Dynamic Throttling for Power and Thermal Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In scenarios where a processor is extremely power and/or thermal constrained, reducing frequency and voltage may not be sufficient to bring the system-on-a-chip (SoC) within programmed power limits, especially during high concurrency and skin temperature management scenarios.
Innovation Solution
The implementation of advanced throttling actions, including dynamic simultaneous multi-threading (SMT) scheduling and core isolation, to manage power and thermal limits. This involves coordinating IP states, reducing I/O device activity, and using thread runtime telemetry to dynamically park SMT core siblings, thereby optimizing power and thermal management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If frequency and voltage are reduced to manage power consumption, then power consumption decreases, but performance deteriorates and may not be sufficient to meet power limits in extreme scenarios
Solution Approach 1:
The patent segments the processor into multiple independent IP blocks (core, IPU, graphics, media, AI/ML accelerators, VPU) that can be individually controlled and throttled. This allows selective power management of specific high-consumption blocks rather than uniformly reducing all frequencies and voltages, thereby achieving better power control with less performance penalty.
Solution Approach 2:
The system dynamically adjusts the operational state of different IP blocks based on real-time power and thermal conditions. The power management unit continuously monitors power consumption and thermal states, then dynamically enables or disables specific IP blocks as needed, allowing adaptive power management that maintains performance when possible while meeting power limits when necessary.
2Use of energy by moving object
If aggressive throttling actions are taken to meet power limits, then power and thermal constraints are managed, but system performance and responsiveness deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the power management unit continuously monitors power consumption, thermal state, and operational conditions of various IP blocks. Based on this real-time feedback, the system dynamically adjusts throttling actions to maintain power within limits while minimizing impact on system responsiveness. The feedback loop allows the system to react to changing conditions and adjust power management strategies accordingly.
3Productivity
If multiple IP blocks operate concurrently to maintain performance, then productivity increases, but power consumption and thermal generation exceed programmed limits
Solution Approach 1:
The patent applies local quality control by allowing different IP blocks to operate at different performance levels simultaneously. High-performance blocks can continue operating at full capacity while power-hungry blocks are throttled or disabled, creating a non-uniform operational state across the processor that optimizes overall power efficiency while maintaining necessary processing throughput.
Data Source
AI summary
Examples of throttling and unthrottling for power and/or thermal management of a computing device are described. In some examples, a power management unit coupled to the processor core is to determine when one or more of power and thermal operating characteristics of a system, of which the apparatus is a part of, are outside of a system constraint (e.g., a desired power or thermal level), and place one or more of devices coupled to the power management unit into a lower power level to meet the system constraint. In some examples, the one or more devices are input/output devices.