Processor Power Management via Segmented Subsystems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multipurpose processor systems, power management is often limited to an 'all or nothing' approach, failing to effectively save power due to the difficulty in selectively powering down unused resources, especially when applications have unpredictable computational needs and mutual awareness is hard to maintain.
Innovation Solution
The system implements independently controllable subsystems with dedicated memory resources, allowing for separate power management of data processing and data collection domains, enabling power-down or reduced power states based on idleness predictions without mutual dependency, using firmware to control power levels in on-die SRAM without the need for special hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If power management is performed in an all or nothing way, then power savings are achieved when no tasks are active, but power savings opportunities are limited and resources cannot be selectively powered down
Solution Approach 1:
The system divides the processor into multiple independently controllable subsystems, each with its own power management. This allows selective powering down of unused subsystems while keeping others active, resolving the contradiction between achieving power savings and maintaining selective resource control capability.
2Productivity
If resources are spread across different entities in multipurpose applications, then resource utilization is improved, but selective power-down of unused resources becomes difficult
Solution Approach 1:
The processor is segmented into independent subsystems with dedicated resources, allowing each to be managed separately. This maintains high resource utilization across multiple applications while simplifying power management through independent control of each subsystem.
Solution Approach 2:
Each subsystem includes self-contained resources and can operate independently. This self-service capability allows subsystems to be selectively activated or deactivated based on workload requirements, reducing power management complexity while maintaining resource utilization.
3Device complexity
If firmware architecture is used to control power levels, then hardware complexity is reduced, but traditional firmware architecture lacks clean isolation of data collection and processing resources
Solution Approach 1:
The system segments data collection and data processing resources into separate subsystems with dedicated memory and control logic. This architectural segmentation enables clean isolation detectable by firmware, reducing hardware complexity while maintaining resource isolation capability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In an embodiment, a processor includes first logic to determine first power to be provided to a first portion of a computational resource during a time period. The first portion may be reserved for execution by the processor of a first workload to be executed during the time period. The first power may be determined based at least in part on the first workload and independently of a second workload. The processor may include second logic to determine second power to be provided to a second portion of the computational resource during the time period. The second portion may be reserved for execution by the processor of the second workload during the time period. The second power may be determined based at least in part on the second workload and independently of the first workload.