Bi-Directional Computing Unit Frequency Control With Thread Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing system on chip (SOC) implementations lack granular control over computing unit frequencies, leading to over- or under-subscription due to cross-thread contamination, resulting in power inefficiencies and reduced battery life.
Innovation Solution
Implementing a bi-directional closed-loop feedback control between computing units and software, using performance hints and power and performance statistics to dynamically adjust frequencies based on software threads, with anti-abuse logic to prevent non-compliant instructions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If hardware autonomous frequency selection techniques are used to accommodate quick execution variations, then responsiveness is improved, but power efficiency deteriorates due to over- or under-subscription
Solution Approach 1:
The patent implements a closed-loop feedback control system where the operating system monitors computing unit performance and dynamically adjusts frequency settings. The OS receives feedback from hardware performance counters and autonomously modifies frequency requests to computing units, creating a continuous optimization loop that balances responsiveness with power efficiency.
Solution Approach 2:
The system transitions from static frequency selection to dynamic frequency adjustment. The operating system continuously adapts frequency settings based on real-time workload characteristics and performance metrics, allowing the system to optimize power consumption while maintaining responsiveness to execution variations.
2Use of energy by moving object
If frequency control is implemented at SOC level with hardware techniques, then power efficiency is improved, but granular control over individual computing units is lost
Solution Approach 1:
The patent divides frequency control authority into discrete segments assigned to individual computing units. Each computing unit can have independent frequency adjustments made by the operating system based on its specific workload characteristics, enabling granular control while maintaining system-wide power optimization.
Solution Approach 2:
The operating system acts as an intermediary layer between hardware frequency control mechanisms and application workloads. It translates high-level performance requirements into specific frequency adjustments for individual computing units, providing both granular control and power efficiency through intelligent mediation.
3Device complexity
If cross-thread contamination occurs in frequency subscription, then device complexity is reduced, but measurement precision of performance metrics deteriorates
Solution Approach 1:
The patent segments performance monitoring and frequency control to thread-level granularity. Separate performance counters and control mechanisms are maintained for each thread, eliminating cross-thread contamination while preserving implementation feasibility through modular organization of monitoring resources.
Data Source
AI summary
Methods and apparatus for bi-directional control of computing unit frequency are disclosed. An example apparatus to control a frequency of a computing unit includes instructions, memory in the apparatus, and processor circuitry. The processor circuitry is to determine a performance hint from a first register, the performance hint corresponding to a requested performance of the computing unit for executing a thread associated with software, determine power and performance (PnP) statistics pertaining to the thread from a second register, control the frequency of the computing unit based on the performance hint and the PnP statistics, and provide a pressure of the computing unit to the software.