GPU Subsystem Monitoring for Safety-Critical Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In safety-critical environments, such as aviation, graphical applications require assurance that the images created by GPUs correspond to the intended images, but existing monitoring methods are inadequate as they either focus on limited subsets of GPU subsystems or are computationally intensive, failing to continuously verify the operation of all GPU subsystems.
Innovation Solution
A method where a CPU provides sample rendering data and subsystem commands to the GPU, receives test results, and determines if they match expected outcomes, identifying any subsystems that have ceased to operate correctly, while selectively testing active and unexposed subsystems to minimize computational impact.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If comprehensive monitoring of all GPU subsystems is implemented, then reliability is improved, but computational overhead and processing time increase
Solution Approach 1:
The patent divides GPU monitoring into separate subsystem-level tests rather than comprehensive whole-GPU testing. Each subsystem (vertex shader, fragment shader, geometry processing, etc.) is tested independently with targeted test seeds, allowing parallel execution and reducing total monitoring time while maintaining coverage of all critical components.
Solution Approach 2:
The patent implements selective monitoring that focuses test resources on subsystems actually used by the running application. By identifying active subsystems through command analysis and testing only those, the system achieves adequate reliability verification without the excessive computational overhead of testing all possible subsystems regardless of usage.
2Measurement precision
If all GPU subsystems are continuously tested, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary analysis of application commands to identify which GPU subsystems will be active before execution. This advance knowledge allows the monitoring system to prepare targeted test seeds and validation criteria for only the relevant subsystems, simplifying the monitoring architecture while maintaining precise verification of operational accuracy.
Solution Approach 2:
The patent applies different monitoring strategies to different subsystems based on their specific characteristics and usage patterns. Each subsystem receives customized test seeds and validation criteria tailored to its function, achieving high measurement precision for each component without requiring a complex uniform monitoring approach across the entire GPU.
3Productivity
If selective monitoring of active subsystems is implemented, then productivity is improved, but reliability may be compromised
Solution Approach 1:
The patent implements a feedback mechanism where test results from each subsystem are immediately compared against expected outcomes. When deviations are detected, the system can trigger alerts, retry tests, or escalate to comprehensive monitoring modes, ensuring that selective monitoring does not compromise reliability while maintaining high productivity through efficient resource utilization.
Solution Approach 2:
The patent makes the monitoring system dynamic by adjusting the scope and intensity of testing based on application behavior and system state. The monitoring granularity can be adapted in real-time, expanding to cover additional subsystems when anomalies are detected or when application usage patterns change, thus maintaining reliability while optimizing productivity under varying conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and computer-readable media for monitoring a graphics processing unit (GPU) of a host computer system, including providing at least one test seed and at least one subsystem command to the GPU; receiving at least one test result from the GPU in response to providing the at least one test seed and at least one subsystem command to the GPU; and if the at least one test result does not correspond to at least one expected result, identifying the GPU as being in an undesired state. The subsystems to be tested may be a subset of all subsystems. The determination of subsystems to be tested may be determined in real-time, based on graphical application instructions. The subsystems to be tested may also be pre-determined.