explain like ELI5 why the left one(virtualized) is much faster than the right one(native)?
>>110077789
>>110077909I’m not gonna post wall of text it is too technical, they are plainly vibecode.
You seem to have correctness issues, why don't you fix those first
Are you sure you get the exact same results if you ran the application independently?The root cause is because the slower native window is unable to utilize 100% of the CPU.It's waiting for something instead of rendering.If both had the same CPU/GPU% utilization, and one of the windows ran faster, that would be very strange and worth looking into.This problem is very specific towards a system without a GPU. It's possible that whatever VNC software is the bottleneck, not anything related to rpi, wayland, x11 and etc (it's possible the xvfb uses 4:2:0 sampling but the other one uses 4:4:4 or something, we can't see because your video is heavily compressed, but a screenshot might be able to show the artifacts).And note that if you plugged your rpi to a monitor directly, it should be able to use the GPU with a much better fps. And maybe there is a way to get it working remotely.
>>110078348If the problem was due to the video encoder, you should be able to see this inside of a task manager (like "tops" if you ssh into it)
full hud.
>>110078435>llvmpipeare you retarded
>>110078472What is the matters?I’m too lazy to make the gpu work, after all they both are the same except more fps when running software in one of them.
>>110078348> that would be very strange and worth looking into.It is definitely worth looking into it. I can’t find any answer even the AI has not been able to answer it it tell me should be slower.
>>110078514I'm sorry for making you install wayland, it seems like the triple buffering effect from the mailbox compositor only improves fps if you are hitting the monitor's max refresh rate.Maybe building vkcube from source and set FRAME_LAG to 3 is the last thing I can think of, and also I would try running the PI without the VNC since it may have some sort of overhead.
>>110078671>is the last thing I can think ofdude your last thing you can think of is everybody can think of too, do you even try to think?
>>110077975
>>110077789Why the left one is "faster" while the right one does nothing? The left on is dropping frames due to sorcery. But it's sorcery that makes fps go zoom. The one on the right is ol' reliable freeware grandpa. Cut grandpa some slack. How do I score?
>>110078435At a guess, the raspi GPU is ass and copying frame data to it is more expensive than copying the frame into RAM. Running it in a VM decouples your program and entire rendering pipeline from the GPU, leaving it free to render frames faster than they can be copied to the GPU. Instead it sends them to your VNC client, and if it can't get a frame rendered before another arrives, it simply drops a frame silently.
>>110079804Grandpa is waiting for the blackbox to load whatever is eating up all the processing, some some weird rasi
>>110079713The cpu is a fire hazard. And the fps is unreal or probably just fake.
>>110079736> is dropping framesI can’t tell whether they’re dropping frames, both of them look to be in same speed but the counter tell differently, but one cube spinning faster than another is the factual.
>>110080136I've seen 20,000 fps in vkcube. Not on my machine.
>>110080136Ever try starting a fire? You need way more than 96 C. That isn't a healthy temperature for a CPU though and OP should probably see about a (better) cooler.
>>110077789Only 2 types of peopleOne who say that VLC is most garbage video player in known universeand Retards
>>110080222> 96 CIt is 96 C when barely used, cpu at 10% use. What can go wrong?
>>110080189> 20,000 fpsTry running multiple benchmarks and see them tank hard.
>>110080276It will throttle ad it approaches 100, which still won't start a fire.
>>110077789Saar redeem some bitches and make better posts saar
Sweet, a schizo thread.
>>110080651It would take 100 to 300 watt btw. No house fire but house warming.
>>110077789try again with a working webm
>>110079804>copying frame data to it (gpu) is more expensive than copying the frame into RAMKinda make some sense btw, but I’m not able to show any actually convincing metrics or numbers or anything. I can only tell by gut feeling it is what it is.
