IC Power Mode Scheduling for Grid-Responsive Computing Loads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for managing power consumption in distributed computing systems, such as cryptocurrency mining or AI tasks, lack flexibility and responsiveness, leading to potential grid overload and inefficiencies.
Innovation Solution
A system for controlling power events in computing devices with integrated circuit chips, utilizing an asset management server to schedule power modes based on target parameters, including random time intervals and energy prices, enabling flexible and efficient power management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional power management methods are used in distributed computing systems, then system simplicity is maintained, but power consumption control flexibility and responsiveness deteriorate, leading to grid overload and inefficiencies
Solution Approach 1:
The system receives and stores event data in advance that provides target values for scheduling power events. Computing devices determine random times ahead of actual power mode transitions, allowing proactive power management rather than reactive responses. This preliminary action enables the system to adapt to varying power conditions while maintaining operational simplicity.
Solution Approach 2:
The patent implements dynamic power mode switching between first and second power modes based on monitored current values reaching target values. The system continuously monitors parameters and adjusts power modes in real-time, creating a dynamic power management approach that improves flexibility and responsiveness to changing conditions.
2Reliability
If power mode switching is implemented without random time intervals, then power control responsiveness is improved, but power surges and grid instability increase
Solution Approach 1:
Computing devices determine a random time within a time range in advance of the actual power mode transition. This preliminary determination of random transition times allows the system to prepare for power mode changes while distributing them across different time intervals, preventing simultaneous transitions that would cause power surges and improving grid stability.
Solution Approach 2:
The system uses periodic monitoring of current values against target values with random time intervals between power mode transitions. This periodic action with randomized timing maintains reliability by preventing synchronized transitions across multiple devices while still responding efficiently to power conditions.
3Ease of operation
If centralized power control is used without local autonomy, then system coordination is simplified, but responsiveness to local power conditions deteriorates
Solution Approach 1:
The patent segments power management authority between centralized event data provision and local autonomous execution. The asset management server provides target values and event data, while individual computing devices autonomously monitor their own power conditions and independently switch power modes based on local measurements. This segmentation maintains coordination simplicity while improving local responsiveness.
Solution Approach 2:
Computing devices perform self-service power management by locally monitoring current values for parameters and autonomously determining when to switch power modes based on received event data. Each device independently compares its own monitored values against target values and executes power mode transitions without requiring continuous centralized control, enhancing responsiveness while maintaining ease of operation through pre-provided event data.
Data Source
AI summary
A system includes a plurality of computing devices. Each of the plurality of computing devices includes a plurality of integrated circuit (IC) chips configured to perform a common type of computation in parallel with one another. At least one of the plurality of computing devices is configured to: receive, from an asset management server, an event data providing target values for one or more parameters to schedule a power event; store the event data in a storage coupled to the computing device; monitor one or more current values for the one or more parameters; and, based on determining that the current values for the one or more parameters have reached the target values, switch one or more of the plurality of IC chips corresponding to the computing device from a first power mode to a second power mode.