
现有技术中,GPU EDPP的控制方法为:在用于为GPU供电的PSU(Power supply unit,电源供应器)内或系统总电源板上增设大电容来阻止GPU在短时间内的峰值电流,并如图1所示,在PSU(系统总电源)和各GPU之间增设管理单元,若系统总电源的使用率过高,则由管理单元同时触发各GPU的power brake讯号,各GPU将高效能计算改为低效能计算,以降低GPU功耗,阻止系统总电源关机或重开机,但这也会让整个AI服务器的计算性能有很大程度地下降。
因此,如何提供一种解决上述技术问题的方案是本领域的技术人员目前需要解决的问题。
发明内容
本发明的目的是提供一种服务器的电源控制方法、系统及装置,采用系统总电源的使用率分等级的方式,在系统总电源的使用率越高的情况下对系统GPU的计算能力越抑制,即系统GPU的功耗降低越多,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
为解决上述技术问题,本发明提供了一种服务器的电源控制方法,包括:
预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;
获取所述系统总电源的实际使用率,并根据所述系统总电源的使用率等级划分结果确定与所述实际使用率对应的目标使用率等级;
按照所述目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
优选地,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略的过程,包括:
预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;
为所述低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;
为所述中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的第二GPU电源控制策略;
为所述高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的第三GPU电源控制策略。
优选地,调节系统中各GPU的负载均衡分配的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;
将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
优选地,按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;其中,所述目标GPU为任一GPU;
依次触发处于高工作负载状态的GPU的power brake讯号,直至所述系统总电源的实际使用率的等级降至低等级。
优选地,按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;
优先关闭处于低工作负载状态的GPU的电源,再关闭处于高工作负载状态的GPU的电源,直至所述系统总电源的实际使用率的等级降至中等级。
优选地,所述服务器的电源控制方法还包括:
将利用所述集成芯片获取的目标GPU的各电源参数分别与其对应的预设参数安全阈值作比较,当所述目标GPU的任一电源参数大于其对应的预设参数安全阈值时,借助所述集成芯片直接关闭所述目标GPU的电源。
为解决上述技术问题,本发明还提供了一种服务器的电源控制系统,包括:
预设模块,用于预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;
确定模块,用于获取所述系统总电源的实际使用率,并根据所述系统总电源的使用率等级划分结果确定与所述实际使用率对应的目标使用率等级;
控制模块,用于按照所述目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
优选地,所述预设模块具体用于:
预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;
为所述低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;
为所述中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的第二GPU电源控制策略;
为所述高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的第三GPU电源控制策略。
优选地,调节系统中各GPU的负载均衡分配的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;
将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
为解决上述技术问题,本发明还提供了一种服务器的电源控制装置,包括:
存储器,用于存储计算机程序;
设于系统总电源和系统中各GPU之间的处理器,用于在执行所述计算机程序时实现上述任一种服务器的电源控制方法的步骤。
本发明提供了一种服务器的电源控制方法,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;获取系统总电源的实际使用率,并根据系统总电源的使用率等级划分结果确定与实际使用率对应的目标使用率等级;按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。可见,本申请采用系统总电源的使用率分等级的方式,在系统总电源的使用率越高的情况下对系统GPU的计算能力越抑制,即系统GPU的功耗降低越多,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
本发明还提供了一种服务器的电源控制系统及装置,与上述电源控制方法具有相同的有益效果。
为了更清楚地说明本发明实施例中的技术方案,下面将对现有技术和实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出 创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为现有技术中的一种服务器的电源控制原理图;
图2为本发明实施例提供的一种服务器的电源控制方法的流程图;
图3为本发明实施例提供的一种服务器内GPU的电源系统的结构示意图。
本发明的核心是提供一种服务器的电源控制方法、系统及装置,采用系统总电源的使用率分等级的方式,在系统总电源的使用率越高的情况下对系统GPU的计算能力越抑制,即系统GPU的功耗降低越多,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
为使本发明实施例的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
请参照图2,图2为本发明实施例提供的一种服务器的电源控制方法的流程图。
该服务器的电源控制方法包括:
步骤S1:预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略。
具体地,本申请提前将系统总电源的使用率进行等级划分,需要说明的是,越高等级的系统总电源的使用率越高,即越低等级的系统总电源的使用率越低,比如,系统总电源的使用率在80%以下时,系统总电源没有关机或重开机风险,没必要对80%以下的系统总电源的使用率进行等级划分,可将80%及其以上的系统总电源的使用率进行等级划分,如设置80%-90%为第一等级,90%-100%为第二等级,100%及其以上为第三等级,等级排序为:第一等级<第二等级<第三等级。
而且,本申请还提前在不同等级的系统总电源的使用率下一一设置GPU电源控制策略,即在每一等级的系统总电源的使用率下,均设置一个GPU电源控制策略,可以理解的是,系统总电源的使用率越高,系统总电源关机或重开机的风险越大,所以越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度应越强,即在系统总电源的使用率越高时,系统GPU计算能力越弱,系统GPU的功耗降低越多,以阻止系统总电源关机或重开机;在系统总电源的使用率越低时,系统GPU计算能力越强,系统GPU的功耗降低越少,以保证服务器的计算性能,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
步骤S2:获取系统总电源的实际使用率,并根据系统总电源的使用率等级划分结果确定与实际使用率对应的目标使用率等级。
具体地,本申请获取系统总电源的实际使用率,然后根据系统总电源的使用率等级划分结果,确定与系统总电源的实际使用率对应的使用率等级(称为目标使用率等级),以为后续确定在系统总电源的实际使用率下所需的GPU电源控制策略。
步骤S3:按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
具体地,本申请在确定与系统总电源的实际使用率对应的目标使用率等级之后,根据使用率等级与GPU电源控制策略的对应关系,确定与目标使用率等级对应的GPU电源控制策略,然后按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
本发明提供了一种服务器的电源控制方法,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;获取系统总电源的实际使用率,并根据系统总电源的使用率等级划分结果确定与实际使用率对应的目标使用率等级;按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。可见, 本申请采用系统总电源的使用率分等级的方式,在系统总电源的使用率越高的情况下对系统GPU的计算能力越抑制,即系统GPU的功耗降低越多,从而在阻止系统总电源关机或重开机的同时尽可能保证服务器的计算性能。
在上述实施例的基础上:
作为一种可选的实施例,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略的过程,包括:
预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;
为低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;
为中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发目标触发GPU的power brake讯号的第二GPU电源控制策略;
为高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭目标关闭GPU的电源的第三GPU电源控制策略。
需要说明的是,本申请的预设是提前设置好的,只需要设置一次,除非根据实际情况需要修改,否则不需要重新设置。
具体地,本申请提前将系统总电源的使用率进行低、中、高三等级划分,得到低等级使用率、中等级使用率及高等级使用率。考虑到越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度应越强,所以本申请为低等级使用率设置的第一GPU电源控制策略为:调节系统中各GPU的负载均衡分配,即将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态,处于高工作负载状态的GPU的电流通常是处于低工作负载状态的GPU的电流的2到3倍,从而降低了系统GPU的功耗,以阻止系统总电源关机或重开机,且并未影 响到服务器的计算性能。为中等级使用率设置的第二GPU电源控制策略为:按照预设GPU触发选择策略从各GPU中选择出预将高效能计算改为低效能计算的GPU(称为目标触发GPU),然后触发目标触发GPU的power brake讯号,从而降低了系统GPU的功耗,以阻止系统总电源关机或重开机。为高等级使用率设置的第三GPU电源控制策略为:按照预设GPU关闭选择策略从各GPU中选择出预关闭电源的GPU(称为目标关闭GPU),然后关闭目标关闭GPU的电源,从而降低了系统GPU的功耗,以阻止系统总电源关机或重开机。
比如,设置80%-90%为低等级使用率(警告),90%-100%为中等级使用率(严重),100%及其以上为高等级使用率(致命),其各自对应的GPU电源控制策略如下表1所示:
表1
电源使用率 | GPU电源控制策略 |
80%-90% | 负载重新分配 |
90%-100% | 启动power brake讯号 |
100%及其以上 | 关闭部分GPU的电源 |
作为一种可选的实施例,调节系统中各GPU的负载均衡分配的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定目标GPU处于高工作负载状态;若否,则确定目标GPU处于低工作负载状态;其中,目标GPU为任一GPU;
将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
具体地,本申请提前为各GPU一一配置用于获取GPU电源参数的集成芯片,以利用集成芯片获取各GPU的电流参数。考虑到处于高工作负载状态的GPU的电流>处于低工作负载状态的GPU的电流,所以本申请设置一个电流阈值,使其:处于高工作负载状态的GPU的电流>电流阈值>处于低工作负载状态的GPU的电流,则在获取目标GPU的电流参数后,判断目标GPU的电流参数是否大于预设电流阈值;若大于预设电流阈值,说明目标GPU处于高工作负载状态;若不大于预设电流阈值,说明目标GPU处于低工作负载状态,从而实现根据GPU的电流大小判断GPU的工作负载状态。
基于此,调节系统中各GPU的负载均衡分配的具体过程为:将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态,从而降低了系统GPU的功耗,以阻止系统总电源关机或重开机,且并未影响到服务器的计算性能。
作为一种可选的实施例,按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发目标触发GPU的power brake讯号的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定目标GPU处于高工作负载状态;其中,目标GPU为任一GPU;
依次触发处于高工作负载状态的GPU的power brake讯号,直至系统总电源的实际使用率的等级降至低等级。
具体地,GPU的工作负载状态的判定原理在上述实施例已经提及,本申请在此不再赘述。
基于此,从各GPU中选择目标触发GPU并触发目标触发GPU的power brake讯号的具体过程为:从各GPU中依次选择处于高工作负载状态的GPU作为目标触发GPU,并触发当前的目标触发GPU的power brake讯号,直至系统总电源的实际使用率的等级降至低等级。
需要说明的是,这种GPU依次触发方式能够在阻止系统总电源关机或重开机的同时,尽可能保证服务器的最大允许计算性能。
作为一种可选的实施例,按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭目标关闭GPU的电源的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定目标GPU处于高工作负载状态;若否,则确定目标GPU处于低工作负载状态;其中,目标GPU为任一GPU;
优先关闭处于低工作负载状态的GPU的电源,再关闭处于高工作负载状态的GPU的电源,直至系统总电源的实际使用率的等级降至中等级。
具体地,GPU的工作负载状态的判定原理在上述实施例已经提及,本申请在此不再赘述。
基于此,从各GPU中选择目标关闭GPU并关闭目标关闭GPU的电源的具体过程为:从各GPU中优先选择处于低工作负载状态的GPU依次作为目标关闭GPU,在处于低工作负载状态的GPU选择完毕后,再选择处于高工作负载状态的GPU依次作为目标关闭GPU,并关闭当前的目标关闭GPU的电源,直至系统总电源的实际使用率的等级降至中等级。
需要说明的是,这种处于低工作负载状态的GPU的电源优先关闭且电源依次关闭的方式能够在阻止系统总电源关机或重开机的同时,尽可能保证服务器的最大允许计算性能。
作为一种可选的实施例,服务器的电源控制方法还包括:
将利用集成芯片获取的目标GPU的各电源参数分别与其对应的预设参数安全阈值作比较,当目标GPU的任一电源参数大于其对应的预设参数安全阈值时,借助集成芯片直接关闭目标GPU的电源。
进一步地,本申请还可将利用集成芯片获取的目标GPU的各电源参数分别与其对应的预设参数安全阈值作比较,如目标GPU的电流参数与其对应的预设电流参数安全阈值作比较;目标GPU的电压参数与其对应的预设电压参数安全阈值作比较;若目标GPU的任一电源参数大于其对应的预设 参数安全阈值,则借助集成芯片直接关闭目标GPU的电源,从而避免GPU芯片被烧毁。
综上,如图3所示,服务器内GPU的电源系统包括PSU和电源控制装置,电源控制装置包括处理器和用于获取GPU电源参数的集成芯片(Hotswap IC),处理器用于在执行存储器存储的计算机程序时实现上述任一种服务器的电源控制方法的步骤。
本申请还提供了一种服务器的电源控制系统,包括:
预设模块,用于预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;
确定模块,用于获取系统总电源的实际使用率,并根据系统总电源的使用率等级划分结果确定与实际使用率对应的目标使用率等级;
控制模块,用于按照目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
作为一种可选的实施例,预设模块具体用于:
预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;
为低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;
为中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发目标触发GPU的power brake讯号的第二GPU电源控制策略;
为高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭目标关闭GPU的电源的第三GPU电源控制策略。
作为一种可选的实施例,调节系统中各GPU的负载均衡分配的过程,包括:
预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用集成芯片获取各GPU的电流参数;
判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定目标GPU处于高工作负载状态;若否,则确定目标GPU处于低工作负载状态;其中,目标GPU为任一GPU;
将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
本申请提供的电源控制系统的介绍请参考上述电源控制方法的实施例,本申请在此不再赘述。
本申请还提供了一种服务器的电源控制装置,包括:
存储器,用于存储计算机程序;
设于系统总电源和系统中各GPU之间的处理器,用于在执行计算机程序时实现上述任一种服务器的电源控制方法的步骤。
本申请提供的电源控制装置的介绍请参考上述电源控制方法的实施例,本申请在此不再赘述。
还需要说明的是,在本说明书中,诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来,而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
对所公开的实施例的上述说明,使本领域专业技术人员能够实现或使用本发明。对这些实施例的多种修改对本领域的专业技术人员来说将是显 而易见的,本文中所定义的一般原理可以在不脱离本发明的精神或范围的情况下,在其他实施例中实现。因此,本发明将不会被限制于本文所示的这些实施例,而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。
Claims (10)
- 一种服务器的电源控制方法,其特征在于,包括:预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;获取所述系统总电源的实际使用率,并根据所述系统总电源的使用率等级划分结果确定与所述实际使用率对应的目标使用率等级;按照所述目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
- 如权利要求1所述的服务器的电源控制方法,其特征在于,预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略的过程,包括:预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;为所述低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;为所述中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的第二GPU电源控制策略;为所述高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的第三GPU电源控制策略。
- 如权利要求2所述的服务器的电源控制方法,其特征在于,调节系统中各GPU的负载均衡分配的过程,包括:预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
- 如权利要求2所述的服务器的电源控制方法,其特征在于,按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的过程,包括:预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;其中,所述目标GPU为任一GPU;依次触发处于高工作负载状态的GPU的power brake讯号,直至所述系统总电源的实际使用率的等级降至低等级。
- 如权利要求2所述的服务器的电源控制方法,其特征在于,按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的过程,包括:预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;优先关闭处于低工作负载状态的GPU的电源,再关闭处于高工作负载状态的GPU的电源,直至所述系统总电源的实际使用率的等级降至中等级。
- 如权利要求3-5任一项所述的服务器的电源控制方法,其特征在于,所述服务器的电源控制方法还包括:将利用所述集成芯片获取的目标GPU的各电源参数分别与其对应的预设参数安全阈值作比较,当所述目标GPU的任一电源参数大于其对应的预设参数安全阈值时,借助所述集成芯片直接关闭所述目标GPU的电源。
- 一种服务器的电源控制系统,其特征在于,包括:预设模块,用于预先将系统总电源的使用率进行等级划分,并在不同等级的系统总电源的使用率下一一设置GPU电源控制策略;其中,越高等级的系统总电源的使用率越高,越高等级的系统总电源的使用率下设置的GPU电源控制策略对系统GPU计算能力的抑制程度越强;确定模块,用于获取所述系统总电源的实际使用率,并根据所述系统总电源的使用率等级划分结果确定与所述实际使用率对应的目标使用率等级;控制模块,用于按照所述目标使用率等级对应的GPU电源控制策略,对系统中的GPU进行电源控制。
- 如权利要求7所述的服务器的电源控制系统,其特征在于,所述预设模块具体用于:预先将系统总电源的使用率进行三等级划分,得到低等级使用率、中等级使用率及高等级使用率;为所述低等级使用率设置用于调节系统中各GPU的负载均衡分配的第一GPU电源控制策略;为所述中等级使用率设置用于按照预设GPU触发选择策略从各GPU中选择目标触发GPU并触发所述目标触发GPU的power brake讯号的第二GPU电源控制策略;为所述高等级使用率设置用于按照预设GPU关闭选择策略从各GPU中选择目标关闭GPU并关闭所述目标关闭GPU的电源的第三GPU电源控制策略。
- 如权利要求8所述的服务器的电源控制系统,其特征在于,调节系统中各GPU的负载均衡分配的过程,包括:预先为各GPU一一配置用于获取GPU电源参数的集成芯片,并利用所述集成芯片获取各GPU的电流参数;判断目标GPU的电流参数是否大于预设电流阈值;若是,则确定所述目标GPU处于高工作负载状态;若否,则确定所述目标GPU处于低工作负载状态;其中,所述目标GPU为任一GPU;将处于高工作负载状态的GPU的部分运算工作量均衡分配给处于低工作负载状态的各GPU,以使原处于高工作负载状态的GPU降至处于低工作负载状态。
- 一种服务器的电源控制装置,其特征在于,包括:存储器,用于存储计算机程序;设于系统总电源和系统中各GPU之间的处理器,用于在执行所述计算机程序时实现如权利要求1-6任一项所述的服务器的电源控制方法的步骤。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011181718.7A CN112114647B (zh) | 2020-10-29 | 2020-10-29 | 一种服务器的电源控制方法、系统及装置 |
CN202011181718.7 | 2020-10-29 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022088800A1 true WO2022088800A1 (zh) | 2022-05-05 |
Family
ID=73794658
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/109190 WO2022088800A1 (zh) | 2020-10-29 | 2021-07-29 | 一种服务器的电源控制方法、系统及装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN112114647B (zh) |
WO (1) | WO2022088800A1 (zh) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112114647B (zh) * | 2020-10-29 | 2022-06-10 | 苏州浪潮智能科技有限公司 | 一种服务器的电源控制方法、系统及装置 |
CN113721747B (zh) * | 2021-07-29 | 2023-08-29 | 苏州浪潮智能科技有限公司 | 一种服务器及其防烧板电路和方法 |
CN116667268B (zh) * | 2022-12-15 | 2024-07-12 | 荣耀终端有限公司 | 防止触发过流保护的方法及电子设备 |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260003A (zh) * | 2015-11-30 | 2016-01-20 | 浪潮(北京)电子信息产业有限公司 | 一种服务器整机自动保护方法及系统 |
US20170262953A1 (en) * | 2016-03-14 | 2017-09-14 | Dell Products, Lp | System and Method for Normalization of GPU Workloads Based on Real-Time GPU Data |
CN208903299U (zh) * | 2018-11-21 | 2019-05-24 | 厦门科一物联网科技有限公司 | 一种ai智能专用计算卡及其构成的边缘网络 |
CN111009883A (zh) * | 2019-11-29 | 2020-04-14 | 苏州浪潮智能科技有限公司 | 一种防止pcie设备过电流误触发的方法 |
CN111290560A (zh) * | 2020-01-19 | 2020-06-16 | 苏州浪潮智能科技有限公司 | 一种防止服务器过流掉电的方法及系统 |
CN111475293A (zh) * | 2020-03-27 | 2020-07-31 | 苏州浪潮智能科技有限公司 | 一种服务器及其供电保护系统 |
CN112114647A (zh) * | 2020-10-29 | 2020-12-22 | 苏州浪潮智能科技有限公司 | 一种服务器的电源控制方法、系统及装置 |
CN112670948A (zh) * | 2020-11-20 | 2021-04-16 | 山东云海国创云计算装备产业创新中心有限公司 | 一种板卡保护方法、系统及装置 |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103188277B (zh) * | 2011-12-27 | 2016-05-18 | 中国电信股份有限公司 | 负载能耗管理系统、方法和服务器 |
CN105068915B (zh) * | 2015-08-10 | 2019-03-15 | 合肥联宝信息技术有限公司 | 电源管理装置及方法 |
CN109446026A (zh) * | 2018-10-22 | 2019-03-08 | 郑州云海信息技术有限公司 | 整机柜gpu服务器供电方法、服务器、电源装置及存储介质 |
CN111352815A (zh) * | 2020-02-26 | 2020-06-30 | 苏州浪潮智能科技有限公司 | 一种服务器系统的性能均衡检测方法、系统及装置 |
-
2020
- 2020-10-29 CN CN202011181718.7A patent/CN112114647B/zh active Active
-
2021
- 2021-07-29 WO PCT/CN2021/109190 patent/WO2022088800A1/zh active Application Filing
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260003A (zh) * | 2015-11-30 | 2016-01-20 | 浪潮(北京)电子信息产业有限公司 | 一种服务器整机自动保护方法及系统 |
US20170262953A1 (en) * | 2016-03-14 | 2017-09-14 | Dell Products, Lp | System and Method for Normalization of GPU Workloads Based on Real-Time GPU Data |
CN208903299U (zh) * | 2018-11-21 | 2019-05-24 | 厦门科一物联网科技有限公司 | 一种ai智能专用计算卡及其构成的边缘网络 |
CN111009883A (zh) * | 2019-11-29 | 2020-04-14 | 苏州浪潮智能科技有限公司 | 一种防止pcie设备过电流误触发的方法 |
CN111290560A (zh) * | 2020-01-19 | 2020-06-16 | 苏州浪潮智能科技有限公司 | 一种防止服务器过流掉电的方法及系统 |
CN111475293A (zh) * | 2020-03-27 | 2020-07-31 | 苏州浪潮智能科技有限公司 | 一种服务器及其供电保护系统 |
CN112114647A (zh) * | 2020-10-29 | 2020-12-22 | 苏州浪潮智能科技有限公司 | 一种服务器的电源控制方法、系统及装置 |
CN112670948A (zh) * | 2020-11-20 | 2021-04-16 | 山东云海国创云计算装备产业创新中心有限公司 | 一种板卡保护方法、系统及装置 |
Also Published As
Publication number | Publication date |
---|---|
CN112114647B (zh) | 2022-06-10 |
CN112114647A (zh) | 2020-12-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022088800A1 (zh) | 一种服务器的电源控制方法、系统及装置 | |
US7596705B2 (en) | Automatically controlling processor mode of multi-core processor | |
US8719605B2 (en) | Method for detecting a trigger to a program not actively being reviewed by the user and performing a power saving action without placing the device as a whole into a sleep state | |
US20220229482A1 (en) | Systems and methods for power outage protection of storage device | |
US9377841B2 (en) | Adaptively limiting a maximum operating frequency in a multicore processor | |
US8442697B2 (en) | Method and apparatus for on-demand power management | |
US10115442B2 (en) | Demand-based provisioning of volatile memory for use as non-volatile memory | |
US20150082076A1 (en) | Dynamic clock regulation | |
US10963028B2 (en) | System, method and apparatus for energy efficiency and energy conservation by configuring power management parameters during run time | |
US8862918B2 (en) | Efficient frequency boost operation | |
US10936038B2 (en) | Power control for use of volatile memory as non-volatile memory | |
US20230315189A1 (en) | Controlling a Processor Clock | |
US20190042418A1 (en) | Power button override for persistent memory enabled platforms | |
CN108369488B (zh) | 使用易失性存储器作为非易失性存储器 | |
WO2022052626A1 (zh) | 功耗管理的方法和相关设备 | |
US10534420B2 (en) | Electronic devices, electronic systems, and control methods therefor | |
CN106292987B (zh) | 一种处理器掉电时序控制系统及方法 | |
TW201428471A (zh) | 具有電源控制功能之電子裝置 | |
WO2014084842A1 (en) | Enforcing a power consumption duty cycle in a processor | |
CA3241222A1 (en) | Method and apparatus for unit control and photovoltaic multi-split air conditioning system | |
CN114327883A (zh) | 一种频率调控方法、装置、电子设备及介质 | |
TW201535100A (zh) | 電子裝置與電源管理方法 | |
US12124319B2 (en) | Dynamic peak power control | |
US20240264651A1 (en) | Energy efficient vmin architecture for shared rails | |
US20160062449A1 (en) | Computing platform power consumption level adjustment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21884535 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21884535 Country of ref document: EP Kind code of ref document: A1 |