CN-122019130-A - GPU application system, application method thereof, electronic terminal and storage medium

CN122019130ACN 122019130 ACN122019130 ACN 122019130ACN-122019130-A

Abstract

The application provides a GPU application system and an application method thereof, an electronic terminal and a storage medium, wherein the GPU application system comprises a physical machine and an acceleration card, the physical machine is connected with the acceleration card, a GPU module at least comprises a first GPU resource pool deployed on the physical machine and a second GPU resource pool deployed on the acceleration card, the first GPU resource pool comprises at least one container, the second GPU resource pool comprises at least one business process, and a regulation and control module is used for determining a matched target acceleration card according to configuration parameters of a task to be processed in a target container so that the target acceleration card processes the task to be processed. According to the application, the GPU module is deployed on the physical machine and the accelerator card, so that a container in a first GPU resource pool deployed on the physical machine can occupy CPU and memory resources on the physical machine, and meanwhile, a business process in a second GPU resource pool deployed on the accelerator card can call AI resources on the accelerator card, thereby effectively utilizing AI resources, memory resources and CPU computing power on the accelerator card.

Inventors

HU DIANRONG
DANG JIANFENG
SHEN JIA
DU WEI
NIU JIAN

Assignees

浙江大华技术股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251223

Claims (12)

1. A GPU application system, the GPU application system comprising: The system comprises a physical machine and an acceleration card, wherein the physical machine is connected with the acceleration card; The system comprises a physical machine, a GPU module, a service platform and a service platform, wherein the physical machine comprises at least a first GPU resource pool deployed on the physical machine and a second GPU resource pool deployed on the accelerator card; And the regulation and control module is connected with the physical machine and the acceleration card and is used for determining a matched target acceleration card according to the configuration parameters of the task to be processed in the target container so as to enable the target acceleration card to process the task to be processed.
2. The GPU application of claim 1, wherein the regulation module is further configured to control sending the task to be processed and program information in the target container to the matched target accelerator card, so that the target accelerator card establishes the business process on the accelerator card based on the received task to be processed and the program information, and processes the task to be processed in the business process based on the program information.
3. The GPU application of claim 1, wherein the configuration parameters of the task to be processed comprise a computational effort demand, the regulation module is further configured to compare the computational effort demand of the task to be processed in the target container with a remaining resource and performance of the accelerator card, and take the accelerator card with the remaining resource greater than the computational effort demand as a target accelerator card.
4. A GPU application according to claim 3, wherein the configuration parameters of the task to be processed further comprise task types, and the regulation module is further configured to, in response to at least two accelerator cards having a residual resource greater than the computational power requirement of the task to be processed, use the accelerator card having any task type inconsistent with the task type of the task to be processed as the target accelerator card matched with the target container.
5. The GPU application of claim 3 or 4, wherein if the accelerator cards that respond to the remaining resources being greater than the computational power requirements of the task to be processed are at least two, the regulator module is configured to take the accelerator card corresponding to the least of the remaining resources as the target accelerator card corresponding to the target container.
6. The GPU application system of claim 1, wherein the regulation module is further configured to determine that each accelerator card is connected to the same physical machine through different preset connection manners, where the preset connection manners include PCIE channels or Socket channels, and/or determine that server types corresponding to each service process are different when at least two service processes are included in the target accelerator card.
7. The GPU application of claim 1, wherein the control module is configured to control the accelerator card to save the task to be processed and a processing result corresponding to the task to be processed in response to the task to be processed being completed based on the program information in the business process.
8. An application method of a GPU application system, wherein the GPU application system is a GPU application system according to any one of claims 1 to 7, the application method comprising: the physical machine determines program information for processing the task to be processed based on configuration parameters of the task to be processed in the target container; Packaging the task to be processed and the program information and sending the packaged task to be processed and the program information to a matched acceleration card, so that the acceleration card processes the task to be processed based on the program information and sends a processing result of the task to be processed to the physical machine; And the physical machine receives the processing result of the task to be processed.
9. An application method of a GPU application system, wherein the GPU application system is a GPU application system according to any one of claims 1 to 7, the application method comprising: The acceleration card builds a business process based on the received task to be processed and the program information; Processing the task to be processed based on the program information in the business process to obtain a processing result of the task to be processed; And sending the physical result of the task to be processed to a physical machine.
10. The method of claim 9, wherein, The processing the task to be processed based on the program information in the business process to obtain a processing result of the task to be processed, including: And responding to the acceleration card to receive the packed task to be processed and the program information, stopping running and deleting the history program corresponding to the business process, and starting the program information to process the task to be processed.
11. An electronic terminal comprising a memory and a processor coupled to each other, the processor configured to execute program instructions stored in the memory, the processor configured to execute program data to implement steps in a method for applying a GPU application as claimed in any of claims 8-10.
12. A computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and when executed by a processor, the computer program implements the steps in the method for applying the GPU application system according to any of claims 8-10.

Description

GPU application system, application method thereof, electronic terminal and storage medium Technical Field The present application relates to the field of computer technologies, and in particular, to a GPU application system, an application method thereof, an electronic terminal, and a storage medium. Background In the artificial intelligence era, graphics processing units (Graphics Processing Unit, GPUs) are becoming increasingly popular. The graphic processing unit is a microprocessor that performs image operation work on personal computers, workstations, game machines, and some mobile devices such as tablet computers, smart phones, etc. Although the GPU service node for deploying the GPU has better processing performance in the image processing field, in view of the high use cost of the GPU service node, more GPU service nodes are not deployed generally, and the requirement of large-scale data calculation on the computing performance cannot be met by utilizing the CPU (Central Processing Unit, processor) service node for computing. Disclosure of Invention The application mainly solves the technical problem of providing a GPU application system, an application method thereof, an electronic terminal and a storage medium, and solves the problem that the GPU application system in the prior art cannot effectively utilize resources on an acceleration card. In order to solve the technical problems, the first technical scheme adopted by the application is to provide a GPU application system, wherein the GPU application system comprises: the physical machine is connected with the acceleration card; The system comprises a GPU module, a service processing module, a processing module and a control module, wherein the GPU module at least comprises a first GPU resource pool deployed on a physical machine and a second GPU resource pool deployed on an acceleration card; And the regulation and control module is connected with the physical machine and the acceleration card and is used for determining a matched target acceleration card according to the configuration parameters of the task to be processed in the target container so as to enable the target acceleration card to process the task to be processed. The regulation and control module is also used for controlling the task to be processed and the program information in the target container to be sent to the matched target acceleration card, so that the target acceleration card establishes a business process on the acceleration card based on the received task to be processed and the program information, and the task to be processed is processed based on the program information in the business process. The configuration parameters of the task to be processed comprise calculation force requirements, the regulation and control module is further used for comparing the calculation force requirements of the task to be processed in the target container with the residual resources and performance of the acceleration card, and the acceleration card with the residual resources larger than the calculation force requirements is used as the target acceleration card. The configuration parameters of the task to be processed further comprise task types, and the regulation and control module is further used for taking the acceleration card with any task type inconsistent with the task type of the task to be processed as a target acceleration card matched with the target container when the residual resources of at least two acceleration cards are larger than the calculation power requirement of the task to be processed. And the regulation and control module is used for taking the acceleration card corresponding to the least residual resource as the target acceleration card corresponding to the target container, wherein at least two acceleration cards responding to the calculation power requirement of the residual resource larger than the task to be processed are arranged. The control module is further configured to determine that each acceleration card is connected with the physical machine through different preset connection modes in response to connection of at least two acceleration cards with the same physical machine, where the preset connection modes include a PCIE channel Socket channel. And the regulation and control module is also used for determining that the server types corresponding to the business processes are different when at least two business processes are contained in the target accelerator card. When the task to be processed is completed based on the program information in the service process, the regulation and control module is used for controlling the acceleration card to store the task to be processed and the processing result corresponding to the task to be processed. In order to solve the technical problems, a second technical scheme adopted by the application is to provide an application method of a GPU application system, wherein the GPU application system