search

GameTechDev / PresentMon

Capture and analyze the high-level performance characteristics of graphics applications on Windows.
https://game.intel.com/story/intel-presentmon/
MIT License
1.7k stars 155 forks source link

Fake frametime spikes due to ETW events flushing in PresentMon 2.2.0? #354

Open AlexUnwinder opened 1 month ago

AlexUnwinder commented 1 month ago

Good job on optimizing ETW backend and implementing flushing there, it really reduced reporting lag drastically and made PresentMon's data more realtime!

However, I noticed that ETW flushing seem to cause periodic fake spikes to be detected on frametime graph on my system. The screenshot displays PresentMon 2.2.0 native overlay with raw frametime and RTSS overlay. Pink (top right) graph is RTSS frame-start-to-frame-start frametime measured inside 3D application's hooked Present() call. There are no real frametime spikes on 3D app side. Violet (bottom right) frametime graph is reported by PresentMon service's API. Unrealistic spikes are displayed there and on PresentMon's own raw frametime overlay. The frequency of spikes on that graph seem to be directly related to the service's ETW flushing period. If I increase ETW flushing period with pmSetEtwFlushPeriod() inside my PresentMon service's client - the spikes become less frequent too, and they completely disappear if I disable flushing via setting period to 0. Which makes me believe that ETW flushing is direct reason of spikes. 

image

planetchili commented 1 month ago

Thank you for the report and the results of your experiments. We will prioritize reproducing this on our end and see what can potentially be done to fix the issue.