Multicore Power Density Multiplier Timing for Thermal Power Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current thermal management techniques for multicore processors are inefficient due to the assumption of equal power distribution among thermal entities, leading to inadequate power allocation from inactive to active entities, resulting in thermal excursions and reduced performance.
Innovation Solution
Implementing a time-based moving average method to delay the application of power density multipliers (PDMs), allowing inactive thermal entities to act as heat sinks for active ones, thereby optimizing power allocation while adhering to thermal constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If equal power distribution is assumed among all thermal entities, then power allocation is simplified, but power allocation efficiency deteriorates because inactive entities cannot contribute to active entities
Solution Approach 1:
The patent implements dynamic power allocation where power density multipliers are adjusted based on real-time thermal states of processor entities. Instead of static equal distribution, the system continuously monitors which entities are active or inactive and dynamically redistributes power budgets, allowing inactive entities to contribute their unused power to active entities through time-based moving average calculations.
Solution Approach 2:
The system changes the power distribution parameter from a fixed equal value to a variable value based on thermal entity states. Power density multipliers are modified according to the moving average of inactive entities, transforming the power allocation from a constant parameter to a dynamic parameter that adapts to changing thermal conditions.
2Productivity
If power is immediately reallocated from inactive to active thermal entities, then power utilization is improved, but thermal constraints are violated due to insufficient cooling time
Solution Approach 1:
The system performs preliminary cooling by identifying inactive thermal entities before reallocating their power budgets to active entities. The time-based moving average mechanism ensures that sufficient time elapses for thermal cooling to occur before power reallocation, preparing the thermal state in advance to prevent constraint violations.
Solution Approach 2:
The patent introduces a time-based moving average mechanism as an intermediary between power consumption and power reallocation. This intermediary smooths out immediate power transitions by considering the thermal inertia and cooling time required, acting as a buffer that prevents direct, abrupt power reallocation that would violate thermal constraints.
3Productivity
If higher power density multipliers are applied to active thermal entities, then performance is boosted, but thermal excursions occur due to rapid heating
Solution Approach 1:
The system applies power density multipliers periodically rather than continuously, using time-based moving averages to determine when to apply or adjust multipliers. This periodic application allows thermal entities to cool between power increases, preventing continuous thermal excursions while still achieving performance boosts during authorized intervals.
Solution Approach 2:
The patent applies beforehand cushioning by using the time-based moving average to anticipate thermal buildup before it occurs. The system cushions against thermal excursions by limiting the rate at which power density multipliers are applied, ensuring that thermal constraints are not exceeded even as performance is boosted.
4Reliability
If thermal-aware techniques are implemented at architecture level, then thermal management is improved, but device complexity increases
Solution Approach 1:
The patent implements self-service thermal management where the system automatically monitors its own thermal states and adjusts power allocation without external intervention. The processor entities themselves provide thermal state information, and the system uses this information to automatically adjust power density multipliers, reducing the need for complex external thermal management infrastructure.
Solution Approach 2:
The thermal management mechanism serves multiple functions: it monitors thermal states, calculates moving averages, determines power reallocation opportunities, and adjusts power density multipliers. This universal approach consolidates what could be separate complex subsystems into a unified power management framework that handles multiple thermal management tasks.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances performance by efficiently transferring power from inactive to active thermal entities, reducing thermal excursions and maximizing single-thread performance within thermal design limits.
Implementation Method 1
allowing inactive thermal entities to act as heat sinks for active ones, thereby optimizing power allocation while adhering to thermal constraints
Data Source
AI summary
A method and an apparatus are described that delay application of a higher order Power Density Multiplier (PDM) using a time based moving average of a number of active cores in a multicore system. A PDM is applied to a thermal design power budget of a thermal entity and performance of the thermal entity is increased by transferring available power from a thermal entity not in an active state to a thermal entity in an active state. Sufficient time is allowed for the cooling effect of reduced active cores, to influence the active core that receives the extra power (a higher PDM). Similarly delaying application of a lower PDM with the same moving average, but a different threshold, allows a core to retain a higher power allocation until the more active neighbor core(s) cause it to heat up, thereby boosting core performance.