Our application framework is doing this on purpose because this is the typical scenario for videogames. We have plans to create a specific example showcasing render on demand. For that we need extra things on the application framework (we need to detect the sync rate and sleep, this is natively available in many platforms like Choreographer in Android - we are already using it -).

Could you please elaborate about 2 ( It is a shame sample code do not promote one of the strongest selling points of NoesisGUI) and 3 (Currently there are some subtle issues with variable framerate in NoesisGUI)?

Sample code I've mentioned is ApplicationLauncher\Src\Window.cpp, it doesn't use the return value of Noesis::IRenderer::UpdateRenderTree() to skip render.

As for my code, it is not easy to reproduce properly, but the issue I am having is:
When rendering on demand, there are occasional spikes in performance of Noesis::IRenderer::Render()/RenderOffscreen().

Jank happens after somewhat long periods of update w/o render.
The typical timeline is:

• .......
• update, skipped render (0ms)
• update, skipped render (0ms)
• update, skipped render (0ms)
• update, skipped render (0ms)
• update, render (30ms) // abnormally long render
• update, render (2ms) // back to normal
• update, render (2ms)
• update, render (2ms)
• .......

If I do not use the return value of Noesis::IRenderer::UpdateRenderTree() to skip render, then there is no jank. The performance is consistent.

When skipping rendering, I assume only Render() and RenderOffscreen() are skipped, right? You should keep invoking Update and UpdateRenderTree consistently.

Yes, Update and UpdateRenderTree are invoked consistently on vsync signal.