Task Segmentation for Multi-Core Neural Network Processors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional task scheduling methods for multi-core neural network processors result in low service quality, low kernel utilization, unbalanced load, and high energy consumption due to inefficient task segmentation and scheduling.
Innovation Solution
A task segmentation device and method that segments tasks into subtasks based on various granularities such as sample data, layer types, interlayer, and intra-layer structures of neural networks, combined with a task scheduling device that monitors core working states and allocates tasks using heuristic algorithms to optimize scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional task scheduling method is applied to multi-core neural network processor, then task scheduling can be performed, but service quality is low and kernel utilization rate is low
Solution Approach 1:
The patent segments neural network tasks into multiple subtasks based on different granularities (sample-level, layer-level, computation-type-level) and schedules them on multi-core processors. This segmentation enables finer-grained task distribution, improving both service quality through better task management and kernel utilization by matching subtasks to available cores more effectively
Solution Approach 2:
The patent implements dynamic task scheduling that adapts to real-time changes in core working states and resource availability. The scheduling algorithm dynamically adjusts task allocation based on monitored core states, improving both service quality through responsive task management and kernel utilization by optimizing task-core matching in real-time
2Productivity
If traditional task scheduling method is applied to multi-core neural network processor, then task scheduling can be performed, but load distribution is unbalanced
Solution Approach 1:
By segmenting tasks into multiple granular subtasks, the system can distribute work more evenly across multiple cores. The segmentation allows the scheduling algorithm to balance load by assigning smaller, more manageable subtasks to cores based on their current states, preventing load imbalance while maintaining scheduling efficiency
Solution Approach 2:
The patent monitors core working states in real-time and uses this feedback to adjust task scheduling decisions. This feedback mechanism ensures balanced load distribution by adapting to changing core conditions, while maintaining high scheduling efficiency through continuous optimization based on actual core performance
3Productivity
If traditional task scheduling method is applied to multi-core neural network processor, then task scheduling can be performed, but energy consumption is high
Solution Approach 1:
The dynamic scheduling algorithm adjusts task allocation based on real-time core states and energy conditions. By monitoring and adapting to changing system states, the system optimizes energy consumption while maintaining processing throughput, avoiding unnecessary energy expenditure on tasks that can be deferred or redistributed
Solution Approach 2:
The patent changes scheduling parameters and task allocation strategies based on system conditions. By adjusting scheduling policies dynamically according to core states and resource availability, the system optimizes the balance between processing throughput and energy consumption, reducing overall energy expenditure while maintaining productivity
Data Source
AI summary
The disclosure provides a task segmentation device and method, a task processing device and method, a multi-core processor. The task segmentation device includes a granularity task segmentation unit configured to segment a task by adopting at least one granularity to form subtasks, and a task segmentation granularity selection unit configured to select the granularity to be adopted.