Dynamic Workload Classification for Processor Power Limit Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer processor systems fail to optimize performance between different workloads and within workloads due to fixed power limits (PL1 and PL2) and averaging window (Tau), which limits the effectiveness of the turbo feature, particularly for workloads with longer sustained power usage.
Innovation Solution
A system that predicts workload classifications using hardware data and adjusts power limits dynamically, allowing for personalized performance management by modifying thermal constraints and power limits (PL1 and PL2) based on predicted workload behavior, thereby optimizing the use of the turbo feature for both short and long-duration workloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fixed power limit PL2 values are set to exceed three times power limit PL1 values to realize full processor core performance, then peak performance is improved, but turbo on-time becomes very short which prevents sustained workloads from benefiting from the turbo feature
Solution Approach 1:
The patent dynamically adjusts power limit PL2 values based on workload classification. The system categorizes workloads into different types (e.g., bursty, sustained, mixed) and applies different PL2 multipliers appropriate to each type. This transforms the static PL2 configuration into a dynamic, workload-adaptive setting that optimizes both peak performance and turbo duration according to actual workload requirements.
Solution Approach 2:
The system changes the parameter values of power limits based on detected workload characteristics. By monitoring workload behavior patterns and classifying them, the system adjusts PL2 as a variable parameter rather than a fixed value, enabling optimal performance for different workload types while resolving the contradiction between peak performance and sustained turbo availability.
2Ease of operation
If fixed power limit PL1 and PL2 values are configured based on industry standard benchmarks, then standardization and ease of configuration are improved, but performance optimization between different workloads is prevented
Solution Approach 1:
The system automatically classifies workloads and adjusts power limits without requiring manual configuration or user intervention. The workload classification mechanism and dynamic PL2 adjustment operate autonomously, detecting workload patterns and optimizing performance parameters in real-time, thus maintaining ease of operation while achieving workload-specific optimization.
Solution Approach 2:
The system implements a feedback loop where workload behavior is continuously monitored, classified, and used to adjust power limit settings. This closed-loop control enables the system to adapt to different workload types automatically, resolving the contradiction between fixed configuration simplicity and dynamic performance optimization.
Data Source
AI summary
In one embodiment, a method receives data regarding processing of a workload by a processor. The data is input into a prediction engine configured to classify the data into a plurality of workload classifications. Each workload classification describes different temporal behavior of the workload. Then, the method outputs a prediction for at least one of the plurality of workload classifications, wherein the prediction is used to control performance of the processor in an upcoming period of time.