Throttling Computational Units by Performance Sensitivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power allocation methods in computer systems are inefficient as they treat all active CPU cores homogeneously when adjusting frequency, failing to selectively distribute power based on individual core sensitivity to frequency changes, leading to suboptimal performance in heterogeneous workloads.
Innovation Solution
A method that analyzes the performance sensitivity of each computational unit to determine which cores can benefit most from frequency changes, allowing for selective power reallocation from less sensitive cores to more sensitive ones, thereby optimizing overall system throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If power is allocated homogeneously to all CPU cores, then implementation is simple, but performance gains are suboptimal for heterogeneous workloads
Solution Approach 1:
The patent applies local quality by differentiating power allocation across individual CPU cores based on their specific workload characteristics and sensitivity to frequency changes. Each core is evaluated independently to determine its performance sensitivity, allowing the system to allocate power selectively to cores that will benefit most, rather than treating all cores uniformly. This resolves the contradiction by maintaining simple overall implementation while achieving optimized performance through localized decision-making at the core level.
Solution Approach 2:
The patent segments the CPU core population into different groups based on their performance sensitivity to frequency changes. By dividing cores into segments (e.g., high-sensitivity cores vs. low-sensitivity cores), the system can apply different power allocation strategies to each segment. This segmentation enables the system to achieve high productivity by directing power to the right cores while keeping the allocation logic manageable through clear categorization.
2Speed
If frequency is increased to improve performance, then processing speed increases, but power consumption and thermal output increase
Solution Approach 1:
The patent applies partial action by increasing frequency only for the subset of CPU cores that exhibit high performance sensitivity, rather than increasing frequency system-wide. By identifying and targeting only the cores that will benefit most from frequency increases, the system achieves improved processing speed for critical workloads while limiting the overall power consumption and thermal output to what is necessary for those specific cores. This resolves the contradiction by applying frequency enhancement partially rather than universally.
Data Source
AI summary
A power allocation strategy limits performance of a subset of a plurality of computational units in a computer system according to performance sensitivity of each of the plurality of computational units to a change performance capability, e.g., frequency change. The performance of the subset of computational units may be limited by setting a power state in which the subset may be operated and/or reducing a current power state of the subset to a lower power state. The subset whose performance is limited includes computational units that are least performance sensitive according to stored sensitivity data. The subset may include one or more processing cores and performance of the one or more processing cores may be limited in response to a CPU-bounded application or graphics processing unit (GPU)-bounded application being executed.