GPU Scheduling Based on Cooling Characteristics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing GPU scheduling frameworks fail to optimize workload allocation on multi-GPU systems due to neglecting thermal and connectivity considerations, leading to non-optimal performance and inefficient resource utilization.
Innovation Solution
An information handling system that receives cooling characteristics of special-purpose processing units and assigns identification numbers based on these characteristics to optimize workload distribution, taking into account thermal performance and connectivity, allowing for improved airflow and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing GPU scheduling frameworks assign workloads to GPUs without considering cooling characteristics, then the scheduling process is simple and fast, but the thermal performance and application performance deteriorate
Solution Approach 1:
The patent changes the scheduling parameter from simple GPU ID assignment to assignment based on thermal zones and cooling characteristics. The system divides GPUs into different thermal zones based on their cooling characteristics and uses this information to make scheduling decisions, thereby improving thermal performance while maintaining scheduling efficiency.
Solution Approach 2:
The patent performs preliminary classification of GPUs into thermal zones before workload assignment. By pre-establishing the thermal zone information and cooling characteristics of each GPU, the system can make informed scheduling decisions without adding complexity to the real-time scheduling process.
2Productivity
If GPU scheduling ignores physical position and connectivity characteristics, then the scheduling algorithm is simpler, but the resource utilization and performance deteriorate
Solution Approach 1:
The patent applies local quality by recognizing that different GPUs have different local characteristics (thermal zones, connectivity, physical position). Instead of treating all GPUs uniformly, the system assigns workloads to specific GPUs based on their local characteristics, optimizing resource utilization for different workload types.
Solution Approach 2:
The patent segments the GPU system into different thermal zones and connectivity groups. This segmentation allows the scheduling framework to make targeted assignments based on workload requirements, improving resource utilization without requiring a completely complex scheduling algorithm.
3Speed
If workloads are assigned to GPUs without considering airflow characteristics, then the allocation process is faster, but the application performance and thermal performance worsen
Solution Approach 1:
The patent performs preliminary characterization of GPU airflow characteristics and thermal zones before scheduling. This pre-processing step stores the cooling characteristics information, allowing the scheduling process to quickly retrieve and use this information without slowing down the actual workload assignment.
Solution Approach 2:
The patent introduces thermal zone information as an intermediary between the workload requirements and GPU selection. This intermediary layer translates airflow characteristics into scheduling decisions, maintaining scheduling speed while improving application performance through informed GPU selection.
Data Source
AI summary
An information handling system may include at least one central processing unit (CPU); and a plurality of special-purpose processing units. The information handling system may be configured to: receive information regarding cooling characteristics of the plurality of special-purpose processing units; and assign identification (ID) numbers to each of the plurality of special-purpose processing units in an order that is determined based at least in part on the cooling characteristics.