QML comes with gigantically more memory usage and it's very inefficient at rendering text. If you want to render something that looks like a wikipedia page, prepare to pay a heavy price.

GPU were designed for graphics, not UIs, we can certainly make them work within constraints, but it's not always a huge andvantage, and has drawbacks. 99% of what you use a UI for doesn't really get that much advantage from the GPU.

That said, I think there is advantage there, we just have to learn how to use it well.

Oh of course. I was talking about embedded use cases (Automotive dashboards etc..), where it is just mostly moving around various textures, playing with colors/opacity etc... (Basically the GUI lvgl youtube videos are showing off: https://www.youtube.com/watch?v=EqTG-3NHHAs would be a lot smoother and less CPU hungry in QtQuick )

For desktop I mostly suggest QWidget based UIs for more than one reason.

Yeah, I think you would use other native technology for videos, but for simple UIs? If it can work on an ESP-32, it can work on x86 big-iron. ( I don't have a dog in this fight, yet, so tell me how I'm wrong.)

Oh it really depends on your use case and the hardware you use for embedded devices. If all you need are some buttons and some kind of static lists, lvgl should be fine too. But for anything where you want things like smooth scrolling/higher resolutions you probably want to use the GPU.

For eg. A badly implemented "speedometer widget" was making our whole embedded GUI stutter at 15-20 fps and 100% cpu usage (which caused other problems like overheating, timing problems etc...). 5 lines of change to make the GPU do most of the work and we were back to smooth 60fps and low cpu usage.

At my last job, the requirements were like: you might as well directly target embedded linux socs instead of microcontrollers like ESP-32, because the price difference between them is not that much for the number of these devices we wish to sell.