Predictive Thermal Management for Sensorless Workload Assistants
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Equipping each workload assistant device in a compute device with temperature sensors to measure and report temperature adds cost and complexity, necessitating a more efficient thermal management solution.
Innovation Solution
Implementing predictive thermal management within compute devices that infer the temperature of workload assistant devices based on utilization metrics, using profiles that relate these metrics to temperature, allowing for adjustments in fan speed or device activity to maintain thermal thresholds without temperature sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If temperature sensors are equipped on each workload assistant device to directly measure and report temperature, then temperature measurement accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent introduces an intermediary approach by using the CPU's existing thermal management infrastructure and utilization metrics as a mediator to infer workload assistant device temperatures, rather than directly measuring them with sensors. This intermediary method allows temperature estimation without adding physical sensing components to each workload assistant device.
Solution Approach 2:
The patent creates a virtual copy of the temperature measurement function by modeling and predicting temperatures based on utilization metrics and thermal profiles, rather than physically measuring each device. This copying approach replicates the temperature information needed for thermal management without requiring physical sensors on every component.
2Reliability
If temperature sensors are equipped on each workload assistant device to directly measure and report temperature, then temperature monitoring capability is improved, but cost increases
Solution Approach 1:
The patent makes the CPU's thermal management system universal by enabling it to monitor temperatures of multiple workload assistant devices using a single centralized approach. The CPU utilizes its existing thermal management capabilities and shared thermal profiles to monitor all workload assistant devices, eliminating the need for dedicated sensors on each device.
Solution Approach 2:
The system implements self-service thermal management where the CPU autonomously monitors and manages the thermal state of workload assistant devices using its own existing resources and metrics. The CPU serves itself and the workload assistant devices by inferring their temperatures from utilization data, without requiring external sensing infrastructure.
3Device complexity
If predictive thermal management is implemented using utilization metrics instead of temperature sensors, then device complexity is reduced, but temperature measurement precision deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-characterizing workload assistant devices with thermal profiles that correlate utilization metrics to temperature ranges. These profiles are established in advance through testing and characterization, allowing the system to predict temperatures based on current utilization without real-time sensing, thus reducing complexity while maintaining acceptable precision.
Solution Approach 2:
The patent changes the measurement parameter from direct temperature sensing to utilization metric monitoring. By shifting from measuring temperature directly to measuring utilization (power consumption, activity level) and mapping it to temperature through profiles, the system reduces hardware complexity while providing sufficient thermal management capability.
Data Source
AI summary
Technologies for providing predictive thermal management include a compute device. The compute device includes a compute engine and an execution assistant device to assist the compute engine in the execution of a workload. The compute engine is configured to obtain a profile that relates a utilization factor indicative of a present amount of activity of the execution assistant device to a predicted temperature of the execution assistant device, determine, as the execution assistant device assists in the execution of the workload, a value of the utilization factor of the execution assistant device, determine, as a function of the determined value of the utilization factor and the obtained profile, the predicted temperature of the execution assistant device, determine whether the predicted temperature satisfies a predefined threshold temperature, and adjust, in response to a determination that the predicted temperature satisfies the predefined threshold temperature, an operation of the compute device to reduce the predicted temperature. Other embodiments are also described and claimed.