Computing device and electronic equipment

Abstract

The embodiment of the invention provides a computing device and electronic equipment, and relates to the technical field of computers. The computing device comprises a main board and a plurality of computing power boards, the plurality of computing power boards are connected to the main board, and the main board is used for distributing computing tasks to at least part of the plurality of computing power boards according to computing tasks to be processed and running states of the plurality of computing power boards. According to the implementation mode, resources of the multiple computing power boards in the computing device can be fully and evenly applied, high power consumption caused by overload operation of the computing power boards is reduced, heat generated by the computing power boards is reduced, and the overall performance of the computing power boards and the computing device is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a computing device and an electronic device. Background Art

[0002] With the rapid development of artificial intelligence, big data analysis and scientific computing, the demand for high-computing computing devices is becoming increasingly urgent. High-computing devices can provide powerful data processing capabilities and efficient parallel computing performance, and have gradually covered various fields in scientific research, business and society. With the increase in computing power, when the processor of the computing device performs tasks, a large amount of calculation and data processing will be performed. This process requires frequent electrical signal conversion and transmission, which causes the hardware to generate a lot of heat. If the heat cannot be dissipated in time, it may cause a series of problems such as reduced hardware performance, shortened life, and reduced stability of the computing device.

[0003] In view of this, how to reduce the heat of the computing device and improve the overall performance of the computing device is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The embodiments of the present application provide a computing device and an electronic device. The computing device generates relatively low heat during operation, which is beneficial to improving the overall performance of the computing device.

[0005] In a first aspect, a computing device is provided, comprising: a mainboard and a plurality of computing boards, wherein the plurality of computing boards are connected to the mainboard, and the mainboard is used to allocate computing tasks to at least some of the plurality of computing boards according to computing tasks to be processed and operating states of the plurality of computing boards.

[0006] In an embodiment of the present application, the mainboard can obtain the operating status of multiple computing boards, and according to the operating status of the multiple computing boards, assign tasks to at least some of the computing boards. As an example, when a computing board among multiple computing boards is in a high-load operating state, the mainboard can stop assigning tasks to the computing board, or, when another computing board among multiple computing boards is in a low-load operating state, the mainboard can assign new tasks to the computing board. Under this implementation mode, it is conducive to the full and balanced application of the resources of multiple computing boards in the computing device, reducing the higher power consumption of the computing board due to overload operation, thereby reducing the heat generation of the computing board, which is conducive to improving the overall performance of the computing board and the computing device.

[0007] In some possible implementations, the operating status of multiple computing boards includes resource utilization, and the main board is used to monitor the resource utilization of the multiple computing boards. When the resource utilization of one computing board among the multiple computing boards is greater than or equal to a first preset threshold, the main board is used to allocate at least part of the data in the computing task to another computing board among the multiple computing boards.

[0008] In the technical solution of this embodiment, the main board monitors the resource utilization rates of multiple computing boards, thereby monitoring the operating status and load status of the multiple computing boards. According to the operating status and load status, data allocation of computing tasks can be achieved, and the resources of multiple computing boards in the computing device can be effectively integrated and applied, reducing the possibility of a large load on a certain computing board, balancing the operation of multiple computing boards, and thus effectively reducing the overall power consumption and heat of the computing device.

[0009] In some possible embodiments, the operating status of multiple computing boards includes temperature, and the computing device further includes a sensor. The sensor is used to detect the temperature of the multiple computing boards and send it to the main board; when the temperature of one of the multiple computing boards is greater than or equal to a second preset threshold, the main board is used to allocate at least part of the data in the computing task to another one of the multiple computing boards.

[0010] Through the technical solution of this embodiment, the main board can monitor the temperatures of multiple computing boards through the sensor, thereby determining whether the computing board is overloaded and whether the heat generation problem is serious, and realizing data allocation of computing tasks according to the temperature of the computing board, which can directly and effectively reduce the overall power consumption and heat of the computing device.

[0011] In some possible embodiments, at least some of the multiple computing boards have different computing powers. The main board is used to obtain the computing powers of the multiple computing boards, and according to the data volume of the computing task to be processed and the computing powers of the multiple computing boards, task allocation is performed to at least some of the multiple computing boards.

[0012] Under this embodiment, the main board can adaptively allocate appropriate data of computing tasks to them according to the computing powers of multiple computing boards, can achieve the implementation of computing tasks with higher efficiency, and ensure resource balance among multiple computing boards, thereby improving the overall performance of the computing device.

[0013] In some possible embodiments, at least some of the multiple computing boards have different types. The main board is used to obtain the types of the multiple computing boards, and according to the type of the computing task to be processed and the types of the multiple computing boards, task allocation is performed to at least some of the multiple computing boards.

[0014] Through the technical solution of this embodiment, the main board can adaptively allocate data of computing tasks suitable for their processing to at least some of the multiple computing boards according to the type of the computing task and the types of the multiple computing boards, so that the multiple computing boards can better exert their respective performances, efficiently execute computing tasks, and the resources of the computing boards are also relatively balanced. While reducing the power consumption and heat of the computing device, it can improve the processing efficiency and processing effect of computing tasks, thereby improving the comprehensive performance of the computing device.

[0015] In some possible embodiments, the computing tasks include: real-time tasks, complex tasks, and hybrid tasks, and the multiple computing boards include a first computing board and a second computing board; if the computing task includes a real-time task, the main board is used to allocate the real-time task to the first computing board, if the computing task includes a complex task, the main board is used to allocate the complex task to the second computing board, and if the computing task includes a hybrid task, the main board is used to allocate the real-time task in the hybrid task to the first computing board and allocate the complex task in the hybrid task to the second computing board.

[0016] In this embodiment, multiple first computing boards in the first computing board group can be suitable for processing real-time data with low latency and high concurrency, that is, the first computing board group can be preferably used to process real-time tasks. Multiple second computing boards in the second computing board group can be suitable for large-scale data processing and complex calculations, that is, the second computing board group can be preferably used to process complex tasks.

[0017] In some possible embodiments, the multiple computing boards include: a first computing board group and a second computing board group, the first computing board group includes multiple first computing boards, and the second computing board group includes multiple second computing boards; at least some of the first computing boards among the multiple first computing boards have different computing powers, and / or at least some of the second computing boards among the multiple second computing boards have different computing powers.

[0018] Based on the technical solution of this embodiment, the computing device can not only have different types of computing boards, but also the same type can include multiple computing boards with different computing powers, so that multiple different types of computing board groups can be preferably suitable for different computing tasks, enabling the computing device to flexibly adapt to the execution of different computing tasks and effectively improving the overall performance of the computing device.

[0019] In some possible embodiments, the main board is used to sequentially allocate real-time tasks to multiple first computing boards in the first computing board group. When the resource utilization rate of the previous first computing board is greater than or equal to the first preset threshold, the main board is used to allocate at least part of the data in the real-time task to the next first computing board; and / or the main board is used to sequentially allocate complex tasks to multiple second computing boards in the second computing board group. When the resource utilization rate of the previous second computing board is greater than or equal to the first preset threshold, the main board is used to allocate at least part of the data in the complex task to the next second computing board.

[0020] Based on the technical solution of this embodiment, the main board can preferably allocate different tasks to two different types of computing board groups, thereby preferably invoking the resources of the computing device and improving the overall performance of the computing device.

[0021] In some possible embodiments, the main board is configured to obtain the computing power of multiple first computing power boards in the first computing power board group, and allocate real-time tasks to at least some of the multiple first computing power boards according to the computing power of the multiple first computing power boards; and / or, the main board is configured to obtain the computing power of multiple second computing power boards in the second computing power board group, and allocate complex tasks to at least some of the multiple second computing power boards according to the computing power of the multiple second computing power boards.

[0022] In this embodiment, the main board can perform more accurate task allocation to multiple computing power boards in the computing power board group according to the computing power of the computing power boards, so as to utilize appropriate resources to process tasks, thereby greatly reducing the temperature and power consumption of the computing power boards and improving the overall performance of the computing device.

[0023] In some possible embodiments, during the execution of the computing task by the first computing power board and / or the second computing power board, the main board is further configured to adjust the power consumption of the first computing power board and / or the second computing power board according to the execution status of the computing task.

[0024] In some possible embodiments, the first computing power board includes at least one of the following hardware modules: a real-time processing unit, a cache, or a parallel computing unit; the second computing power board includes at least one of the following hardware modules: a batch processing computing unit, a high-capacity memory, or a high-speed network interface.

[0025] In some possible embodiments, the computing task further includes a regular task, and the main board is configured to execute the regular task.

[0026] In some possible embodiments, the main board includes an AI computing power main board.

[0027] Based on the technical solution of this embodiment, the AI computing power main board has powerful computing capabilities and dedicated hardware acceleration modules to meet the requirements of AI applications for large-scale data processing and complex algorithms.

[0028] In a second aspect, there is provided an electronic device, including: the computing device provided in the first aspect or any one of the possible embodiments in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 FIG. shows a schematic diagram of a computing power board provided by an embodiment of the present application.

[0030] Figure 2 FIG. shows a schematic diagram of a computing device provided by an embodiment of the present application.

[0031] Figure 3 FIG. shows another schematic diagram of a computing device provided by an embodiment of the present application.

[0032] Figure 4Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0033] Figure 5 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0034] Figure 6 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0035] Figure 7 A data processing flow chart of a computing task provided by the embodiment of the present application is shown.

[0036] Figure 8 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0037] Figure 9 A schematic diagram of the main board provided by the embodiment of the present application is shown.

[0038] Figure 10 Another schematic diagram of the computing device provided by the embodiment of the present application is shown. Detailed implementation manners

[0039] The technical solutions in the present application will be described below with reference to the accompanying drawings.

[0040] The present application relates to a computing device, which may also be referred to as a computing power device or a computing power module, and includes a hardware device or system for executing computing tasks. Its core function is to solve large-scale computing problems in the fields of science, engineering, business, etc. through efficient data processing and computing capabilities. A computing board is a hardware module designed for high-performance computing tasks and can be used as the core component of a computing device. A computing board usually integrates a processor, high-speed memory, and an interconnect interface, aiming to provide intensive computing capabilities and optimize the energy efficiency ratio. It can be used as an independent computing unit or embedded in a server or an edge device to support high-load scenarios such as artificial intelligence training, scientific simulation, and real-time inference.

[0041] Figure 1 A schematic diagram of the computing board provided by the embodiment of the present application is shown.

[0042] Such as Figure 1As shown, the computing power board 100 may include a processor 110, a memory 120, an interconnect interface 130, a power supply device 140, and a heat dissipation device 150. Among them, the processor 110 may include: a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), etc. Optionally, in addition to the CPU, GPU, and TPU, the processor 110 may also include other types of application-specific integrated circuits (ASICs) such as a neural network processing unit (NPU), or a field programmable gate array (FPGA), etc. The specific type of the processor 110 is not limited in the embodiments of the present application.

[0043] The memory 120 may include internal memory and / or external memory for implementing data storage. Among them, the internal memory may include, for example, on-chip cache, high-bandwidth memory, or dynamic random access memory, etc. The capacity and bandwidth of the internal memory will directly affect the computing power efficiency of the computing device. The external memory may include storage devices such as a solid state drive (SSD) or a hard disk drive (HDD), etc., for long-term storage of data, models, etc., and may have a relatively high read / write speed. The specific type of the memory 120 is not limited in the embodiments of the present application.

[0044] The interconnect interface 130 may implement intra-board interconnect and / or external expansion of the computing power board 100, so as to enable data transmission and collaborative computing between multiple chips within the computing power board or between multiple computing power boards, and can improve the computing performance and parallel efficiency of the system. Among them, intra-board chip-level interconnect may be implemented by technologies such as network-on-chip (NoC) or NVLink, etc. External expansion interconnect may be implemented through high-speed interfaces such as optical communication modules.

[0045] The power supply device 140 may include a power supply and a power management module, etc. Among them, the power management module may adjust the supply voltage according to the load to reduce the idle power consumption. The heat dissipation device 150 may include, for example, heat dissipation devices in various ways such as phase change cooling or liquid cooling, etc., for dissipating heat from the hardware of the computing power board.

[0046] In high-computing scenarios, during the process of a large amount of computing and data processing by a computing device (such as the aforementioned computing board), a relatively large amount of heat is generated. Additionally, while software is becoming more versatile nowadays, it also occupies more processor performance, exerting the performance limit of the hardware as much as possible. This practice of occupying the hardware performance to the limit makes the heat generation problem of the hardware more severe. Although a heat dissipation device is configured in the computing device for dissipating heat from the hardware, the cost of the heat dissipation device is relatively high, and it cannot fundamentally solve the problem of high heat in the computing device.

[0047] In view of this, the present application provides a computing device that can effectively solve the above-mentioned high-heat problem, thereby improving the overall performance of the computing device.

[0048] Figure 2 A schematic diagram of a computing device provided by an embodiment of the present application is shown.

[0049] As Figure 2 shown, the computing device 200 includes: a main board 210 and a plurality of computing boards 220. The plurality of computing boards 220 are connected to the main board 210, and the main board 210 is configured to allocate computing tasks to at least some of the plurality of computing boards 220 according to the type of the computing tasks to be processed and the operating states of the plurality of computing boards 220.

[0050] In Figure 2 , the plurality of computing boards 220 are denoted as computing board 220-1 to computing board 220-N, where N is a positive integer greater than 1. Optionally, the computing powers of the N computing boards 220 may be the same, or the computing powers of the N computing boards 220 may be different from each other, or the computing powers of some of the N computing boards 220 may be the same and the computing powers of some of the N computing boards 220 may be different.

[0051] In some examples, the computing boards with different computing powers may include computing boards with the same hardware architecture but different hardware parameters. For example, the computing boards with different computing powers may include processors of the same type but different parameters. By way of example, among the plurality of computing boards 220, both computing board 220-1 and computing board 220-2 include a CPU, but the CPU parameters of the two are different, such as the number of cores being different, so that the computing powers of computing board 220-1 and computing board 220-2 are different.

[0052] In some other examples, the computing boards with different computing powers may also include computing boards with different hardware architectures. For example, the computing boards with different computing powers may include processors of different types. By way of example, among the plurality of computing boards 220, computing board 220-1 includes a CPU, and computing board 220-2 includes a GPU, and the types of the processors of the two are different, so that the computing powers of computing board 220-1 and computing board 220-2 are different.

[0053] Based on the computing boards with different computing powers in the multiple computing boards 220, the main board 210 can flexibly allocate computing tasks to them. The main board 210 may include a processor with task scheduling and system control functions. For example, the main board 210 may include a CPU to execute task scheduling and control of the multiple computing boards 220.

[0054] In the embodiment of the present application, the main board 210 can obtain the operating states of the multiple computing boards 220, and allocate tasks to at least some of the multiple computing boards 220 according to the operating states of the multiple computing boards 220. As an example, when a certain computing board among the multiple computing boards is in a high-load operating state, the main board can stop allocating tasks to this computing board, or when another computing board among the multiple computing boards is in a low-load operating state, the main board can allocate new tasks to this computing board. In this implementation manner, it is beneficial to fully and evenly apply the resources of the multiple computing boards in the computing device, reduce the relatively high power consumption caused by the overloaded operation of the computing boards, thereby reducing the heat generation of the computing boards, and is beneficial to improving the overall performance of the computing boards and the computing device.

[0055] Figure 3 Another schematic diagram of the computing device 200 provided by the embodiment of the present application is shown.

[0056] As Figure 3 shown, in this computing device 200, the main board 210 can be used to monitor the resource utilization rates of the multiple computing boards 220. When the resource utilization rate of a certain computing board among the multiple computing boards 220 is greater than or equal to the first preset threshold, the main board 210 is used to allocate at least some of the data in the computing task to another computing board among the multiple computing boards.

[0057] Among them, the resource utilization rate of the computing board 220 may include the utilization rates (utilizations) of one or more hardware modules in the computing board 220. For example, as Figure 1 shown, the computing board 220 may include hardware modules such as a processor and a memory. The resource utilization rate of the computing board 220 may include at least one of the following: the utilization rate of the core computing unit in the processor, the occupancy rate of the physical memory, the utilization rate of the memory bandwidth, etc. The resource utilization rate of the computing board 220 is a key indicator for measuring its hardware performance and task load balance. By monitoring the resource utilization rates of the multiple computing boards 220, the operation and load conditions of the multiple computing boards 220 can be characterized. Optionally, in addition to the several indicators described above, the resource utilization rate of the computing board 220 may also include other indicators such as the amount of operations completed per unit time and the communication efficiency between the computing board and external devices. The embodiment of the present application does not limit the specific indicator types for characterizing this resource utilization rate.

[0058] When the resource utilization rate of one of the multiple computing boards 220 is greater than or equal to the first preset threshold, it can be characterized that the load of this one computing board is relatively large, and it is not suitable to continue allocating new calculation data to it, so as to cause a significant increase in its power consumption and serious heat generation problems. In this case, the main board 210 can allocate some data in the computing task to other computing boards with smaller loads, for example, the resource utilization rate of this computing board is less than the first preset threshold.

[0059] Optionally, when the resource utilization rate of the computing board includes the utilization rate of the core computing unit in the processor, the value range of the first preset threshold can be greater than or equal to 85%, for example, the first preset threshold can be 85%, or, in some emergency situations, the first preset threshold can be 93%.

[0060] Optionally, when the resource utilization rate of the computing board includes the occupancy rate of the physical memory, the value range of the first preset threshold can be greater than or equal to 88%, for example, the first preset threshold can be 88%, or, in some emergency situations, the first preset threshold can be 95%.

[0061] In some embodiments, the main board 210 can monitor the resource utilization rate of the computing board in real time. When the resource utilization rate of the computing board exceeds the first preset threshold, the main board 210 can monitor the duration of exceeding the first preset threshold. As an example, when the resource utilization rate of a certain computing board is greater than or equal to the first preset threshold and the duration is greater than or equal to the time threshold, the main board 210 can allocate some data in the computing task to other computing boards with smaller loads. Optionally, when the resource utilization rate of the computing board includes the utilization rate of the core computing unit in the processor, the value range of the above time threshold can be greater than or equal to 3 minutes. When the resource utilization rate of the computing board includes the occupancy rate of the physical memory, the value range of the above time threshold can be greater than or equal to 5 minutes.

[0062] In the technical solution of this embodiment, by the main board monitoring the resource utilization rates of multiple computing boards, the operating states and load states of multiple computing boards can be monitored, and the data allocation of the computing task can be realized according to the operating states and load states, so as to effectively integrate and apply the resources of multiple computing boards in the computing device, reduce the possibility of a relatively large load on a certain computing board, balance the operation of multiple computing boards, and thus effectively reduce the overall power consumption and heat of the computing device.

[0063] Figure 4 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0064] As Figure 4As shown, in the computing device 200, in addition to the main board 210 and multiple computing power boards 220, at least one sensor 230 may further be included. As an example, Figure 4 sensors 230-1 to 230-N are shown, that is, each computing power board 220 is correspondingly configured with a sensor 230. In some alternative examples, the number of sensors may also be less than the number of computing power boards, and one sensor may be used to detect the temperatures of multiple computing power boards in a certain area.

[0065] In this embodiment, the main board 210 can obtain the temperatures of the multiple computing power boards 220 detected by the sensor 230. When the temperature of one of the multiple computing power boards 220 is greater than or equal to the second preset threshold, it indicates that the current heat generation problem of this computing power board is relatively serious and it is not suitable to continue allocating new calculation data to it. In this case, the main board 210 can allocate some of the data in the computing task to other computing power boards with lower temperatures, for example, the temperature of this computing power board is less than the second preset threshold. In some examples, the normal temperature of consumer electronics is 40°C to 70°C, and the temperature may reach 80°C to 90°C under high load. Optionally, for consumer electronics, the range of the above second preset threshold includes but is not limited to between 70°C and 90°C. For example, the second preset threshold can be 80°C.

[0066] Through the technical solution of this embodiment, the main board can monitor the temperatures of multiple computing power boards through the sensor, thereby determining whether the computing power board is overloaded and whether the heat generation problem is serious, and realizing the data allocation of the computing task according to the temperature of the computing power board, which can directly and effectively reduce the overall power consumption and heat of the computing device.

[0067] In some other embodiments, in addition to monitoring the temperature and / or resource utilization rate of the computing power board to perform the allocation of the computing task, the power consumption or other performance parameters of the computing power board may also be monitored to characterize the current operating state and load condition of the computing power board.

[0068] In some other embodiments, in addition to detecting the temperature of the computing power board, the sensor can also detect other parameters of the computing power board, such as humidity, pressure and other information. The main board can adjust the working mode, communication parameters, etc. of the computing power board based on this information to adapt to different application scenarios and environmental conditions.

[0069] Figure 5 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0070] As Figure 5As shown, in the computing device 200, the main board 210 can be used to obtain the computing power of multiple computing power boards 220, and allocate tasks to at least some of the multiple computing power boards 220 according to the data volume of the computing tasks to be processed and the computing power of the multiple computing power boards 220. Optionally, the computing power of at least some of the multiple computing power boards 220 is different.

[0071] In this embodiment, the main board 210 can obtain relevant parameters of multiple computing power boards 220, such as relevant parameters of one or more hardware modules (such as processors, etc.), so as to determine the computing power range of the computing power board 220. After the main board 210 obtains the computing tasks to be processed, it can determine a computing power board with appropriate computing power from multiple computing power boards based on the data volume of the computing task. For example, the data volume that the computing power board can carry can be greater than the data volume of the computing task. If the computing power of a single computing power board cannot carry the computing task, multiple computing power boards can be selected to jointly execute the computing task, and the main board 210 can allocate the data of the computing task to the multiple computing power boards according to the computing power of the multiple computing power boards.

[0072] In this implementation manner, the main board can adaptively allocate appropriate computing task data to multiple computing power boards according to their computing power, which can efficiently implement the computing task and ensure resource balance among multiple computing power boards, thereby improving the overall performance of the computing device.

[0073] Continue to refer to Figure 5 As shown, optionally, the main board 210 can also obtain the types of multiple computing power boards 220, and allocate tasks to at least some of the multiple computing power boards 220 according to the data volume of the computing tasks to be processed and the types of the multiple computing power boards 220. Optionally, the types of at least some of the multiple computing power boards are different.

[0074] In this embodiment, the main board 210 can obtain relevant parameters of multiple computing power boards 220, such as relevant parameters of one or more hardware modules (such as processors, etc.), so as to determine the type of the computing power board 220. The type of the computing power board 220 can include: the type of the processor in the computing power board 220. Optionally, the type of the processor can include the type and / or model of the processor chip, etc. According to the type of the computing power board 220, the data types (such as floating-point operations, integer operations, image processing, etc.), task types (such as real-time tasks, complex tasks, etc.) that the computing power board 220 is good at processing, and the applicable application scenarios can be determined.

[0075] Optionally, the computing tasks may include: real-time tasks, complex tasks, and hybrid tasks, where the hybrid tasks include both real-time tasks and complex tasks. Real-time tasks have high requirements for real-time performance. For example, they may include online game data processing, real-time video call data processing tasks, etc. Complex tasks have a large amount of data and high computational complexity, but may have a lower time sensitivity. For example, they may include big data analysis, log data processing, etc. Hybrid tasks need to process real-time data and perform complex data calculations. For example, they may include data processing of real-time recommendation systems, etc.

[0076] After the motherboard 210 obtains the types of multiple computing power boards 220, according to the type of computing tasks, it can match them to the appropriate computing power boards. That is, among the multiple computing power boards 220, there may be a first computing power board suitable for processing real-time tasks, and there may also be a second computing power board suitable for processing hybrid tasks. The first computing power board can meet the computing tasks with high real-time performance and has a low latency. As an example, the processors on the first computing power board may include real-time processors, FPGAs, etc. The second computing power board can achieve high-volume data, high parallelism, or high-precision data calculations. As an example, the processors on the second computing power board may include multi-core CPUs, GPUs, TPUs, custom ASICs, etc. The embodiments of the present application do not limit the specific hardware types of the first computing power board and the second computing power board.

[0077] In some examples, the first computing power board may include at least one of the following hardware modules: a real-time processing unit, a cache, or a parallel computing unit. The first computing power board can utilize its cache and real-time processing unit to reduce the latency of task processing.

[0078] In some examples, the second computing power board may include at least one of the following hardware modules: a batch processing computing unit, a high-capacity memory, or a high-speed network interface. The second computing power board can utilize its powerful computing ability and large memory space to quickly process a large amount of data.

[0079] When the computing tasks received by the motherboard 210 include real-time tasks, the motherboard 210 can be used to allocate the real-time tasks to the first computing power board. If the computing tasks include complex tasks, the motherboard 210 is used to allocate the complex tasks to the second computing power board. If the computing tasks include hybrid tasks, the motherboard 210 is used to allocate the real-time tasks in the hybrid tasks to the first computing power board and allocate the complex tasks in the hybrid tasks to the second computing power board. Optionally, when multiple computing power boards execute hybrid tasks, through a task distribution mechanism and a data synchronization mechanism, the coordinated work between multiple computing power boards can be ensured.

[0080] Optionally, in some embodiments, during the execution of computing tasks by the first computing board and / or the second computing board, the main board 210 is further configured to adjust the power consumption of the first computing board and / or the second computing board according to the execution status of the computing tasks. For example, during the execution of real-time tasks by the first computing board, the power consumption can be optimized by adjusting the voltage and frequency to ensure that the real-time requirements can still be met under low power consumption. During the execution of complex tasks by the second computing board, the parallelism of the computing units in the second computing board can be dynamically adjusted and the energy-saving mode can be used to balance the power consumption and performance. During the execution of hybrid tasks by the first computing board and the second computing board, the power consumption and performance of multiple computing boards can be dynamically adjusted according to the real-time nature and computational amount of the tasks.

[0081] Optionally, in some other embodiments, the main board can further optimize the configuration and power consumption scheme according to factors such as the priority of the tasks, the user's usage habits, and the system load. For example, for high-priority tasks, computing resources can be preferentially allocated and the performance can be improved. For low-priority tasks, the power consumption can be reduced to extend the battery life of the device while ensuring the basic performance.

[0082] Through the technical solution of the embodiments of the present application, the main board can adaptively allocate data of computing tasks suitable for processing to at least some of the multiple computing boards according to the types of the computing tasks and the types of the multiple computing boards, so that the multiple computing boards can better exert their respective performances, efficiently execute the computing tasks, and the resources of the computing boards are relatively balanced. While reducing the power consumption and heat of the computing device, the processing efficiency and processing effect of the computing tasks can be improved, thereby improving the comprehensive performance of the computing device.

[0083] Figure 6 Another schematic diagram of the computing device provided by the embodiments of the present application is shown.

[0084] As Figure 6 shown, in the computing device 200, the multiple computing boards include: a first computing board group (schematically shown as computing board A group in the figure) and a second computing board group (schematically shown as computing board B group in the figure). The first computing board group includes multiple first computing boards (schematically shown as including computing boards A1, A2, A3, etc. in the figure), and the second computing board group includes multiple second computing boards (schematically shown as including computing boards B1, B2, B3, etc. in the figure). Among them, the computing powers of at least some of the multiple first computing boards are different, and / or the computing powers of at least some of the multiple second computing boards are different.

[0085] In the embodiments of the present application, multiple first computing power boards in the first computing power board group can be applicable to processing real-time data with low latency and high concurrency, that is, the first computing power board group can be preferably used for processing real-time tasks. Multiple second computing power boards in the second computing power board group can be applicable to large-scale data processing and complex calculations, that is, the second computing power board group can be preferably used for processing complex tasks.

[0086] In the first computing power board group and / or the second computing power board group, computing power boards with different computing powers can be respectively set. For example, any computing power board group can include computing power boards with three different computing powers: strong computing power boards, medium computing power boards, and weak computing power boards. Among them, the strong computing power board has powerful computing capabilities and is usually equipped with high-performance computing chips, such as high-end GPUs, FPGAs, or specially customized ASIC chips, etc. In addition, the strong computing power board can have a large number of computing cores and high-speed memory, and can process complex computing tasks. The medium computing power board has moderate computing capabilities, and the configured chips and hardware resources are relatively lower than those of the strong computing power board. The medium computing power board is suitable for some tasks that do not have particularly high computing requirements but still require a certain processing ability. The weak computing power board has relatively weak computing capabilities and is mainly used for some simple computing tasks or cost-sensitive scenarios. The weak computing power board may use low-power and low-cost chips.

[0087] Optionally, in some embodiments, the memory capacities of at least some of the first computing power boards in the first computing power board group are different, and / or the memory capacities of at least some of the second computing power boards in the second computing power board group are different.

[0088] Based on the technical solution of the embodiments of the present application, the computing device can not only have different types of computing power boards, but also the same type can include multiple computing power boards with different computing powers, so that multiple different types of computing power board groups can be preferably applicable to different computing tasks, enabling the computing device to flexibly adapt to the execution of different computing tasks and more effectively improving the overall performance of the computing device.

[0089] Optionally, the main board can be used to sequentially allocate real-time tasks to multiple first computing power boards in the first computing power board group. When the resource utilization rate of the previous first computing power board is greater than or equal to the preset threshold, the main board is used to allocate at least part of the data in the real-time task to the next first computing power board; and / or the main board can be used to sequentially allocate complex tasks to multiple second computing power boards in the second computing power board group. When the resource utilization rate of the previous second computing power board is greater than or equal to the preset threshold, the main board is used to allocate at least part of the complex task to the next second computing power board.

[0090] Figure 7 A data processing flow chart of a computing task provided by the embodiments of the present application is shown.

[0091] As Figure 7As shown, the main board can receive computing tasks, obtain the data therein, and the main board can judge the complexity and real-time requirements of the data. When the complexity of the data is relatively high, the computing task belongs to a complex task, and the main board can allocate the data in the complex task to the B group of computing power boards (the second group of computing power boards). When the real-time nature of the data is relatively high, the computing task belongs to a real-time task, and the main board can allocate the data in the real-time task to the A group of computing power boards (the first group of computing power boards).

[0092] Optionally, the main board can first allocate the real-time task to any one of the computing power boards in the A group of computing power boards, such as computing power board A1. The main board can further monitor the resource utilization rate of computing power board A1 and judge whether the resource utilization rate of computing power board A1 is greater than or equal to a preset threshold. As an example, the preset threshold can be 85%. If the resource utilization rate of computing power board A1 is less than the preset threshold, then the current computing power board A1 is not operating in an overloaded state, and computing power board A1 can process the data received for the real-time task, where the data can include, for example, image data, signal data, etc., and the type of data is not limited in this application. The main board can continue to monitor the resource utilization rate of computing power board A1 and analyze the power consumption of computing power board A1, so as to adjust the power consumption of computing power board A1 according to the operating conditions of computing power board A1.

[0093] If the resource utilization rate of computing power board A1 is greater than or equal to the preset threshold, then the current resource utilization rate of computing power board A1 is relatively high, which will generate relatively large power consumption and heat, and it is not suitable to continue allocating new data to computing power board A1. The main board can divert some of the data in the implementation task to computing power board A2 and judge whether the resource utilization rate of computing power board A2 is greater than or equal to the preset threshold. If not, then computing power board A2 can process the data received for the real-time task. If so, the main board can continue to divert some of the data in the implementation task to computing power board A3. And so on, the main board can control multiple computing power boards in the A group of computing power boards to jointly execute the real-time task, and the resource utilization rate of each computing power board among the multiple computing power boards does not exceed the preset threshold, and the temperature and power consumption are both at a relatively low level.

[0094] Similar to the processing process of the A group of computing power boards above, the main board can allocate complex tasks to the B group of computing power boards. When multiple computing power boards in the B group of computing power boards jointly execute complex tasks, the resource utilization rate of each computing power board among the multiple computing power boards does not exceed the preset threshold, and the temperature and power consumption are both at a relatively low level.

[0095] Optionally, continue to refer to Figure 7As shown, when the data received by the motherboard neither meets the real-time requirement nor the complexity requirement, the motherboard can be used to process the data. Optionally, a task carrying the data that neither meets the real-time requirement nor the complexity requirement can be called a regular task. When the motherboard executes a regular task, the motherboard can monitor its own resource utilization rate and analyze its own power consumption, so as to adjust its own power consumption according to the running situation.

[0096] In some alternative embodiments, when the resource utilization rate of a certain computing board is greater than or equal to a preset threshold, in addition to performing resource balancing and diverting the received data to other computing boards, the motherboard can also allocate all the received data to other computing boards with stronger computing power. Optionally, during the process of the motherboard allocating tasks to multiple computing boards, the computing power of the subsequent computing board assigned with tasks can be greater than that of the previous computing board.

[0097] Optionally, the motherboard can be used to obtain the computing power of multiple first computing boards in the first computing board group, and allocate real-time tasks to at least some of the multiple first computing boards according to the computing power of the multiple first computing boards; and / or, the motherboard is used to obtain the computing power of multiple second computing boards in the second computing board group, and allocate complex tasks to at least some of the multiple second computing boards according to the computing power of the multiple second computing boards.

[0098] For example, when the motherboard allocates tasks to a computing board group, if the computing power of the subsequent computing board is greater than or equal to the sum of the computing powers of the previous two computing boards, the motherboard can allocate all the data on the previous two computing boards to the subsequent computing board. If the computing power of the subsequent computing board is less than the sum of the computing powers of the previous two computing boards, the motherboard can divert some of the data in the task to the subsequent computing board.

[0099] In this embodiment, the motherboard can perform more accurate task allocation to multiple computing boards in the computing board group according to the computing power of the computing boards, so as to use appropriate resources to process tasks, thereby greatly reducing the temperature and power consumption of the computing boards and improving the overall performance of the computing device.

[0100] To further improve the data interaction performance between the motherboard and multiple computing boards, at least one of the following solutions can be executed:

[0101] (1) Task distribution mechanism: Through an intelligent task distribution mechanism (such as a priority-based queue, load balancing algorithm, etc.), the motherboard dynamically allocates tasks to the most suitable computing board according to the characteristics of the tasks such as real-time nature, computing volume, and data type.

[0102] (2) Data synchronization and sharing: To enable collaborative work among multiple computing boards, an efficient data synchronization and sharing mechanism needs to be established. This can be achieved by using shared memory, message queues, databases, etc., to ensure data consistency and integrity across different computing boards.

[0103] (3) Asynchronous calls and callbacks: In some cases, one computing board may need to wait for another to complete certain calculations before it can continue execution. In such situations, asynchronous calls and callbacks can be adopted to ensure the continuity and efficiency of task execution.

[0104] In any of the above embodiments, the main board may include an artificial intelligence (AI) computing main board.

[0105] The AI computing main board, which can also be simply referred to as the AI main board, is a main board specifically designed for artificial intelligence (AI) computing tasks. It has powerful computing capabilities and dedicated hardware acceleration modules to meet the requirements of AI applications for massive data processing and complex algorithms. The AI main board can process the following types of information.

[0106] Data parallel processing: The AI main board can efficiently process parallel data, thanks to its built-in high-performance AI chips and optimized software algorithms. This parallel processing ability is crucial for complex computing tasks such as deep learning and machine learning, and can significantly improve the overall computing efficiency.

[0107] Complex algorithm calculations: The AI main board supports the running of complex algorithms such as deep learning and machine learning. These algorithms have extensive applications in fields such as image recognition, speech processing, and natural language processing, and require powerful computing capabilities to support algorithm training and inference. The AI main board accelerates the processing speed of these algorithms by providing interfaces and protocols specifically designed for AI.

[0108] Audio and video data: The AI main board performs excellently in audio and video processing and can provide users with high-quality AI audio and video processing services in real time. For example, in the field of autonomous driving, the AI main board can process a large amount of audio and video data from sensors and make real-time responses; in smart homes, it supports intelligent adjustment of the home environment, such as optimizing home security, entertainment, etc. through the analysis of audio and video data.

[0109] Industrial-grade communication and data: The AI main board has rich peripheral interfaces and supports multiple communication protocols and data transmission standards. This enables it to play an important role in fields such as industrial automation and intelligent manufacturing, achieving efficient communication and data exchange between devices.

[0110] In addition, the AI motherboard also has technical advantages such as efficient energy consumption management, support for fast model training and inference, making the deployment and update of AI applications more efficient. With the continuous progress of AI technology and the expansion of application fields, the AI motherboard will play an important role in more fields, promoting the popularization and development of artificial intelligence technology.

[0111] In summary, the data mainly processed by the AI motherboard includes parallel data, complex algorithm calculation results, audio-visual data, and industrial-level communication and data, etc. The processing capabilities of these data enable the AI motherboard to have a wide range of application prospects in the field of artificial intelligence.

[0112] Figure 8 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0113] As Figure 8 shown, the computing device 200 schematically shows a motherboard 210, computing power boards 220-1 to 220-6, and sensors 230-1 to 230-4. Optionally, the computing power boards 220-1 to 220-6 can be divided into two groups, for example, the first computing power board group and the second computing power board group described in the above embodiment. The sensors 230-1 to 230-4 can be used to monitor information such as the temperature, humidity, and pressure of multiple computing power boards. Based on these status information and other information (such as power) of multiple computing power boards monitored by the motherboard 210, the motherboard 210 can automatically adjust the working modes, communication parameters, etc. of multiple computing power boards to adapt to different application scenarios and environmental conditions.

[0114] Optionally, as Figure 8 shown, the processor in the motherboard 210 may include one or more of a CPU, a GPU, and an NPU. In some examples, the motherboard 210 may at least include a CPU, and on this basis, other types of processors such as a GPU and an NPU may be further included.

[0115] Figure 9 A schematic diagram of the motherboard provided by the embodiment of the present application is shown.

[0116] As Figure 9 shown, in addition to at least one of a CPU, a GPU, and an NPU, the motherboard 210 may further include a microcontroller. Optionally, one or more types of interfaces may be provided on the motherboard 210. As an example, Figure 9The main board 210 may include at least one of the following interfaces: high definition multimedia interface (HDMI), display port (DP), serial digital interface (SDI), serial interface for digital audio transmission between integrated circuits (I2S), inter-integrated circuit bus interface (I2C), universal serial bus (USB) interface, bi-directional serial bus interface (inter integrated circuit), general purpose input / output (GPIO) pin interface, peripheral component interconnect express (PCIe) interface, universal asynchronous receiver / transmitter (UART) interface, etc.

[0117] Among them, the HDMI interface can be used to receive HDMI input. The USB interface can be used to connect USB devices, camera devices (CAM), or network devices (such as Wifi). The I2S interface can be used to connect functional chips, such as the ACM8622M digital audio power amplifier chip shown in the figure. The DP interface can be used to receive display data input. The I2C interface can be used to receive, and the GPIO interface can be used to connect external devices such as network ports or fans.

[0118] In addition to the above interfaces, the following at least one functional component may be provided on the main board 210: crystal oscillator (XTAL), pulse-width modulation (PWM), pulse density modulation (PDM), embedded multi-media card (eMMC), double data rate (DDR) synchronous dynamic random access memory, successive approximation register (SAR) analog-to-digital converter, etc. Among them, the PWM can be used to control the fan speed regulation. The PDM can be used to control the microphone (MIC). The SAR analog-to-digital converter can be used to connect to a negative temperature coefficient thermistor (NTC). DDR4 is the 4th generation of DDR, which can perform faster and lower-power storage.

[0119] Figure 10 Another schematic diagram of the computing device provided by the embodiment of the present application is shown.

[0120] As Figure 10 shown, the computing device 200 may include a CPU and multiple computing power board sub-boards (also referred to as computing power boards above). In some embodiments, the CPU may be located on the main board, and the multiple computing power board sub-boards are connected to the CPU through the main board. As an example, the CPU may include a control unit, a temperature control unit, a logic analysis unit, and a computing power analysis unit. These multiple units can work together to control the operation of the multiple computing power board sub-boards.

[0121] Figure 10 As a schematic illustration, 6 computing power board sub-boards are shown. Among them, the processors in computing power board sub-boards 1 to 3 may include CPUs and CPUs, and the processors in computing power board sub-boards 4 to 6 may include CPUs and NPUs. These two different types of computing power board sub-boards can be used to process different types of data, such as the real-time data and complex data described above. The computing power board sub-boards can be interconnected with the CPU through high-speed interfaces to achieve high-speed data transmission. In addition, a storage unit may be provided in the computing power board sub-board to store the data processed by the computing power board sub-board. In addition to data transmission, the main board can also detect the temperature information of the computing power board sub-board, so that the main board can adjust the operating state of the computing power board sub-board according to the temperature of the computing power board sub-board.

[0122] In addition to the computing power board sub-board and the CPU, the computing device may further include a storage unit, sensors, a camera, a high definition multimedia interface (HDMI), a DisplayPort (DP), a serial digital interface (SDI), a direct current to direct current (DC to DC) conversion unit, etc., connected to the CPU.

[0123] In Figure 10 In the illustrated example, different types of computing power boards are good at different computing tasks. For example, the CPU is good at general computing, the GPU is suitable for graphics and parallel computing, and the NPU is optimized for AI tasks. Integrating them into a computing device enables heterogeneous computing, allowing various computing tasks to run on the most suitable hardware, giving full play to the advantages of each computing power board, thereby improving the overall computing performance and better meeting the requirements for multiple computing capabilities in complex AI tasks.

[0124] Figures 8 to 10 Merely for illustration, several schematic structural block diagrams of the computing device provided by the present application are shown, and this illustration should not limit the present application.

[0125] Based on the computing device provided by the present application, at the underlying driver and operating system levels, the invocation and cooperation of the computing modules (such as including the computing power boards) also rely on efficient driver programs and operating system support. The driver program is responsible for controlling the hardware resources inside the module to ensure that computing units, communication modules, etc. can work properly. The operating system provides a unified software platform to support functions such as multitasking, resource management, and security protection. By optimizing the driver program and the operating system, the performance and stability of the module can be further improved, while reducing power consumption and costs.

[0126] At the level of dynamic computing power allocation, in order to adapt to different computing power requirements, the module also needs to have the ability of dynamic computing power allocation. This is usually achieved through intelligent scheduling algorithms, which can dynamically adjust the working mode and resource allocation of the computing units according to the requirements of the current task and the actual state of the module. For example, during peak task periods, the parallel processing ability of the computing units can be increased to improve the processing speed; while during low task periods, the power consumption of the computing units can be reduced to save energy.

[0127] At the level of status monitoring and adaptive adjustment, for the adaptation to different status quantities, the module needs to have the capabilities of status monitoring and adaptive adjustment. By integrating sensors and status monitoring modules, the module can perceive the status changes of itself and the external environment in real time, such as temperature, humidity, pressure, power, etc. Based on this status information, the module can automatically adjust its working mode, communication parameters, etc., to adapt to different application scenarios and environmental conditions.

[0128] Analyze the computing power adaptability. For each type of module, analyze in detail its computing power range, the types of data it is good at processing (such as floating-point operations, integer operations, image processing, etc.), and the applicable application scenarios. Based on this, the principle of how modules perform computing power scheduling, data sharing, and collaborative work according to task requirements can be deeply analyzed. Technical means include but are not limited to hardware acceleration technology, parallel processing strategies, task queue management, etc.

[0129] Combined with the analysis of actual usage scenarios, tasks can be classified according to business requirements into tasks with high real-time requirements (such as video processing), compute-intensive tasks (such as deep learning training), etc., and an optimal computing power module configuration plan can be designed for each task type. For each configuration plan, provide detailed power consumption estimates and energy-saving measures. For example, adopt dynamic voltage and frequency scaling (DVFS), task merging to reduce startup power consumption, and use idle periods for low-power mode switching, etc.

[0130] As an example, the AI motherboard first analyzes the input task to determine its type, such as whether it is a deep learning training, inference, or image / video processing, natural language processing, etc. Different types of tasks have different requirements and emphases on computing power. Then the AI motherboard performs a computing power requirement assessment, and evaluates the computing power resources required to complete the task according to factors such as the scale, complexity, and real-time requirements of the task. For example, training a large neural network model may require multiple high-performance GPUs or specialized AI chips to provide sufficient computing power, while a simple image classification task can be completed by some low-power NPUs. Subsequently, the AI motherboard can allocate the task to different computing power boards. For example, for compute-intensive deep learning tasks, it will be preferentially allocated to computing power boards such as GPUs or NPUs that are good at parallel computing. For general computing tasks, such as data preprocessing, logical control, etc., they can be processed by the CPU.

[0131] This application also provides an electronic device, including the computing device of any of the above embodiments.

[0132] Optionally, the electronic device can be applied to multiple fields such as display, payment, vehicle-mounted, robotics, security, industrial inspection and control. In some examples, the electronic device may include a video conferencing device, which may include, for example, AI capabilities.

[0133] The computing device and the electronic device provided by the embodiments of the present application can have the advantages of combining general computing and heterogeneous computing, can cover different levels of AI computing power intervals, and fully match the requirements of edge-side computing and edge computing. With the continuous expansion of application scenarios, the computing device and various types of consumer or Internet of Things terminals and other electronic devices in which it is located have gradually become new carriers of computing power.

[0134] The term "and / or" in this application is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: the existence of A, the simultaneous existence of A and B, and the existence of B.

[0135] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by those skilled in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments, and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects, rather than to describe a specific order or primary-secondary relationship.

[0136] Referring to "embodiment" in this application means that a specific feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of this application. The appearance of this phrase in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments.

[0137] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0138] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0139] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0140] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0141] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0142] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A computing device, characterized in that, Comprising: A main board and a plurality of computing power boards, the plurality of computing power boards being connected to the main board, and the main board being configured to allocate the computing task to at least some of the plurality of computing power boards according to the computing task to be processed and the operating states of the plurality of computing power boards.

2. The computing device according to claim 1, wherein The operating states of the plurality of computing power boards include resource utilization rates. The main board is configured to monitor the resource utilization rates of the plurality of computing power boards. When the resource utilization rate of one of the plurality of computing power boards is greater than or equal to a first preset threshold, the main board is configured to allocate at least some of the data in the computing task to another one of the plurality of computing power boards.

3. The computing device according to claim 1, wherein The operating states of the plurality of computing power boards include temperatures. The computing device further includes a sensor configured to detect the temperatures of the plurality of computing power boards and send them to the main board. When the temperature of one of the plurality of computing power boards is greater than or equal to a second preset threshold, the main board is configured to allocate at least some of the data in the computing task to another one of the plurality of computing power boards.

4. The computing device according to any one of claims 1 to 3, characterized in that, At least some of the plurality of computing power boards have different computing powers. The main board is configured to obtain the computing powers of the plurality of computing power boards and perform task allocation to at least some of the plurality of computing power boards according to the data volume of the computing task to be processed and the computing powers of the plurality of computing power boards.

5. The computing device according to any one of claims 1 to 3, characterized in that, At least some of the plurality of computing power boards have different types. The main board is configured to obtain the types of the plurality of computing power boards and perform task allocation to at least some of the plurality of computing power boards according to the type of the computing task to be processed and the types of the plurality of computing power boards.

6. The computing device according to claim 5, wherein The computing task includes: real-time tasks, complex tasks, and hybrid tasks. The plurality of computing power boards includes a first computing power board and a second computing power board; If the computing task includes a real-time task, the main board is configured to allocate the real-time task to the first computing power board. If the computing task includes a complex task, the main board is configured to allocate the complex task to the second computing power board. If the computing task includes a hybrid task, the main board is configured to allocate the real-time task in the hybrid task to the first computing power board and allocate the complex task in the hybrid task to the second computing power board.

7. The computing device according to claim 6, wherein The plurality of computing power boards includes: a first computing power board group and a second computing power board group. The first computing power board group includes a plurality of first computing power boards, and the second computing power board group includes a plurality of second computing power boards; At least some of the plurality of first computing power boards have different computing powers, and / or at least some of the plurality of second computing power boards have different computing powers.

8. The computing device according to claim 7, wherein The main board is configured to sequentially allocate the real-time task to the plurality of first computing power boards in the first computing power board group. When the resource utilization rate of the previous first computing power board is greater than or equal to the first preset threshold, the main board is configured to allocate at least some of the data in the real-time task to the next first computing power board; And / or The main board is used to sequentially allocate the complex task to multiple second computing power boards in the second computing power board group. When the resource utilization rate of the previous second computing power board is greater than or equal to the first preset threshold, the main board is used to allocate at least part of the data in the complex task to the next second computing power board.

9. The computing device according to claim 7, wherein The main board is used to obtain the computing power of multiple first computing power boards in the first computing power board group, and allocate the real-time task to at least part of the first computing power boards among the multiple first computing power boards according to the computing power of the multiple first computing power boards; and / or The main board is used to obtain the computing power of multiple second computing power boards in the second computing power board group, and allocate the complex task to at least part of the second computing power boards among the multiple second computing power boards according to the computing power of the multiple second computing power boards.

10. The computing device according to claim 6, wherein During the process of the first computing power board and / or the second computing power board executing the computing task, the main board is further used to adjust the power consumption of the first computing power board and / or the second computing power board according to the execution status of the computing task.

11. The computing device according to claim 6, wherein The first computing power board includes at least one of the following hardware modules: a real-time processing unit, a cache, or a parallel computing unit; The second computing power board includes at least one of the following hardware modules: a batch processing computing unit, a high-capacity memory, or a high-speed network interface.

12. The computing device according to claim 6, wherein The computing task further includes a regular task, and the main board is used to execute the regular task.

13. The computing device according to any one of claims 1 to 3, characterized in that, The main board includes an artificial intelligence computing power main board.

14. An electronic device, characterized in that, including: The computing device according to any one of claims 1 to 13.

Images