A GPU internal energy consumption optimization method based on balanced task scheduling

Abstract

The invention belongs to the technical field of task balance scheduling based on equipment state and task characteristics, and discloses a GPU internal energy consumption optimization method based on task balance scheduling, using the balance influence factor CB of the program to be executed and the SM use frequency information PHB to propose a balance CB‑HRV task scheduling algorithm based on influence factor and frequency information used by SM; CB‑HRV task scheduling algorithm assigns a group of tasks to the corresponding SM according to the balanced impact factor of the task according to the task execution time; the policy scheduler balances the tasks At the same time, use the frequency information of SM to improve the utilization efficiency of equipment resources; through the integrated application of the task information to be executed and the resource information, the task scheduling method of energy consumption optimization is realized. The invention realizes the dynamic balance scheduling among multiple SMs, and achieves the goal of optimizing the energy consumption of the GPU.

Description

technical field [0001] The invention belongs to the technical field of task balance scheduling based on equipment status and task characteristics, and in particular relates to a GPU internal energy consumption optimization method based on task balance scheduling. Background technique [0002] At present, the commonly used existing technology in the industry is as follows: Graphics processing unit (English: Graphics Processing Unit, abbreviation: GPU), also known as display core, visual processor, display chip, is a special And some microprocessors for image computing work on mobile devices (such as tablet computers, smart phones, etc.). The purpose is to convert and drive the display information required by the computer system, and provide line scan signals to the display to control the correct display of the display. It is an important component for connecting the display and the main board of the personal computer, and is also one of the important devices for "human-machin...

AI Technical Summary

Problems solved by technology

[0026] (1) The loss caused by the migration of tasks on each SM is not considered. For example, the scheduling algorithm based on device release (device history-based algorithm) considers the priority of using the released SM, but does not consider the migration of tasks on the SM; it is very likely Due to the repeated migration of multiple tasks between SMs, a large amount of energy consumption is generated, which weakens the energy consumption performance of the GPU

[0027] (2) There is no comprehensive consideration of equipment resources and tasks performed; although the SM utilization rate is considered in the algorithm strategy based on SM historical performance (Performance History Based), it does not combine the migration characteristics of tasks and cannot Balanced use of SM results in inefficient task execution

[0029] (1) The current task scheduling algorithm does not have a high utilization rate of the computing unit (stream processor SM) of the GPU, and cannot comprehensively consider the utilization efficiency of SM and the impact of task characteristics on energy consumption

[0030] (2) Migration phenomenon in task scheduling is a common problem. Migration of tasks in the stream processor will cause energy loss. The current scheduling strategy cannot effectively alleviate the phenomenon of task migration.

[0031] (3) At present, there is no effective energy consumption optimization task scheduling algorithm that can solve the above two problems, and algorithm design is needed

Images