Method and device for dynamically metering computing power of intelligent computing center cloud platform based on computing power operation task

Abstract

The invention provides a method and a device for dynamically metering computing power based on computing power operation tasks of an intelligent computing center cloud platform, and relates to the technical field of intelligent computing centers, intelligent computing centers, computing power infrastructures and intelligent computing cloud. The method for dynamically metering the computing power of the intelligent computing center cloud platform based on the computing power operation task comprises the following steps: S1, acquiring a parameter value of first operation data of a first GPU in a first time period, wherein the first GPU is a GPU for operating a first computing power operation task; s2, determining a first task type of the first computing power operation task based on a parameter value of the first operation data; s3, determining a weighting coefficient corresponding to the first task type; and S4, based on the weighting coefficient and the parameter value of the first operation data, calculating to obtain the calculation power value of the first calculation power operation task in the first time period. According to the method, the computing power metering accuracy of the computing power operation task in the intelligent computing center cloud platform can be greatly improved.

Description

Technical Field

[0001] This invention relates to the fields of intelligent computing centers, smart computing centers, computing infrastructure, and smart cloud technologies, specifically to a method and apparatus for dynamically measuring computing power based on computing power operation tasks on an intelligent computing center cloud platform. Background Technology

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "smart computing centers" have emerged.

[0003] An "intelligent computing center" refers to a facility that provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models) by utilizing large-scale heterogeneous computing resources, including general-purpose and intelligent computing power. Intelligent computing centers encompass facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] "Intelligent computing center" includes, but is not limited to, "intelligent computing center".

[0005] "Intelligent computing center" or artificial intelligence computing center is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing center" and "smart computing center". It is the ability of computer equipment or computing / data center to process parameters. It is the ability of computer hardware and software to work together to execute a certain computing requirement. It is the computing power to achieve the target result output by processing parameter data. It is a new type of productivity that integrates parameter computing power, network carrying capacity and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] During the execution of computing tasks in an intelligent computing center, it is necessary to measure these tasks to allocate appropriate computing resources. In the applicant's related patent application (application number No. 202411224087.0), the computing power value that the processor can provide is calculated by collecting the processor's half-precision floating-point computing power, video memory bandwidth, and video memory capacity. However, in existing technologies, the computing power value that the processor can provide can only be calculated based on the processor's parameters. In actual computing task execution, different tasks have different resource usage patterns, with some types of resources being used more than others, such as consuming more floating-point computing resources and less video memory bandwidth. In such cases, it is impossible to accurately measure the computing power required by the task based solely on the processor's available computing power, resulting in a very low accuracy rate in measuring computing power in intelligent computing centers.

[0008] It is evident that since the emergence of intelligent computing centers, the accuracy of computing power measurement for computing power operation tasks has been very low, and this problem has always been a pressing issue that needs to be addressed in this field. Summary of the Invention

[0009] This invention provides a method and apparatus for dynamically measuring computing power based on computing power running tasks in an intelligent computing center cloud platform, in order to solve the problem of very low accuracy in measuring computing power of computing power running tasks in the prior art.

[0010] To solve the above problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for dynamically measuring computing power based on computing power in a cloud platform for intelligent computing centers, comprising: Step S1: Obtain the parameter values ​​of the first running data of the first GPU within the first time period, where the first GPU is the GPU running the first computing power running task; Step S2: Determine the first task type of the first computing power running task based on the parameter values ​​of the first running data; Step S3: Determine the weighting coefficients corresponding to the first task type; Step S4: Based on the weighting coefficient and the parameter values ​​of the first running data, calculate the computing power value of the first computing power running task within the first time period.

[0011] In one embodiment, step S2 includes: Step S21: Determine the first task type of the first computing power running task based on the first mapping table and the first running data. The first mapping table includes multiple task types and a running data value range corresponding to each task type. The multiple task types include the first task type, and the parameter values ​​of the first running data are within the running data value range corresponding to the first task type.

[0012] In one embodiment, step S3 includes: Step S31: Determine the weighting coefficient corresponding to the first task type based on the second mapping table. The second mapping table includes multiple task types and the weighting coefficient corresponding to each task type.

[0013] In one embodiment, the first runtime data includes at least one of floating-point computing utilization, video memory capacity, and video memory bandwidth; The number of weighting coefficients corresponding to the first task type matches the number of parameters included in the first running data.

[0014] In one embodiment, the first running data includes floating-point calculation utilization, and step S1 includes: Step S11: Obtain the duration percentage and utilization parameter value of each floating-point calculation type among multiple floating-point calculation types within the first time period; Step S12: Calculate the floating-point calculation utilization parameter value based on the time percentage and utilization parameter value of the multiple floating-point calculation types.

[0015] In one embodiment, step S4 includes: Step S41: Calculate the relative parameter value, wherein the relative parameter value is the ratio of the parameter value of the first running data to the preset rated value; Step S42: The product of the relative parameter value and the weighting coefficient is set as the computing power value of the first computing power running task in the first time period.

[0016] In one embodiment, when the first running data includes at least two parameters among floating-point calculation utilization, video memory capacity, and video memory bandwidth, step S4 includes: Step S41': Obtain the preset rated value corresponding to each of the at least two parameters; Step S42': Calculate the relative parameter value of each parameter. The relative parameter value of the first parameter is the ratio of the parameter value of the first parameter to the preset rated value corresponding to the first parameter. The first parameter is any one of the at least two parameters. Step S43': Based on the weighting coefficient, the relative parameter values ​​of each parameter are weighted to obtain the computing power value of the first computing power running task within the first time period.

[0017] In one embodiment, the method further includes: Step S5: Obtain parameter values ​​of the first running data of the first GPU within multiple consecutive time periods, wherein the multiple time periods are the time periods during which the first GPU executes the first computing power running task, and the multiple time periods include the first time period; Step S6: Based on the parameter values ​​of the first running data of the first GPU in the multiple time periods, calculate the computing power value of the first computing power running task in each time period; Step S7: Set the sum of the computing power values ​​of the multiple time periods as the total computing power value of the first computing power running task.

[0018] Secondly, the present invention also provides a device for dynamically measuring computing power based on computing power operation tasks in an intelligent computing center cloud platform, comprising: The first acquisition module is used to acquire parameter values ​​of the first running data of the first GPU within a first time period, wherein the first GPU is the GPU running the first computing power running task; The first determining module is used to determine the first task type of the first computing power running task based on the parameter values ​​of the first running data. The second determining module is used to determine the weighting coefficients corresponding to the first task type; The first calculation module is used to calculate the computing power value of the first computing power running task within the first time period based on the weighting coefficient and the parameter values ​​of the first running data.

[0019] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the method for dynamically measuring computing power based on computing power in the intelligent computing center cloud platform described in the first aspect above.

[0020] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the method for dynamically measuring computing power based on computing power in a cloud platform for intelligent computing centers as described in the first aspect above.

[0021] Fifthly, the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the method for dynamically measuring computing power based on computing power in a cloud platform for intelligent computing centers as described in the first aspect above.

[0022] In this invention, step S1 involves obtaining parameter values ​​of the first running data of the first GPU within a first time period, where the first GPU is the GPU running the first computing power task; step S2 involves determining the first task type of the first computing power task based on the parameter values ​​of the first running data; step S3 involves determining the weighting coefficient corresponding to the first task type; and step S4 involves calculating the computing power value of the first computing power task within the first time period based on the weighting coefficient and the parameter values ​​of the first running data. Thus, by determining the first task type of the first computing power task within the first time period, and then determining the weighting coefficient based on the first task type, the weighting coefficient can determine the contribution of the first running data to the computing power value for the first task type. Furthermore, by calculating the computing power value of the first computing power task within the first time period based on the weighting coefficient and the parameter values ​​of the first running data, the computing power resources occupied by the first computing power task within the first time period can be accurately measured based on the parameter values ​​of the first running data of the first computing power task within the first time period, greatly improving the accuracy of computing power measurement for computing power tasks in the intelligent computing center cloud platform. Attached Figure Description

[0023] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a method for dynamically measuring computing power based on computing power in a cloud platform for intelligent computing centers, provided by the present invention. Figure 2 This is a schematic diagram of the measurement factor for the computing power of the hardware device provided by the present invention; Figure 3 This is a schematic diagram of the computing power of the first computing power running the task provided by the present invention; Figure 4 This is a structural diagram of a device for dynamically measuring computing power based on computing power to run tasks in an intelligent computing center cloud platform, provided by the present invention. Figure 5 This is a structural diagram of an electronic device provided by the present invention. Detailed Implementation

[0025] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] The “computing power” mentioned in this invention refers to: the ability of computer equipment or computing / data center to process information; the ability of computer hardware and software to work together to perform a certain computing requirement; the computing power to achieve the target result output by processing information data; and a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0027] The "computational power" (CP) described in this invention refers to the ability of a data center server to process data and output results. It is a comprehensive indicator of a data center's computing power, encompassing general computing power, supercomputing power, and intelligent computing power. The commonly used unit of measurement is floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), with higher values ​​indicating stronger overall computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A supercomputers, 500,000 mainstream server CPUs, or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0028] The "Network Power" (NP) mentioned in this invention refers to the performance of data transmission capability of computing facilities, which includes comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, and involves network transmission within and between data centers. It is a comprehensive indicator for measuring network transmission scheduling capability.

[0029] The "Storage Power" (SP) described in this invention refers to the comprehensive capabilities of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon operation. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and internal storage devices within servers. The commonly used unit of measurement for storage capacity is exabytes (EB, 1EB = 2^60 bytes), while the commonly used unit of measurement for performance is the number of read / write operations per second (IOPS / TB). Disaster recovery ratio is an important indicator of security and reliability.

[0030] The "computing infrastructure" mentioned in this invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, enabling centralized computing, storage, transmission, and application of information.

[0031] The "new information infrastructure" mentioned in this invention refers to network infrastructure such as 5G networks, fiber optic broadband networks, backbone networks, international communication networks, and satellite internet; computing infrastructure such as data centers, general computing centers, intelligent computing centers, and supercomputing centers; and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0032] The “computing power” mentioned in this invention includes: general computing power, intelligent computing power, and supercomputing power.

[0033] The "general computing power" mentioned in this invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0034] The "intelligent computing power" mentioned in this invention refers to: a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for various artificial intelligence innovative applications, such as natural language processing and machine vision.

[0035] The “supercomputing power” mentioned in this invention refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0036] The "intelligent computing center" described in this invention refers to a facility that, through the use of large-scale heterogeneous computing resources, including general-purpose computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), primarily provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models). The intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0037] The "intelligent computing center cloud platform" mentioned in this invention, abbreviated as "intelligent computing cloud", refers to a cloud computing platform that integrates hardware and software resources based on an intelligent computing center.

[0038] The "intelligent computing center" mentioned in this invention includes, but is not limited to, "smart computing center".

[0039] The "intelligent computing center" mentioned in this invention, also known as an artificial intelligence computing center, is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0040] The "computing center" mentioned in this invention refers to a facility that is mainly composed of infrastructure such as wind, thermal, hydro, and electricity, and IT hardware and software equipment, and has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0041] The "supercomputing center" mentioned in this invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters. It can provide large-scale computing, storage and network services and is widely used in aerospace, oil exploration, climate modeling and genome sequencing and other application scenarios.

[0042] The “computing resources” mentioned in this invention refer to the technologies and facilities required for the development of the digital society that have the ability to compute, transmit, store and apply information, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guaranteeing resources such as wind, fire, water and electricity.

[0043] The "model" mentioned in this invention includes, but is not limited to, "large language model" and "multimodal large model".

[0044] The "large language model" mentioned in this invention refers to a large-scale language model (LLM), which is a language model with a large number of parameters. It is designed to understand and generate human language, and is trained with a large amount of text data. It can perform a wide range of tasks, including text summarization, translation, and sentiment analysis.

[0045] The “Multimodal Large Models” mentioned in this invention refer to models that combine multimodal information such as text, images, videos, and audio for training, including but not limited to multimodal large language models.

[0046] The “computing power running task” mentioned in this invention refers to a specific workload or job executed on computing power resources that requires a certain amount of computing power support, usually involving complex data processing, numerical calculation, model training or simulation scenarios.

[0047] Please see Figure 1 , Figure 1 This is a flowchart of a method for dynamically measuring computing power based on computing power in a cloud platform for intelligent computing centers, as provided by this invention. Figure 1 As shown, the method includes: Step S1: Obtain the parameter values ​​of the first running data of the first GPU within the first time period, where the first GPU is the GPU running the first computing power running task.

[0048] The aforementioned first time period refers to a time interval during the first GPU's execution of the first computing power task. The aforementioned parameter values ​​are obtained by collecting parameter values ​​from the first GPU within the first time period. It should be noted that when calculating the computing power value of the first computing power task, to accurately measure the computing power value, the frequency of collecting the first running data can be increased as much as possible to improve the accuracy of the calculated computing power value. However, collecting the first running data requires the computing power resources of the intelligent computing center cloud platform. Excessively frequent collection of parameter values ​​from the first running data will consume a significant amount of computing power resources. To achieve accurate collection of the computing power value of the first computing power task without excessive consumption of computing power resources, this invention sets a first time period, collecting the parameter values ​​of the first GPU's first running data only once within the first time period, thereby reducing the consumption of computing power resources on the intelligent computing center cloud platform.

[0049] In this context, the parameter values ​​of the first running data of the first GPU within the first time period are real-time values. This first running data includes GPU-related data such as memory capacity, memory bandwidth, and floating-point utilization. It should be noted that the measurement factors for the computing power of hardware devices can be, for example... Figure 2 As shown, the parameters include GPU resources, CPU resources, storage resources and network resources. In this invention, the computing power is measured only for GPU-related parameters of the intelligent computing center cloud platform.

[0050] The first GPU mentioned above can be a physical GPU or a virtual GPU.

[0051] Step S2: Determine the first task type of the first computing power running task based on the parameter values ​​of the first running data.

[0052] The aforementioned first task type refers to the task type of the first computing power running task, used to characterize the computing power resources occupied by the first computing power running task within a first time period. It should be noted that during the execution of the first computing power running task on the first GPU, the resource usage of the first computing power running task varies. For example, during certain periods, it may require higher floating-point calculation utilization and lower memory capacity, while during other periods, it may require lower floating-point calculation utilization and higher memory capacity and bandwidth. In such cases, the contribution of different resources to the computing power value differs, requiring different weighting coefficients to calculate the computing power value to improve its accuracy.

[0053] The first task type can be pre-set based on experience, such as computationally intensive task type, memory-intensive task type, or large memory capacity task type. The characteristics of computationally intensive task type are high floating-point computing load and low memory bandwidth pressure. It is usually a computing power-intensive task such as scientific simulation (such as molecular dynamics, climate simulation) and deep learning training (such as matrix multiplication, convolution operation). The characteristics of memory-intensive task type are high memory bandwidth requirements and computing units are easy to be idle. The characteristics of large memory capacity task type are that the memory capacity is close to saturation and the bandwidth and computing load are relatively low.

[0054] Step S3: Determine the weighting coefficient corresponding to the first task type.

[0055] The weighting coefficients mentioned above are the weighting coefficients corresponding to the first task type. The computing power value of the first computing power running task in the first time period can be calculated using the weighting coefficients. Note that the weighting coefficients are pre-configured, and the weighting coefficients for different task types can be the same or different.

[0056] Step S4: Based on the weighting coefficient and the parameter values ​​of the first running data, calculate the computing power value of the first computing power running task within the first time period.

[0057] The aforementioned computing power value represents the amount of computing power resources occupied by the first computing power running task in the first time period. This invention enables accurate measurement of computing power for different task types.

[0058] In some implementations, the computing power value of the first computing power operation task within the first time period is calculated based on the weighting coefficient and the parameter value of the first running data. Specifically, the computing power value can be obtained by multiplying the weighting coefficient and the first running data.

[0059] In this invention, step S1 involves obtaining parameter values ​​of the first running data of the first GPU within a first time period, where the first GPU is the GPU running the first computing power task; step S2 involves determining the first task type of the first computing power task based on the parameter values ​​of the first running data; step S3 involves determining the weighting coefficient corresponding to the first task type; and step S4 involves calculating the computing power value of the first computing power task within the first time period based on the weighting coefficient and the parameter values ​​of the first running data. Thus, by determining the first task type of the first computing power task within the first time period, and then determining the weighting coefficient based on the first task type, the weighting coefficient can determine the contribution of the first running data to the computing power value for the first task type. Furthermore, by calculating the computing power value of the first computing power task within the first time period based on the weighting coefficient and the parameter values ​​of the first running data, the computing power resources occupied by the first computing power task within the first time period can be accurately measured based on the parameter values ​​of the first running data of the first computing power task within the first time period, greatly improving the accuracy of computing power measurement for computing power tasks in the intelligent computing center cloud platform.

[0060] In one embodiment, step S2 includes: Step S21: Determine the first task type of the first computing power running task based on the first mapping table and the first running data. The first mapping table includes multiple task types and a running data value range corresponding to each task type. The multiple task types include the first task type, and the parameter values ​​of the first running data are within the running data value range corresponding to the first task type.

[0061] The aforementioned first mapping table is a pre-created mapping table. The first mapping table includes multiple pre-configured task types and the running data value range corresponding to each task type, so that when the parameter value of the first running data of the first GPU in the first time period is obtained, the first task type corresponding to the first computing power running task can be determined according to the running data range in which the parameter value is located.

[0062] In one embodiment, step S3 includes: Step S31: Determine the weighting coefficient corresponding to the first task type based on the second mapping table. The second mapping table includes multiple task types and the weighting coefficient corresponding to each task type.

[0063] The second mapping table mentioned above is a pre-created mapping table. The second mapping table includes multiple pre-configured task types and weighting coefficients corresponding to each task type. In this way, after determining the first task type corresponding to the task to be run by the first computing power, the weighting coefficients can be determined according to the first task type and the second mapping table.

[0064] In some implementations, the first mapping table and the second mapping table can be merged into a single mapping table. This mapping table includes multiple task types, the range of running data values ​​corresponding to each task type, and the weighting coefficient corresponding to each task type. Through this mapping table, the first task type corresponding to the first computing power running task can be determined based on the range of running data where the parameter value is located, and the weighting coefficient can be determined based on the first task type and the second mapping table.

[0065] In one embodiment, the first runtime data includes at least one of floating-point computing utilization, video memory capacity, and video memory bandwidth; The number of weighting coefficients corresponding to the first task type matches the number of parameters included in the first running data.

[0066] In this invention, the first running data includes at least one of floating-point computing utilization, video memory capacity, and video memory bandwidth; the number of weighting coefficients corresponding to the first task type matches the number of parameters included in the first running data. Thus, a weighting coefficient is configured for each parameter in the first running data to adjust for different resource contributions, thereby improving the accuracy of the measured computing power value.

[0067] The first set of running data includes at least one of floating-point computing utilization, video memory capacity, and video memory bandwidth. Floating-point computing utilization can be obtained by calculating the occupancy time of the Streaming Multiprocessor (SM). Video memory capacity can be obtained through preset command codes. Video memory bandwidth can be obtained by acquiring the video memory frequency and bit width, and then calculating it based on the video memory frequency and bit width. For example, video memory bandwidth = video memory frequency × bit width ÷ 8 × 2.

[0068] For example, taking the first running data including floating-point calculation utilization, video memory capacity, and video memory bandwidth as an example, the first mapping table may include: The computationally intensive task type is defined by the following data ranges: floating-point utilization >70%, video memory bandwidth utilization <30%, and video memory capacity utilization <50%. For memory-intensive task types, the corresponding data ranges are: video memory bandwidth utilization >60%, floating-point calculation utilization 30%~50%, and video memory capacity utilization 50%~70%. For tasks with large video memory capacity, the corresponding data range is: video memory usage > 80%, video memory bandwidth usage < 40%, and floating-point calculation utilization < 50%.

[0069] Furthermore, for each parameter included in the first running data, a second mapping table can be set including: For compute-intensive task types, the weighting coefficients are as follows: 0.7 for floating-point computation utilization, 0.1 for video memory bandwidth utilization, and 0.2 for video memory capacity utilization. For memory-intensive task types, the weighting coefficients are as follows: floating-point calculation utilization is 0.1, video memory bandwidth utilization is 0.6, and video memory capacity utilization is 0.3. For tasks with large video memory capacity, the weighting coefficients are as follows: 0.2 for floating-point calculation utilization, 0.5 for video memory bandwidth utilization, and 0.3 for video memory capacity utilization.

[0070] In one embodiment, the first running data includes floating-point calculation utilization, and step S1 includes: Step S11: Obtain the duration percentage and utilization parameter value of each floating-point calculation type among multiple floating-point calculation types within the first time period; Step S12: Calculate the floating-point calculation utilization parameter value based on the time percentage and utilization parameter value of the multiple floating-point calculation types.

[0071] It should be noted that, as Figure 3 As shown, during the operation of the first computing power task, different types of floating-point numbers may be used for calculation, such as 64-bit, 32-bit, 16-bit or mixed precision. The actual floating-point calculation resources of different types of floating-point numbers are different. It is necessary to combine different floating-point calculation types to calculate the floating-point calculation utilization parameter value of the first computing power task in the first time period.

[0072] Specifically, the floating-point computing utilization parameter value is calculated based on the time percentage and utilization parameter values ​​of the multiple floating-point computing types. This can be achieved by weighting the utilization parameter values ​​of the multiple floating-point computing types using the time percentage as a weighting coefficient to obtain the floating-point computing utilization parameter value for the first time period. Specifically, the time percentage and utilization parameter values ​​of the multiple floating-point computing types can be obtained through framework layer monitoring or hardware layer monitoring.

[0073] For example, calculate the percentage of time for each floating-point calculation type within a 15-minute data collection period (the floating-point calculation type can be FP16, FP32, FP8, FP64, etc.). Multiply the parameter value of each floating-point calculation by its respective percentage and then sum them up (e.g., INT8 for H100 is 3958 TFLOPS, FT16 is 1979 TFLOPS, FP32 is 67 TFLOPS, and FP64 is 34 TFLOPS) to obtain an actual floating-point calculation parameter value, which is the floating-point calculation utilization parameter value of the first computing power running task in the first time period.

[0074] In one embodiment, step S4 includes: Step S41: Calculate the relative parameter value, wherein the relative parameter value is the ratio of the parameter value of the first running data to the preset rated value; Step S42: The product of the relative parameter value and the weighting coefficient is set as the computing power value of the first computing power running task in the first time period.

[0075] The aforementioned preset rating values ​​are the rated values ​​of the relevant parameters of the pre-selected accelerator card. The computing power provided by the pre-selected accelerator card can be defined as one unit. In this way, by calculating the relative parameter value, which is the ratio of the parameter value of the first running data to the preset rating value, the product of the relative parameter value and the weighting coefficient is set as the computing power value of the first computing power running task in the first time period, so that the calculated computing power value can be measured by the standard computing power unit.

[0076] The pre-selected accelerator card can be an A100 SXM model accelerator card. The unit computing power is the computing power provided by the accelerator card under full load in FP16 dense computing, which can be expressed as 1DCU=312TFlops*1h*700w, where 1DCU represents 1 unit of computing power.

[0077] Furthermore, after calculating the relative parameter values, the product of the relative parameter values ​​and the weighting coefficients is set as the computing power value of the first computing power operation task in the first time period. This makes the computing power value measured by standard computing power units, which facilitates the subsequent determination of the amount of computing power required by the first computing power operation task.

[0078] In one embodiment, when the first running data includes at least two parameters among floating-point calculation utilization, video memory capacity, and video memory bandwidth, step S4 includes: Step S41': Obtain the preset rated value corresponding to each of the at least two parameters; Step S42': Calculate the relative parameter value of each parameter. The relative parameter value of the first parameter is the ratio of the parameter value of the first parameter to the preset rated value corresponding to the first parameter. The first parameter is any one of the at least two parameters. Step S43': Based on the weighting coefficient, the relative parameter values ​​of each parameter are weighted to obtain the computing power value of the first computing power running task within the first time period.

[0079] In this invention, since the first operating parameters may include at least two parameters, all parameter items included in the first operating parameters need to be combined when calculating the computing power value. Specifically, the relative parameter value of each parameter is calculated, where the relative parameter value of the first parameter is the ratio of the parameter value of the first parameter to the preset rated value corresponding to the first parameter, and the first parameter is any one of the at least two parameters. Based on the weighting coefficient, the relative parameter values ​​of each parameter are weighted to obtain the computing power value of the first computing power operation task within the first time period. In this way, by weighting the relative parameter values ​​of each parameter, the computing power value of the first computing power operation task within the first time period can be obtained by calculating multiple parameters.

[0080] For example, if the first set of running data includes three parameters: floating-point utilization, video memory capacity, and video memory bandwidth, a preset rated value is calculated for each parameter: the preset rated value for floating-point utilization is 'a', the preset rated value for video memory capacity is 'b', and the preset rated value for video memory bandwidth is 'c'. Then, the computing power value is calculated by weighting the parameters according to the weighting coefficients obtained through the second mapping table. If the first task type is a computationally intensive task, then the computing power value is 0.7a + 0.1b + 0.2c; If the first task type is a memory-intensive task, then the computing power value is 0.1a + 0.6b + 0.3c; If the first task type is a large memory capacity task type, then the computing power value is 0.2a+0.5b+0.3c.

[0081] In one embodiment, the method further includes: Step S5: Obtain parameter values ​​of the first running data of the first GPU within multiple consecutive time periods, wherein the multiple time periods are the time periods during which the first GPU executes the first computing power running task, and the multiple time periods include the first time period; Step S6: Based on the parameter values ​​of the first running data of the first GPU in the multiple time periods, calculate the computing power value of the first computing power running task in each time period; Step S7: Set the sum of the computing power values ​​of the multiple time periods as the total computing power value of the first computing power running task.

[0082] It should be noted that the first GPU needs a relatively long time to execute the first computing power task. As mentioned above, in order to improve the accuracy of the measured computing power value as much as possible and to avoid occupying too much computing power resources of the intelligent computing center cloud platform, multiple consecutive time periods are set in this invention. The computing power value of each time period is calculated, and then all computing power values ​​are added together to obtain the total computing power value of the first computing power task.

[0083] In some implementations, the duration of each time segment within a continuous time period can be the same or different. Where the duration of each time segment is the same, the duration of the time segment can be set to 15 minutes, meaning that the parameter value for each parameter is collected every 15 minutes. For example, by collecting the video memory capacity over 15 minutes, the average value is calculated as the parameter value for the video memory capacity during that time segment.

[0084] Please see Figure 4 , Figure 4 This is a structural diagram of a device for dynamically measuring computing power based on computing power operation tasks in an intelligent computing center cloud platform, as provided by the present invention. Figure 4 As shown, the device 400 for dynamically measuring computing power based on computing power to run tasks in the intelligent computing center cloud platform includes: The first acquisition module 401 is used to acquire parameter values ​​of the first running data of the first GPU within a first time period, wherein the first GPU is the GPU running the first computing power running task; The first determining module 402 is used to determine the first task type of the first computing power running task based on the parameter values ​​of the first running data. The second determining module 403 is used to determine the weighting coefficient corresponding to the first task type; The first calculation module 404 is used to calculate the computing power value of the first computing power running task within the first time period based on the weighting coefficient and the parameter value of the first running data.

[0085] In one embodiment, the first determining module 402 includes: The first determining unit is configured to determine the first task type of the first computing power running task based on the first mapping table and the first running data. The first mapping table includes multiple task types and a running data value range corresponding to each task type. The multiple task types include the first task type, and the parameter values ​​of the first running data are within the running data value range corresponding to the first task type.

[0086] In one embodiment, the second determining module 403 includes: The second determining unit is used to determine the weighting coefficient corresponding to the first task type based on the second mapping table, wherein the second mapping table includes multiple task types and the weighting coefficient corresponding to each task type.

[0087] In one embodiment, the first runtime data includes at least one of floating-point computing utilization, video memory capacity, and video memory bandwidth; The number of weighting coefficients corresponding to the first task type matches the number of parameters included in the first running data.

[0088] In one embodiment, the first runtime data includes floating-point calculation utilization, and the first acquisition module 401 includes: The first acquisition unit is used to acquire the duration percentage and utilization parameter value of each floating-point calculation type among multiple floating-point calculation types in the first time period; The first calculation unit is used to calculate the floating-point calculation utilization parameter value based on the time proportion and utilization parameter value of the multiple floating-point calculation types.

[0089] In one embodiment, the first computing module 404 includes: The second calculation unit is used to calculate the relative parameter value, wherein the relative parameter value is the ratio of the parameter value of the first running data to the preset rated value; The third calculation unit is used to set the product of the relative parameter value and the weighting coefficient as the computing power value of the first computing power running task in the first time period.

[0090] In one embodiment, when the first operating data includes at least two parameters among floating-point calculation utilization, video memory capacity, and video memory bandwidth, the first calculation module 404 includes: The second acquisition unit is used to acquire a preset rated value corresponding to each of the at least two parameters; The fourth calculation unit is used to calculate the relative parameter value of each parameter, wherein the relative parameter value of the first parameter is the ratio of the parameter value of the first parameter to the preset rated value corresponding to the first parameter, and the first parameter is any one of the at least two parameters; The fifth calculation unit is used to weight the relative parameter values ​​of each parameter based on the weighting coefficient to obtain the computing power value of the first computing power running task within the first time period.

[0091] In one embodiment, the device 400 for dynamically measuring computing power based on computing power operation tasks in the intelligent computing center cloud platform further includes: The second acquisition module is used to acquire parameter values ​​of the first running data of the first GPU within multiple consecutive time periods, wherein the multiple time periods are the time periods during which the first GPU executes the first computing power running task, and the multiple time periods include the first time period; The second calculation module is used to calculate the computing power value of the first computing power running task in each time period based on the parameter values ​​of the first running data of the first GPU in the multiple time periods. The third calculation module is used to set the sum of the computing power values ​​of the multiple time periods as the total computing power value of the first computing power running task.

[0092] The device for dynamically measuring computing power based on computing power running tasks in the intelligent computing center cloud platform provided by the present invention is capable of realizing the various processes of the above-mentioned method for dynamically measuring computing power based on computing power running tasks in the intelligent computing center cloud platform. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0093] It should be noted that the device for dynamically measuring computing power based on computing power to run tasks in the intelligent computing center cloud platform of this invention can be a device, or it can be a component, integrated circuit, or chip in an electronic device.

[0094] The present invention also provides an electronic device, see below. Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 501, a processor 502, and a program or instructions stored in the memory 501 that run on the processor 502. When the program or instructions are executed by the processor 502, they can achieve the following: Figure 1 The corresponding steps in the method embodiment of the intelligent computing center cloud platform for dynamically measuring computing power based on computing power to run tasks and achieving the same beneficial effect will not be elaborated here.

[0095] The processor 502 can be a CPU, ASIC, FPGA, or GPU.

[0096] Those skilled in the art will understand that all or part of the steps of the above-described method embodiment for dynamically measuring computing power based on computing power to run tasks in an intelligent computing center cloud platform can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0097] The present invention also provides a readable storage medium on which a computer program is stored, and which, when executed by a processor, can perform the above-described functions. Figure 1The corresponding steps in the method embodiment for dynamically measuring computing power based on computing power to run tasks in the intelligent computing center cloud platform can achieve the same technical effect, and will not be described again here to avoid repetition. The storage medium mentioned includes, for example, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] The present invention also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described... Figure 1 The corresponding intelligent computing center cloud platform's method for dynamically measuring computing power based on computing power to run tasks involves various processes that achieve the same technical effect. To avoid repetition, these processes will not be elaborated here.

[0099] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: A alone, B alone, C alone, both A and B present, both B and C present, both A and C present, and A, B, and C present.

[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or second terminal device, etc.) to execute the methods of the various embodiments of this application.

[0102] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for dynamically measuring computing power based on computing power running tasks by an intelligent computing center cloud platform, characterized in that, The method comprises: Step S1, obtaining a parameter value of first running data of a first GPU in a first time period, the first GPU being a GPU running a first computing power running task; Step S2, determining a first task type of the first computing power running task based on the parameter value of the first running data; Step S3, determining a weighting coefficient corresponding to the first task type; Step S4, calculating a computing power value of the first computing power running task in the first time period based on the weighting coefficient and the parameter value of the first running data.

2. The method of claim 1, wherein, The step S2 comprises: Step S21, determining a first task type of the first computing power running task based on a first mapping table and the first running data, the first mapping table comprising a plurality of task types and a running data numerical interval corresponding to each task type, the plurality of task types comprising the first task type, and the parameter value of the first running data being within the running data numerical interval corresponding to the first task type.

3. The method of claim 1, wherein, The step S3 comprises: Step S31, determining a weighting coefficient corresponding to the first task type based on a second mapping table, the second mapping table comprising a plurality of task types and a weighting coefficient corresponding to each task type.

4. The method of claim 3, wherein, The first running data comprises at least one of a floating point calculation utilization rate, a display memory capacity, and a display memory bandwidth; The number of weighting coefficients corresponding to the first task type matches the number of parameters included in the first running data.

5. The method of any one of claims 1 to 4, wherein, The first running data comprises a floating point calculation utilization rate, and the step S1 comprises: Step S11, obtaining a time length proportion and a utilization rate parameter value of each floating point calculation type in a plurality of floating point calculation types in the first time period; Step S12, calculating the floating point calculation utilization rate parameter value based on the time length proportion and the utilization rate parameter value of the plurality of floating point calculation types.

6. The method of any one of claims 1 to 4, wherein, The step S4 comprises: Step S41, calculating a relative parameter value, the relative parameter value being a ratio of the parameter value of the first running data to a preset rated value; Step S42, setting a product of the relative parameter value and the weighting coefficient as the computing power value of the first computing power running task in the first time period.

7. The method of any one of claims 1 to 4, wherein, In the case where the first running data comprises at least two parameters of a floating point calculation utilization rate, a display memory capacity, and a display memory bandwidth, the step S4 comprises: Step S41', obtaining a preset rated value corresponding to each parameter of the at least two parameters; Step S42', calculating a relative parameter value of each parameter, the relative parameter value of a first parameter being a ratio of a parameter value of the first parameter to a preset rated value corresponding to the first parameter, the first parameter being any parameter of the at least two parameters; Step S43', weighting the relative parameter values of the at least two parameters based on the weighting coefficient to obtain the computing power value of the first computing power running task in the first time period.

8. The method of any one of claims 1 to 4, wherein, The method further comprises: Step S5, obtaining parameter values of first running data of the first GPU in a plurality of time periods, the plurality of time periods being time periods in a process in which the first GPU executes the first computing power running task, and the plurality of time periods comprising the first time period; Step S6, calculating, based on the parameter values of the first running data of the first GPU in the plurality of time periods, a computing power value of the first computing power running task in each time period; Step S7, setting a sum of the computing power values of the plurality of time periods as a total computing power value of the first computing power running task.

9. An apparatus for dynamically metering computing power based on computing power running tasks of an intelligent computing center cloud platform, characterized in that, Comprising: A first obtaining module, configured to obtain parameter values of first running data of a first GPU in a first time period, the first GPU being a GPU running a first computing power running task; A first determining module, configured to determine a first task type of the first computing power running task based on the parameter values of the first running data; A second determining module, configured to determine a weighting coefficient corresponding to the first task type; A first calculating module, configured to calculate, based on the weighting coefficient and the parameter values of the first running data, a computing power value of the first computing power running task in the first time period.

10. An electronic device, comprising: Comprising: A processor, a memory, and a program stored on the memory and executable on the processor, the program, when executed by the processor, implementing the steps of the method for dynamically metering computing power of a computing power running task of an intelligent computing center cloud platform according to any one of claims 1 to 8.

11. A computer readable storage medium, characterized in that, A computer program is stored on the computer readable storage medium, and the computer program, when executed by the processor, implements the steps of the method for dynamically metering computing power of a computing power running task of an intelligent computing center cloud platform according to any one of claims 1 to 8.

12. A computer program product, characterised in that, Comprising computer instructions, the computer instructions, when executed by the processor, implementing the steps of the method for dynamically metering computing power of a computing power running task of an intelligent computing center cloud platform according to any one of claims 1 to 8.

Images