Virtual Temperature Sensors for Multi-GPU Cooling Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cooling systems for complex computing components like multi-GPU devices are inefficient due to the limitation of using only 4 temperature sensors per Baseboard Management Controller (BMC) device, ignoring the majority of temperature sensors and requiring manual configuration and firmware modifications.
Innovation Solution
Implementing a BMC system that identifies multiple physical temperature sensors and creates virtual temperature sensors for each, allowing the BMC to retrieve and utilize temperature data from all sensors in a cooling control algorithm to manage a cooling system effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If the BMC device uses only 4 temperature sensors per PLDM device to conserve processing resources, then the processing load on the BMC device is reduced, but the cooling system cannot utilize temperature data from all sensors in complex computing components
Solution Approach 1:
The patent creates a virtual temperature sensor that copies the functionality of multiple physical temperature sensors. This virtual sensor aggregates temperature data from all physical sensors in a computing component, allowing the BMC device to access comprehensive temperature information without directly processing data from all individual sensors, thus resolving the contradiction between processing resource conservation and temperature data utilization
Solution Approach 2:
The virtual temperature sensor acts as an intermediary between the physical temperature sensors and the BMC device. It receives temperature data from all physical sensors and presents a consolidated view to the BMC, enabling the BMC to access complete temperature information while maintaining limited processing resources
2Productivity
If conventional solutions manually configure only 4 temperature sensors in the platform table, then the BMC device processing load is reduced, but the system requires manual configuration and firmware modifications
Solution Approach 1:
The system performs self-configuration by automatically discovering and registering all temperature sensors in the computing component. The virtual temperature sensor is dynamically created and configured without manual intervention or firmware modifications, allowing the BMC device to automatically manage cooling systems with any number of sensors
Solution Approach 2:
The patent introduces dynamic configuration where the number of temperature sensors is no longer fixed at 4 but can adapt to any number of physical sensors in the computing component. The system dynamically adjusts its behavior based on the actual number of sensors present, eliminating the need for manual configuration
3Device complexity
If the BMC device limits temperature sensor monitoring to 4 sensors per PLDM device, then the cooling algorithm processing is simplified, but the majority of temperature sensors are ignored
Solution Approach 1:
The patent merges multiple physical temperature sensors into a single virtual temperature sensor. This consolidation allows the cooling algorithm to process temperature data from all sensors through a unified interface, maintaining algorithm simplicity while achieving comprehensive temperature monitoring that improves reliability
Data Source
AI summary
A multiple-virtual-temperature-sensor-per-computing-component cooling system includes a chassis housing a cooling system and a BMC device coupled to a computing component that includes physical temperature sensors. The BMC device identifies the physical temperature sensors included in the computing component, and creates virtual device(s) that includes a respective virtual temperature sensor for each of the physical temperature sensors. The BMC device then retrieves respective temperature data from each of the physical temperature sensors and, for each virtual temperature sensor included in the virtual device(s), provides the respective temperature data that was retrieved from the physical temperature sensor for which that virtual temperature sensor was created in a cooling system control algorithm to generate cooling system control information. The BMC device then uses the cooling system control information to control the cooling system.