[SOLVED] Random Frame Spike

[zzuegg]

Anyone else getting random frame spikes?
This code on an empty simple application and with a 120hz vsynced monitor prints:

    @Override
    public void simpleUpdate(float tpf) {
        super.simpleUpdate(tpf);
        if(tpf*1000>16){
            System.out.println("Spike : "+tpf*1000);
        }
    }

Spike : 16.539299
Spike : 23.974499
Spike : 16.2832
Spike : 16.6841
Spike : 35.360897
Spike : 27.8832
Spike : 19.6469
Spike : 34.481796
Spike : 33.9422
Spike : 28.845499
Spike : 25.912899
Spike : 43.3629
Spike : 42.961
Spike : 41.0638
Spike : 40.8127
Spike : 44.5886
Spike : 28.017199
Spike : 38.6125
Spike : 85.4524
Spike : 40.575798

The problem is most likely on my side, but i don’t know where to start looking. Happens with lwjgl and lwjgl3

[pspeed]

I’m running an older version of JME… and I have heavily instrumented it to notify me of frame spikes and spit out the app state that caused them. I generally do not see random frame spikes and can always attribute them to one of my app states.

Caveats:
my JME is 3.6 before it was released “sometime”
I run at 60 FPS
most of my watchdogs only trigger after 50 ms.

Note: I also generally consider ‘float’ tpf to be a bit garbage and never use it for anything. So all of my timings are based on System.nanoTime() and so wouldn’t reflect any issues with float-based TPF or how JME tracks time.

I do recommend you switch to a System.nanoTime() per update to see if that aligns with what you see.

Also, have you enabled the detailed profiler to see where the time goes?

[zzuegg]

I used NanoTime in my code. I stripped it down as much i could to have most basic test case.
That’s the frame time graph i recorded. The interval seems to not total random:

JVisualVM might show a slight increase in cpu usage. I say might, because the bump i so little i am not 100% sure. Rather interestingly the frame after the spike makes up for the time mostly. Unfortunately this increases the problem if you are using classical movement*tpf code.

I tried with jm3 3.5.1, 3.6 and 3.6.1. Result are similar.

Nvidia 1080ti
Amd ThreadRipper 1950x

[pspeed]

Are you using vsync on or having JME control the frame rate with sleeps? (Just trying to make sure the picture is complete… theoretically, either way can cause catch-up frames.)

[zzuegg]

vsync

[pspeed]

Which OS?

I feel like I’ve seen similar reports about Windows 11 having random spikes unless you perform some magic config or what-not.

[zzuegg]

Windows 10. Microsoft says my pc is too old for Win11.

I played with the windows graphics settings, “Hardware accelerated GPU scheduling and variable refresh rate”
Enabled both.
. and the spikes are smaller now.