ML-Based GPU Assignment for Power and Thermal Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems inefficiently assign processing tasks between integrated GPUs (iGPUs) and discrete GPUs (dGPUs) based on pre-defined rules, leading to excessive power consumption and heat generation, particularly when non-processing heavy features of applications are executed on dGPUs, which negatively impacts user experience and hardware longevity.
Innovation Solution
A processor uses a machine learning model trained on usage data and contextual information to dynamically determine whether to execute applications on iGPUs or dGPUs, optimizing performance metrics and user experience by predicting power consumption, temperature, and efficiency based on historical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If applications are executed on discrete GPUs (dGPUs), then processing capability is improved, but power consumption increases
Solution Approach 1:
The patent implements dynamic processing unit assignment that adjusts which processing unit (iGPU or dGPU) executes applications based on real-time conditions including user experience metrics, system performance metrics, and workload characteristics. This dynamic switching resolves the contradiction by selecting dGPU only when its superior processing capability is actually needed and justified by predicted outcomes, rather than statically assigning all applications to dGPU.
Solution Approach 2:
The system changes the parameter of processing unit selection based on multiple input factors including application type, current system state, and predicted user experience metrics. By varying this selection parameter dynamically rather than fixed, the system achieves high processing capability only when necessary while reducing power consumption during lighter or less critical tasks.
2Power
If applications are executed on discrete GPUs (dGPUs), then processing capability is improved, but temperature increases
Solution Approach 1:
The dynamic assignment system monitors and responds to thermal conditions by adjusting processing unit selection. When dGPU execution is predicted to generate excessive heat or when thermal thresholds are approached, the system dynamically switches to iGPU execution, thereby maintaining processing capability when safe while preventing overheating.
Solution Approach 2:
The system uses predicted user experience metrics and system performance data to identify when dGPU execution would create harmful thermal effects, then alternatively assigns those workloads to iGPU. This converts the potential harm of thermal management constraints into a benefit by optimizing the overall system operational sustainability and hardware longevity.
3Device complexity
If pre-defined rules are used for processing unit assignment, then system complexity is reduced, but assignment efficiency deteriorates
Solution Approach 1:
The patent introduces machine learning models as intermediary components that process multiple input factors (application characteristics, system state, historical performance data) and translate them into optimized processing unit assignment decisions. This intermediary layer enables sophisticated, efficient assignment without requiring the entire system architecture to become complex, as the ML model encapsulates the decision logic.
Solution Approach 2:
The system replaces traditional rule-based mechanical decision-making with machine learning-based intelligent decision-making. The ML models analyze patterns and predict outcomes more effectively than fixed rules, achieving superior assignment efficiency while managing complexity through the specialized ML component rather than system-wide complexity.
Data Source
AI summary
Apparatuses, systems, and techniques for assigning execution of applications to various processing units using machine learning are disclosed herein. Usage data for an application to be executed using a computing system including an integrated processing unit and a discrete processing unit is identified. At least a portion of operations of the application to be executed using the integrated processing unit or the discrete processing unit based on the usage data and in view of at least one of one or more system performance metrics or one or more user experience metrics associated with executing the application using the integrated processing unit and the discrete processing unit.