Control method and device of artificial intelligence processor, terminal and storage medium
By calculating the task response latency and power consumption of the AI processing module, a control scheme is generated, which solves the problem of balancing latency and energy consumption in the start-up and shutdown control of the AI processing module, and achieves energy-saving effect while meeting latency requirements.
Patent Information
- Application Number
- CN202511059256.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-30
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies struggle to optimize the activation and deactivation control of AI processing modules, leading to energy waste or increased task processing latency, making it difficult to achieve a balance between latency and energy consumption.
By calculating the task response latency and power consumption of each AI processing module, a switching control scheme is generated, and the control strategy with the lowest power consumption is selected to meet the latency requirements.
It achieves a significant reduction in terminal power consumption while meeting latency requirements, thus realizing effective energy saving.
Smart Images
Figure CN120909411A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of artificial intelligence, and particularly relates to a control method and device of an artificial intelligence processor, a terminal and a storage medium. BACKGROUND
[0002] Edge AI refers to deploying the training or inference process of an artificial intelligence algorithm directly on a terminal device for running, rather than relying on a cloud server to process data. It realizes the collection, analysis and decision-making of data locally by sinking the computing power to the device end, marking the evolution of AI technology from "cloud centralization" to "terminal distribution".
[0003] At present, in the AI deployment scheme on the terminal side, at least two AI processing modules are generally deployed on the terminal, the network side issues an AI task to the terminal, and the AI processing module of the terminal executes the AI task and feeds back the processing result. The AI task triggered by the network side is often not a continuous task, but a discrete task with intervals. If the AI processing module remains in an open state in an idle state, it will cause waste of energy. If the AI processing module in the idle state is closed, since the AI processing module will consume extra time when switching from the closed state to the open state, the time delay of subsequent processing tasks will increase. The prior art is difficult to optimize the start and stop control of the AI processing module to balance the time delay and energy consumption, resulting in waste of energy or increase of task processing time delay. SUMMARY
[0004] The present application aims to provide a scheduling method and device of an artificial intelligence processor, a terminal and a storage medium, to solve the technical problem that the prior art is difficult to optimize the start and stop control of the AI processing module to balance the time delay and energy consumption, resulting in waste of energy or increase of task processing time delay.
[0005] To achieve the above-mentioned purpose, the first aspect of the present application provides a control method of an artificial intelligence processor, applied to a terminal, wherein the terminal is internally deployed with a plurality of AI processing modules, and the control method comprises:
[0006] In response to notification information issued by the network side, the theoretical idle time delay of the AI processing module is obtained, and the notification information comprises the issue time of the next task or the earliest issue time;
[0007] The start time delay and model loading time delay of each AI processing module are obtained, and the task response time delay of each AI processing module is calculated, wherein the start time delay is the time delay of the AI processing module switching from the closed state to the open state, the model loading time delay is the time delay of the AI processing module loading the AI model, and the task response time delay is the minimum time delay of the AI processing module switching from the closed state to the state capable of executing the task.
[0008] determining whether there is a target AI processing module whose task response time delay is less than the theoretical idle time delay;
[0009] If yes, determining whether the number of the target AI processing modules is greater than 1;
[0010] If yes, obtaining power consumption information of each target AI processing module, and calculating task power consumption of each target AI processing module under different control strategies based on the power consumption information, the power consumption information including power consumption in an open standby state, a start state and an execution state, and the control strategies including keeping open state and executing tasks and keeping closed state and starting to execute tasks;
[0011] generating a control scheme based on the target AI processing module and the control strategy corresponding to the lowest task power consumption.
[0012] In one or more embodiments, the step of obtaining power consumption information of each target AI processing module and calculating task power consumption of each target AI processing module under different control strategies based on the power consumption information includes:
[0013] calculating the sum of power consumption in the open standby state and power consumption in the execution state of the target AI processing module to obtain task power consumption of the target AI processing module under the control strategy of keeping open state and executing tasks;
[0014] calculating the sum of power consumption in the start state and power consumption in the execution state of the target AI processing module to obtain task power consumption of the target AI processing module under the control strategy of keeping closed state and starting to execute tasks.
[0015] In one or more embodiments, if there is no target AI processing module, further comprising:
[0016] obtaining the power consumption information of each AI processing module, and calculating task power consumption of each AI processing module under the control strategy of keeping open state and executing tasks based on the power consumption information;
[0017] taking the AI processing module corresponding to the lowest task power consumption as an execution module of the next task;
[0018] controlling the execution module to be in an open state, and controlling the remaining AI processing modules to be closed.
[0019] In one or more embodiments, if the number of target AI processing modules is 1, further comprising:
[0020] calculating task power consumption of the target AI processing module under different control strategies.
[0021] generating a control scheme based on the control strategy corresponding to the lowest task power consumption.
[0022] In one or more embodiments, the notification information includes a next task issuance time;
[0023] When the control strategy corresponding to the lowest task power consumption is to keep the closed state and start to execute the task, the step of generating a control scheme based on the target AI processing module corresponding to the lowest task power consumption and the control strategy is specifically:
[0024] using the target AI processing module corresponding to the lowest task power consumption as an execution module of the next task, and obtaining a sleep time of the execution module based on the notification information and the task response time delay of the execution module, the sleep time being a time for the execution module to keep the closed state;
[0025] controlling all the AI processing modules to be closed, and controlling the execution module to be started when the sleep time is reached.
[0026] In one or more embodiments, the notification information includes an earliest next task issuance time;
[0027] When the control strategy corresponding to the lowest task power consumption is to keep the closed state and start to execute the task, the step of generating a control scheme based on the target AI processing module corresponding to the lowest task power consumption and the control strategy is specifically:
[0028] using the target AI processing module corresponding to the lowest task power consumption as an execution module of the next task, and reporting the task response time delay of the execution module to a network side;
[0029] controlling all the AI processing modules to be closed, and controlling the execution module to be started when the next task is received.
[0030] In one or more embodiments, when the control strategy corresponding to the lowest task power consumption is to keep the open state and execute the task, the step of generating a control scheme based on the target AI processing module corresponding to the lowest task power consumption and the control strategy is specifically:
[0031] using the target AI processing module corresponding to the lowest task power consumption as an execution module of the next task;
[0032] controlling the execution module to be in the open state, and controlling the remaining AI processing modules to be closed.
[0033] In one or more embodiments, the terminal is internally deployed with a first AI processing module and a second AI processing module, and the second AI processing module is deployed in a modem.
[0034] The step of obtaining the theoretical idle delay of the AI processing module further comprises:
[0035] obtaining the working state of the first AI processing module, the working state including the presence of other AI tasks or the absence of other AI tasks;
[0036] When the working state of the first AI processing module is the presence of other AI tasks, the first AI processing module is controlled to be in an open state, and the second AI processing module is controlled to be in a closed state.
[0037] To achieve the above-mentioned purpose, the second aspect of the present application provides a control device of an artificial intelligence processor, applied to a terminal, the terminal is internally deployed with a plurality of AI processing modules, and the control device comprises:
[0038] an obtaining module, configured to obtain the theoretical idle delay of the AI processing module in response to notification information issued by a network side, the notification information including the issuance time of the next task or the earliest issuance time;
[0039] a delay calculation module, configured to obtain the startup delay and the model loading delay of each AI processing module, and to calculate the task response delay of each AI processing module, wherein the startup delay is the delay of switching the AI processing module from a closed state to an open state, the model loading delay is the delay of loading an AI model by the AI processing module, and the task response delay is the minimum delay of switching the AI processing module from a closed state to a state capable of executing a task;
[0040] a first judgment module, configured to judge whether there is a target AI processing module with a task response delay less than the theoretical idle delay;
[0041] a second judgment module, configured to judge whether the number of target AI processing modules is greater than 1 when there is a target AI processing module;
[0042] a power consumption calculation module, configured to obtain the power consumption information of each target AI processing module, and to calculate the task power consumption of each target AI processing module under different control strategies based on the power consumption information, the power consumption information including the power consumption in an open standby state, a startup state, and an execution task state, and the control strategies including keeping an open state and executing a task, and keeping a closed state and starting to execute a task;
[0043] An output module is configured to generate a control scheme based on the target AI processing module and the control strategy corresponding to the lowest task power consumption.
[0044] To achieve the above object, the third aspect of the present application provides an electronic device, comprising:
[0045] at least one processor; and
[0046] a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the control method of the artificial intelligence processor according to any one of the above embodiments.
[0047] To achieve the above object, the fourth aspect of the present application provides a machine readable storage medium storing executable instructions, which, when executed, cause the machine to perform the control method of the artificial intelligence processor according to any one of the above embodiments.
[0048] Compared with the prior art, the present application has the following beneficial effects:
[0049] The control method of the present application can generate an on-off control scheme of the AI processing module based on the time delay requirement and the power consumption by calculating the task response time delay and the task power consumption of each AI processing module, balance the time delay and the power consumption, significantly reduce the terminal power consumption on the premise of meeting the time delay requirement, and achieve effective energy saving. BRIEF DESCRIPTION OF DRAWINGS
[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.
[0051] Figure 1 is a schematic diagram of an embodiment of a terminal side AI deployment scheme;
[0052] Figure 2 is a flowchart of an embodiment of the control method of the artificial intelligence processor of the present application;
[0053] Figure 3 is Figure 2 is a flowchart of an embodiment corresponding to S600a in
[0054] Figure 4 is Figure 2 is a flowchart of another embodiment corresponding to S600a in
[0055] Figure 5 isFigure 2 Figure 6 is a flowchart of another embodiment of the method for controlling the artificial intelligence processor of the present application;
[0056] Figure 6 Figure 7 is a flowchart of another embodiment of the method for controlling the artificial intelligence processor of the present application;
[0057] Figure 7 Figure 8 is a structural diagram of an embodiment of the control device for the artificial intelligence processor of the present application;
[0058] Figure 8 Figure 9 is a structural diagram of an embodiment of the electronic device of the present application. DETAILED DESCRIPTION
[0059] In order to enable those skilled in the art to better understand the technical solutions in the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present disclosure.
[0060] Please refer to Figure 1 , Figure 1 Figure 1 is a schematic diagram of an embodiment of the terminal AI deployment scheme. As shown in Figure 1 , in the existing terminal-side AI deployment scheme, there are generally at least two AI processing modules, wherein the first AI processing module 11 as a dedicated AI processing module can adopt NPU, GPU, etc., which is used for end-side application processing, such as picture beautification, voice assistant, etc., and can also be used for processing communication connection AI tasks; in addition, in order to cope with the application of AI in wireless connection, a second AI processing module 13 is generally also deployed in the modem 12 for processing communication connection AI tasks.
[0061] When the first AI processing module 11 and the second AI processing module 13 have no tasks, if they are in an open state, there will be additional basic power consumption, and on the other hand, the AI-based tasks triggered by the network side are not all continuous but discrete. Therefore, when there is no corresponding AI task, turning off the AI processing module is the most power-saving way. On the other hand, the execution of AI tasks by the network side often has a delay requirement, and always wants to execute as quickly as possible. At present, the conventional AI processor control method is difficult to balance the delay requirement and power consumption.
[0062] In order to solve the above problems, the applicant develops a new control method of artificial intelligence processor, which can perform on-off control based on the response time delay and power consumption of multiple AI processing modules deployed in the terminal, effectively reducing the power consumption while meeting the time delay requirement.
[0063] Specifically, please refer to Figure 2 , Figure 2 is a flowchart of an embodiment of the control method of the artificial intelligence processor of the applicant.
[0064] The control method is applied in the terminal, and the terminal is deployed with several AI processing modules, as shown in Figure 2 The control method comprises the following steps:
[0065] S100, in response to the notification information issued by the network side, the theoretical idle time delay of the AI processing module is obtained.
[0066] The notification information includes the next task issue time or the earliest issue time.
[0067] Specifically, in an embodiment, the network side can directly send the notification information including the next task issue time to the terminal, or send the communication information including the specific period of the periodic task to the terminal, so that the terminal can obtain the next task issue time.
[0068] In another embodiment, the network side can also send the communication information including the time within which the terminal will not be instructed to perform the AI task to the terminal, so that the terminal can obtain the earliest issue time of the next task, and there will be no AI task issued within the time.
[0069] It can be understood that after the terminal obtains the notification information, the theoretical idle time of the AI processing module can be obtained, that is, from the time when the information is obtained to the next task issue time or the earliest issue time.
[0070] S200, the start-up time delay and the model loading time delay of each AI processing module are obtained, and the task response time delay of each AI processing module is calculated.
[0071] The start-up time delay is the time delay of the AI processing module switching from the off state to the on state, the model loading time delay is the time delay of the AI processing module loading the AI model, and the task response time delay is the minimum time delay of the AI processing module switching from the off state to the state capable of executing the task.
[0072] After obtaining the theoretical idle time of the AI processing module, the total time delay of each AI processing module switching from the off state to the state capable of executing the task, that is, the task response time delay, needs to be calculated.
[0073] It can be understood that the AI processing module can enter the closed state under the premise of meeting the delay requirement only when the task response delay is less than the theoretical idle time.
[0074] Specifically, the task response delay of each AI processing module is the sum of its startup delay t1 and model loading delay t2, that is, t1+t2; wherein the model loading delay refers to the time required for the AI processing module to load the AI model of the AI task from the memory, which is affected by the transmission speed of the interface and the size of the AI model.
[0075] S300, judge whether there is a target AI processing module whose task response delay is less than the theoretical idle delay.
[0076] Further, compare the task response delay of each AI processing module with the theoretical idle time to determine whether there is a target AI processing module whose task response delay is less than the theoretical idle time.
[0077] If there is a target AI processing module, then:
[0078] S400a, judge whether the number of target AI processing modules is greater than 1.
[0079] If it is greater than 1, then:
[0080] S500a, obtain the power consumption information of each target AI processing module, and calculate the task power consumption of each target AI processing module under different control strategies based on the power consumption information.
[0081] Wherein, the power consumption information includes the power consumption in the open standby state, the startup state and the execution task state, and the control strategy includes keeping the open state and executing the task and keeping the closed state and starting to execute the task.
[0082] It can be understood that the power consumption of each AI processing module in the open standby state, the startup state and the execution task state is different, so based on the known power consumption information, the total power consumption of each AI processing module in the open state and the execution task state, and the total power consumption in the closed state and the execution task state can be calculated.
[0083] Specifically, the power consumption of the AI processing module in the open standby state is P1, the power consumption in the startup state is P2, and the power consumption in the execution task state is P3. The task power consumption of the target AI processing module under the control strategy of keeping the open state and executing the task is P1+P3; the task power consumption of the target AI processing module under the control strategy of keeping the closed state and starting to execute the task is P2+P3.
[0084] S600a, generate a control scheme based on the target AI processing module and the control strategy corresponding to the lowest task power consumption.
[0085] By comparing the task power consumptions of all target AI processing modules under different control strategies, the target AI processing module with the lowest task power consumption can be selected as the execution module for the next task, and a control scheme is generated based on the control strategy with the lowest power consumption.
[0086] Specifically, in one embodiment, when the control strategy corresponding to the lowest task power consumption is to keep the execution module in an open state and execute the task, please refer to Figure 3 , Figure 3 is Figure 2 the flowchart of another embodiment corresponding to S600a in FIG. 6.
[0087] As shown in Figure 3 , the method of generating a control scheme can include:
[0088] S601A, the target AI processing module corresponding to the lowest task power consumption is used as the execution module for the next task.
[0089] S602A, the execution module is controlled to be in an open state, and the remaining AI processing modules are controlled to be closed.
[0090] It can be understood that when the power consumption of the execution module keeping in an open state and executing the task is the lowest, all AI processing modules except the execution module can be controlled to be closed, and the execution module is controlled to keep in an open state until the task is issued, and the execution module is controlled to execute the task.
[0091] In another embodiment, when the control strategy corresponding to the lowest task power consumption is to keep the execution module in a closed state and start executing the task, the control scheme is different for different types of notification information.
[0092] Specifically, when the notification information includes the issue time of the next task, please refer to Figure 4 , Figure 4 is Figure 2 the flowchart of another embodiment corresponding to S600a in FIG. 6.
[0093] As shown in Figure 4 , the method of generating a control scheme includes:
[0094] S601B, the target AI processing module corresponding to the lowest task power consumption is used as the execution module for the next task, and based on the notification information and the task response time delay of the execution module, the sleep time of the execution module is obtained.
[0095] The sleep time is the time during which the execution module keeps in a closed state.
[0096] It can be understood that when the next task issuing time Ts is known and the task response time delay t1+t2 of the execution module is known, the sleep time of the execution module can be calculated, that is, Ts-(t1+t2).
[0097] S602B, control all AI processing modules to be closed, and control the execution module to be started when the sleep time is reached.
[0098] Since the execution module is in a closed state and has the lowest task power consumption when starting to execute a task, all AI processing modules can be controlled to be closed, and the execution module can be controlled to be started when the sleep time Ts-(t1+t2) is reached, thereby meeting the delay requirement while minimizing power consumption.
[0099] When the notification information includes the earliest next task issuing time, please refer to Figure 5 , Figure 5 is Figure 2 the flowchart of another embodiment corresponding to S600a in
[0100] As shown in Figure 5 , the method of generating a control scheme includes:
[0101] S601C, the target AI processing module corresponding to the lowest task power consumption is used as the execution module of the next task, and the task response time delay of the execution module is reported to the network side.
[0102] S602C, control all AI processing modules to be closed, and control the execution module to be started when the next task is received.
[0103] Since the specific task issuing time is unknown, the terminal cannot start the execution module before the task is issued after the execution module is closed, and therefore, when the execution module is in a closed state and starts to execute a task, an additional delay, that is, the above-mentioned task response time delay t1+t2, will be caused.
[0104] The task response time delay needs to be reported to the network side, and the processing time of the network side in expecting the terminal to feed back the task execution result needs to consider the task response time delay. For example, the processing time of the terminal in processing a task is Tx, and then the time of the network side in triggering the terminal to report the task processing result cannot be less than Tx+T0, wherein T0 is greater than or equal to t1+t2.
[0105] It should be noted that the terminal can directly report the task response time delay of the execution module to the network side, or report a preset time delay T0 to the network side. T0 should be greater than or equal to the task response time delay t1+t2 of the execution module.
[0106] Based on the technical solutions of the embodiments S500a-S600a, a plurality of target AI processing modules are first obtained by comparing the task response time delay and the theoretical idle time, and then the target AI processing module with the lowest task power consumption and the corresponding control strategy are obtained by power consumption screening, and then the control scheme is generated to ensure the optimal power consumption under the premise of meeting the delay requirement.
[0107] Further, please continue to refer to Figure 2 If there is no target AI processing module with a task response time delay less than the theoretical idle time, S400a is synchronized with the following:
[0108] S400b, obtain the power consumption information of each AI processing module, and calculate the task power consumption of each AI processing module under the control strategy of keeping the open state and executing the task based on the power consumption information.
[0109] The specific task power consumption calculation method is consistent with the task power consumption calculation method of the target AI processing module. Please refer to S500a, which will not be repeated here.
[0110] S500b, the AI processing module corresponding to the lowest task power consumption is used as the execution module of the next task.
[0111] S600b, control the execution module to be in the open state, and control the remaining AI processing modules to be closed.
[0112] It can be understood that if the task response time delay of all AI processing modules is greater than or equal to the theoretical idle time, the delay requirement cannot be met at this time, and the selection can be directly based on the power consumption to select the AI processing module with the lowest task power consumption as the execution module. In order to minimize the delay, the execution module is controlled to keep the open state until the task is executed, and the remaining AI processing modules are controlled to be in the closed state.
[0113] Further, please continue to refer to Figure 2 If the number of target AI processing modules is 1, S500a is synchronized with the following:
[0114] S500c, calculate the task power consumption of the target AI processing module under different control strategies.
[0115] S600c, generate a control scheme based on the control strategy corresponding to the lowest task power consumption.
[0116] It can be understood that when only one AI processing module has a task response time delay less than the theoretical idle time, the power consumption of the target AI processing module under different control strategies can be calculated, and the control strategy with the lowest task power consumption is selected as the control strategy of the target AI processing module. A control scheme is generated.
[0117] The generation method of the specific control scheme is adjusted based on different control strategies and different notification information, and can be seen from S600a and Figures 3 to 5 , which will not be repeated here.
[0118] In the above embodiments, the control scheme of the terminal AI processing module is generated based on the notification information issued by the network side, and the next task is an AI task based on communication connection. In other embodiments, the terminal can also process end tasks such as picture beautification and voice assistants. For AI processing modules with end tasks, the AI processing module can be kept in an open state and synchronously process the next task issued by the network side.
[0119] Specifically, taking the terminal AI deployment scheme shown in Figure 1 as an example, the terminal is internally deployed with a dedicated first AI processing module 11 and a second AI processing module 13 arranged in the modem 12. Please refer to Figure 6 , Figure 6 is a flowchart of another embodiment of the control method of the artificial intelligence processor of the present application.
[0120] As shown in Figure 6 , the control method comprises:
[0121] S10, in response to the notification information issued by the network side, obtaining the working state of the first AI processing module.
[0122] The working state includes the presence of other AI tasks or the absence of other AI tasks.
[0123] S20, when the working state of the first AI processing module is the presence of other AI tasks, controlling the first AI processing module to be in an open state, and controlling the second AI processing module to be in a closed state.
[0124] It can be understood that when the first AI processing module 11 has other AI tasks, the first AI processing module 11 can be kept in an open state and used as an execution module for the next task issued by the network side, while the second AI processing module 13 is controlled to be in a closed state.
[0125] When the first AI processing module 11 has no other AI tasks, the task response delay and power consumption can be calculated by referring to the technical scheme shown in Figure 1 , and the control scheme is finally determined.
[0126] Based on the control method of the artificial intelligence processor according to the above embodiments, the task response delay and task power consumption of each AI processing module are calculated, the control scheme is generated based on the delay requirement and power consumption, the balance between delay and power consumption is achieved, the terminal power consumption is significantly reduced on the premise of meeting the delay requirement, and effective energy saving is achieved.
[0127] The application also provides a control device of an artificial intelligence processor, please refer to Figure 7 , Figure 7 is a structural schematic diagram of an embodiment of the control device of the artificial intelligence processor of the present application.
[0128] As Figure 7 shown, the control device is applied to a terminal, and a plurality of AI processing modules are deployed in the terminal. The control device comprises an acquisition module 21, a time delay calculation module 22, a first judgment module 23, a second judgment module 24 and a power consumption calculation module 25.
[0129] The acquisition module 21 is configured to acquire a theoretical idle time delay of the AI processing module in response to notification information issued by a network side, and the notification information comprises an issuance time or an earliest issuance time of a next task.
[0130] The time delay calculation module 22 is configured to acquire a starting time delay and a model loading time delay of each AI processing module, and to calculate a task response time delay of each AI processing module, wherein the starting time delay is a time delay of switching the AI processing module from a closed state to an open state, the model loading time delay is a time delay of loading an AI model by the AI processing module, and the task response time delay is a minimum time delay of switching the AI processing module from the closed state to a state capable of executing a task.
[0131] The first judgment module 23 is configured to judge whether there is a target AI processing module with a task response time delay less than the theoretical idle time delay.
[0132] The second judgment module 24 is configured to judge whether the number of target AI processing modules is greater than 1 when there is a target AI processing module.
[0133] The power consumption calculation module 25 is configured to acquire power consumption information of each target AI processing module, and to calculate a task power consumption of each target AI processing module under different control strategies based on the power consumption information, wherein the power consumption information comprises power consumption in an open standby state, a starting state and an execution task state, and the control strategies comprise keeping open state and executing a task and keeping closed state and starting to execute a task.
[0134] The output module 26 is configured to generate a control scheme based on a target AI processing module and a control strategy corresponding to the lowest task power consumption.
[0135] As described above with reference to Figures 1 to 6 , the control method of the artificial intelligence processor according to the embodiments of the present application is described. The details mentioned in the above description of the method embodiments are also applicable to the control device of the artificial intelligence processor of the embodiments of the present application. The control device of the artificial intelligence processor above can be realized by hardware, or by software or a combination of hardware and software.
[0136] The present application also provides an electronic device, please refer to Figure 8 , Figure 8 is a structural schematic diagram of an embodiment of the electronic device of the present application. As shown in Figure 8 , the electronic device 30 can include at least one processor 31, a memory 32 (for example, a non-volatile memory), a memory 33, and a communication interface 34, and the at least one processor 31, the memory 32, the memory 33, and the communication interface 34 are connected together via an internal bus 35. The at least one processor 31 executes at least one computer-readable instruction stored or encoded in the memory 32.
[0137] It should be understood that the computer executable instructions stored in the memory 32, when executed, cause the at least one processor 31 to perform various operations and functions described above in conjunction with Figures 1-6 various embodiments of the present specification.
[0138] In embodiments of the present specification, the electronic device 30 can include, but is not limited to, a personal computer, a server computer, a workstation, a desktop computer, a laptop computer, a notebook computer, a mobile electronic device, a smart phone, a tablet computer, a cellular phone, a personal digital assistant (PDA), a handheld device, a messaging device, a wearable electronic device, a consumer electronic device, and the like.
[0139] According to one embodiment, a program product such as a machine-readable medium is provided. The machine-readable medium can have instructions (i.e., the above-mentioned elements implemented in software) that, when executed by a machine, cause the machine to perform various operations and functions described above in conjunction with Figures 1-5 various embodiments of the present specification. Specifically, a system or apparatus equipped with a readable storage medium on which a software program code implementing the functions of any of the above-mentioned embodiments is stored, and a computer or processor of the system or apparatus can be provided to read and execute the instructions stored in the readable storage medium.
[0140] In this case, the program code read from the readable medium itself can implement the functions of any of the above-mentioned embodiments, so the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of the present specification.
[0141] Embodiments of the readable storage medium include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROM. Alternatively, the program code can be downloaded from a server computer or the cloud over a communication network.
[0142] Those skilled in the art should understand that the various embodiments disclosed above can be variously changed and modified without departing from the essence of the application. Therefore, the scope of protection of the present specification should be defined by the appended claims rather than by the detailed description.
[0143] It should be noted that not all steps and units in the above processes and system structure diagrams are necessary, and some steps or units can be omitted according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in each of the above embodiments can be a physical structure or a logical structure, that is, some units can be implemented by the same physical client, or some units can be implemented by multiple physical clients, or can be implemented by some components in multiple independent devices together.
[0144] In each of the above embodiments, a hardware unit or module can be implemented mechanically or with electronic components. For example, a hardware unit, module or processor can include dedicated circuitry or logic, such as an application specific integrated circuit (ASIC) or field programmable gate array (FPGA), for implementing the relevant operations. The hardware unit or processor can also include programmable logic or circuitry, such as microprocessor or other programmable processor, which can be temporarily configured by software to perform the relevant operations. The specific implementation (mechanical or dedicated permanent circuitry, or temporarily configuration of circuitry) can be determined based on cost and time considerations.
[0145] The specific embodiments described above with reference to the drawings represent just example implementations. The description is not intended to be exhaustive or to limit the embodiments to the precise form disclosed. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the term "exemplary" is meant to mean "serving as an example, instance, or illustration," and is not meant to mean "preferred" or "advantageous over other embodiments." The specific implementations described above can be implemented in software or hardware, or a combination thereof. The specific implementations described above can be implemented in one or more computer programs executing on one or more programmable computer systems (e.g., hardware) to perform a variety of tasks. Each of the computer programs can include one or more software modules that can be stored and executed on the computer system. The specific implementations described above can be implemented in a distributed computing environment, where tasks are performed by one or more computer systems (e.g., hardware) that are located at various geographic locations.
[0146] The above description of the disclosure has been presented to enable any person skilled in the art to implement or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the scope of protection of the disclosure. Therefore, the disclosure is not limited to the examples and designs described herein, but is consistent with the broadest scope of the principles and novel features disclosed herein.
Claims
1. A control method for an artificial intelligence processor, characterized in that, The application is applied to a terminal, wherein a plurality of AI processing modules are deployed in the terminal, and the control method comprises the following steps: In response to notification information issued by a network side, a theoretical idle time delay of the AI processing module is obtained, wherein the notification information comprises a next task issuing time or an earliest issuing time; A starting time delay and a model loading time delay of each AI processing module are obtained, and a task response time delay of each AI processing module is calculated, wherein the starting time delay is a time delay of switching the AI processing module from a closed state to an open state, the model loading time delay is a time delay of loading an AI model by the AI processing module, and the task response time delay is a minimum time delay of switching the AI processing module from the closed state to a state capable of executing a task; It is judged whether there is a target AI processing module with a task response time delay less than the theoretical idle time delay; If yes, it is judged whether the number of the target AI processing modules is greater than 1; If yes, power consumption information of each target AI processing module is obtained, and a task power consumption of each target AI processing module under different control strategies is calculated based on the power consumption information, wherein the power consumption information comprises power consumption in an open standby state, a starting state and a task execution state, and the control strategies comprise keeping an open state and executing a task and keeping a closed state and starting to execute a task; A control scheme is generated based on the target AI processing module corresponding to the lowest task power consumption and the control strategy.
2. The control method according to claim 1, characterized by, The step of obtaining the power consumption information of each target AI processing module and calculating the task power consumption of each target AI processing module under different control strategies based on the power consumption information comprises the following steps: The sum of the power consumption in the open standby state and the task execution power consumption of the target AI processing module is calculated to obtain the task power consumption of the target AI processing module under the control strategy of keeping the open state and executing the task; The sum of the power consumption in the starting state and the task execution power consumption of the target AI processing module is calculated to obtain the task power consumption of the target AI processing module under the control strategy of keeping the closed state and starting to execute the task.
3. The control method according to claim 1, characterized by, If there is no target AI processing module, the following steps are further included: The power consumption information of each AI processing module is obtained, and the task power consumption of each AI processing module under the control strategy of keeping the open state and executing the task is calculated based on the power consumption information; The AI processing module corresponding to the lowest task power consumption is taken as an execution module of the next task; The execution module is controlled to be in the open state, and the remaining AI processing modules are controlled to be closed.
4. The control method according to claim 1, characterized by, If the number of the target AI processing modules is 1, the following steps are further included: The task power consumption of the target AI processing module under different control strategies is calculated; A control scheme is generated based on the control strategy corresponding to the lowest task power consumption.
5. The control method according to claim 1, characterized by, The notification information comprises a next task issuing time; When the control strategy corresponding to the lowest task power consumption is keeping the closed state and starting to execute the task, the step of generating a control scheme based on the target AI processing module corresponding to the lowest task power consumption and the control strategy is specifically as follows: corresponding to the minimum task power consumption as the execution module of the next task, and based on the notification information and the task response time delay of the execution module, a sleep time of the execution module is obtained, the sleep time being a time during which the execution module remains in a closed state; controlling all the AI processing modules to be closed, and controlling the execution module to be started when the sleep time is reached.
6. The control method according to claim 1, characterized by, The notification information includes the earliest issue time of the next task; When the control strategy corresponding to the minimum task power consumption is to remain in a closed state and start executing a task, the step of generating a control scheme based on the target AI processing module corresponding to the minimum task power consumption and the control strategy is specifically: corresponding to the minimum task power consumption as the execution module of the next task, and reporting the task response time delay of the execution module to the network side; controlling all the AI processing modules to be closed, and controlling the execution module to be started when the sleep time is reached.
7. The control method according to claim 1, characterized by, When the control strategy corresponding to the minimum task power consumption is to remain in a closed state and start executing a task, the step of generating a control scheme based on the target AI processing module corresponding to the minimum task power consumption and the control strategy is specifically: corresponding to the minimum task power consumption as the execution module of the next task; controlling the execution module to be in an open state, and controlling the remaining AI processing modules to be closed.
8. The control method according to claim 1, characterized by, The terminal is internally deployed with a first AI processing module and a second AI processing module, and the second AI processing module is deployed in a modem; The step of obtaining the theoretical idle time delay of the AI processing module further includes: obtaining a working state of the first AI processing module, the working state including the presence of other AI tasks or the absence of other AI tasks; When the working state of the first AI processing module is the presence of other AI tasks, the first AI processing module is controlled to be in an open state, and the second AI processing module is controlled to be in a closed state.
9. A control device of an artificial intelligence processor, characterized by, The control device is applied to a terminal, and the terminal is internally deployed with a plurality of AI processing modules, and the control device includes: an obtaining module, configured to obtain a theoretical idle time delay of the AI processing module in response to notification information issued by a network side, the notification information including an issue time or an earliest issue time of a next task; a time delay calculation module, configured to obtain a start time delay and a model loading time delay of each AI processing module, and to calculate a task response time delay of each AI processing module, wherein the start time delay is a time delay of switching the AI processing module from a closed state to an open state, the model loading time delay is a time delay of loading an AI model by the AI processing module, and the task response time delay is a minimum time delay of switching the AI processing module from the closed state to a state capable of executing a task; a first judgment module, configured to judge whether there is a target AI processing module with a task response time delay less than the theoretical idle time delay. a second determining module, configured to determine whether the number of the target AI processing modules is greater than 1 when the target AI processing modules exist; a power consumption calculating module, configured to acquire power consumption information of each target AI processing module, and calculate task power consumption of each target AI processing module under different control strategies based on the power consumption information, the power consumption information including power consumption in an open standby state, a start state and a task execution state, and the control strategies including keeping open state and executing tasks and keeping closed state and starting to execute tasks; an output module, configured to generate a control scheme based on the target AI processing module corresponding to the lowest task power consumption and the control strategy. 10.An electronic device, comprising: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the control method of the artificial intelligence processor according to any one of claims 1 to 8.