GPU IFPC Wake-Up Timing for Periodic XR Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques fail to address the performance penalty associated with inter-frame power collapse (IFPC) exit delay at graphics processing units (GPUs) when processing fixed periodical workloads, leading to unnecessary power consumption and latency.
Innovation Solution
A hint relating to a timer value is provided to the graphics management unit (GMU) firmware to avoid unnecessary hysteresis timeouts and advance the GPU wake-up timeline, eliminating delays in processing commands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If the GPU enters inter-frame power collapse state to save power, then power consumption is reduced, but GPU wake-up latency increases causing performance penalty
Solution Approach 1:
The system performs preliminary actions by providing a hint to the GMU firmware about the expected workload timing, allowing the GPU to be awakened in advance before the actual workload arrives. This eliminates the wake-up latency penalty while still enabling power collapse during idle periods, as the GPU is ready immediately when the workload is submitted.
2Device complexity
If the GPU uses fixed hysteresis timeout for power management, then power management is simplified, but unnecessary timeouts occur for fixed periodical workloads causing performance degradation
Solution Approach 1:
The system transitions from a static, fixed hysteresis timeout approach to a dynamic approach where the GMU firmware adjusts power management behavior based on workload characteristics. By providing a hint about fixed periodical workloads, the system dynamically modifies the power management strategy to eliminate unnecessary timeouts while maintaining simplicity for other workload types.
Solution Approach 2:
The system changes the power management parameters dynamically based on the workload type. When a fixed periodical workload is detected via the hint, the hysteresis timeout parameter is adjusted or eliminated for that specific workload pattern, while other workloads continue to use the standard timeout values, thus optimizing performance without significantly increasing overall system complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This disclosure provides systems, devices, apparatus, and methods, including computer programs encoded on storage media, for optimizing power and performance of XR workloads. A graphics processor may receive, from an application, an indication of a time period for a timer associated with exiting an IFPC state. The graphics processor may process, upon triggering the timer associated with exiting the IFPC state, one or more predefined workloads. The graphics processor may initiate the IFPC state upon the one or more predefined workloads being finished processing. The graphics processor may exit the IFPC state upon detecting an expiration of the timer.