Runtime Dispatching Among Heterogeneous Processors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems with heterogeneous processing engines, such as CPUs and GPUs, typically assign computing operations at design time rather than runtime, leading to inefficiencies in power consumption, speed, and thermal management, as they do not adapt to actual usage conditions.
Innovation Solution
A runtime dispatching system that includes a scheduler module to dynamically distribute computing tasks among heterogeneous processors based on criteria like power conservation, thermal balancing, and speed optimization, allowing for adaptive task assignment at runtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If computing operations are assigned at design time to processing engines, then the system structure is simple and easy to implement, but the system cannot adapt to actual usage conditions leading to inefficiencies in power consumption, speed, and thermal management
Solution Approach 1:
The patent implements dynamic task assignment by introducing a runtime scheduler that continuously monitors system state (power, thermal, performance) and redistributes computing tasks among heterogeneous processors based on current conditions. This transforms the static design-time assignment into a dynamic runtime decision-making process, allowing the system to adapt to changing usage conditions while maintaining manageable complexity through automated scheduling algorithms.
2Speed
If computing tasks are executed on GPU, then processing speed is improved, but power consumption increases
Solution Approach 1:
The runtime scheduler dynamically monitors power consumption metrics and processing speed requirements, then intelligently assigns tasks to GPU only when both speed improvement is needed and power budget allows. When power consumption becomes excessive, the scheduler automatically migrates tasks to more power-efficient processors, creating a dynamic balance between speed and power usage based on real-time system state.
Solution Approach 2:
The system changes the operational parameters of different processors based on system state. The scheduler adjusts task assignment decisions by evaluating multiple parameters including power consumption thresholds, thermal conditions, and performance requirements, then selects the optimal processor configuration to meet current demands while minimizing power usage.
3Use of energy by moving object
If computing tasks are executed on CPU, then power consumption is reduced, but processing speed decreases
Solution Approach 1:
The runtime scheduler continuously evaluates whether tasks should remain on power-efficient CPUs or be migrated to faster GPUs based on changing system conditions. When performance requirements increase or thermal headroom allows, the scheduler dynamically transfers tasks to GPU for accelerated processing, ensuring the system maintains optimal speed while minimizing power consumption under varying load conditions.
4Temperature
If heterogeneous processors are utilized without runtime dispatching, then hardware costs are avoided, but thermal management becomes inefficient
Solution Approach 1:
The patent introduces a runtime scheduler as an intermediary software layer that manages thermal balance across heterogeneous processors. This scheduler monitors thermal conditions and task distribution, then makes intelligent decisions about task migration to prevent thermal hotspots. The intermediary handles the complexity of thermal management, allowing the hardware to remain simple while achieving efficient thermal balance through software-based dispatching.
Data Source
AI summary
Systems, apparatus, articles, and methods are described including operations for runtime dispatching among a heterogeneous group of processors.