GPU资源调度方法

The GPU resource scheduling method, which combines parallel matching and dynamic priority adjustment, solves the problem of unbalanced GPU resource allocation, improves resource utilization and computing performance, and reduces power consumption.

CN116450330BActive Publication Date: 2026-07-17METAX INTEGRATED CIRCUITS (SHANGHAI) CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
METAX INTEGRATED CIRCUITS (SHANGHAI) CO LTD
Filing Date
2022-01-10
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve balanced resource scheduling during GPU resource allocation, resulting in low resource utilization and decreased computing performance, especially inefficient and power-consuming tasks involving complex computing.

Method used

By acquiring the resource requirements of task groups and the remaining resource information of execution modules, and using parallel matching and dynamic priority adjustment methods, combined with the priority and resource type of task groups, the allocation order and resource allocation strategy of task groups are dynamically adjusted to ensure balanced resource allocation.

Benefits of technology

It achieves balanced allocation of GPU resources, improves resource utilization, reduces power consumption, and enhances task allocation and processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116450330B_ABST
    Figure CN116450330B_ABST
Patent Text Reader

Abstract

本发明涉及一种GPU资源调度方法,包括:步骤H1、获取待分发任务组,读取待分配任务数量,所述待分配任务为需要平衡分配至执行模块的执行单元中的任务,所述执行模块包括Q个执行单元;步骤H2、基于任务数切分表确定所述待分配任务数量对应的初始分配组合序列,所述任务数切分表用于存储分配任务数量与初始分配组合序列的映射关系;步骤H3、获取所述执行模块中的预设指针信息,基于所述指针信息确定循环移位数Su,将所述初始分配组合序列向预设的预设方向循环移动Su位,得到目标分配组合信息;步骤H4、基于所述待分发任务组对应的目标分配组合信息匹配所述执行模块的GPU资源。本发明提高了GPU资源利用率,减小了功耗。
Need to check novelty before this filing date? Find Prior Art