Project : minimizing input lag

[Jim Liak]

This is a repost from here.

Hi, i bought my self a new phone which can shoot videos at 100fps ( slow motion ).
I thought it was a good opportunity to try and analyze how Assetto Corsa behaves regarding input lag.

I did a test. I took slow motion videos for the following settings :

1. latency = 0, fullscreen off, vsync=off ( 90 fps )
2. latency = 0, fullscreen on, vsync=off ( 90 fps )
3. latency = 0, fullscreen on, vsync=on ( 60 fps )
4. latency = 0, fullscreen on, vsync=off, lock at 63fps
5. latency = 1, fullscreen on, vsync=on ( 60 fps )
6. latency = 1, fullscreen on, vsync=off, lock at 63fps
7. latency = 1, fullscreen on, vsync=on NVIDIA ( 60 fps )

Lets explain some things :

1. My monitor is a TV, which WILL introduce lag. It was set at game mode.
2. Latency is the known "Maximum frame latency" setting at Steam \SteamApps\common\assettocorsa\system\cfg, which is supposed to minimize input lag.
3. I turn the wheel at my maximum speed.
4. 7th is with vsync ON only from Nvidia control panel. In game was OFF.
5. 90fps was what my system achieved when vsync and lock OFF
6. Fullscreen is the "Fullscreen rendering" inside game.

So here is the video :

Also if someone wants to see it in his/her pc , here is the file https://drive.google.com/file/d/0B1UNl21-h5iHbFVtUE9VX2pybTQ/view?usp=sharing

So i watched the video and counted the frames to see when the virtual wheel will move .
Results :

1. latency = 0, fullscreen off, vsync=off ( 90 fps ) - > 14 frames
2. latency = 0, fullscreen on, vsync=off ( 90 fps ) -> 12 frames
3. latency = 0, fullscreen on, vsync=on ( 60 fps ) -> 15 frames
4. latency = 0, fullscreen on, vsync=off, lock at 63fps -> 9 frames
5. latency = 1, fullscreen on, vsync=on ( 60 fps ) -> 18 frames
6. latency = 1, fullscreen on, vsync=off, lock at 63fps -> 11 frames
7. latency = 1, fullscreen on, vsync=on NVIDIA ( 60 fps ) -> 15 frames

If i did a mistake in counting i am open to suggestions but is seems that the 4th setting is the best.(which kind of surprised me as i thought that it would be the 6th)

Thanks.

 

[bigbawmcgraw]

Yeah, that makes sense. I just wasn't sure how you had measured your numbers. I'll edit the numbers.

It also seems very odd that
2. latency = 0, fullscreen on, vsync=off , lock off ( 90 fps ) -> 12 frames
is slower than this:
4. latency = 0, fullscreen on, vsync=off, lock at 63fps -> 9 frames

I'd expect lower frames lag from higher fps.

The GPU will be working harder at a higher frame rate, which could introduce lag in itself.

 

[Ghoults]

The GPU will be working harder at a higher frame rate, which could introduce lag in itself.

How? What could cause that?

 

[Stereo]

With latency set to 0, the GPU decides how many frames to buffer - if it's working to a shorter deadline (11ms instead of 16ms per frame) it might decide to increase that buffer (eg. if it was taking 15ms to render start to finish, it'd take 22ms to update at the higher framerate instead of 16)

 

[Jim Liak]

I guess it depends from the GPU. Mine is old and has trouble keeping above 90fps with the settings i use.
That is why when i tried to lock to 83fps instead of 63, i had also huge microstuttering.

63 fps is MY magic number. GPU is ok, there is no tearing and i have the least microstuttering . Image is not as fluid as with vsync ON but i can live with it, taking the benefit of the smallest input lag.

What DID surprised me was that latency=1 , made things slightly worse for that specific setting.

Now i use latency=0.

 

[Naliport Halulam]

Did you try with 56fps lock with no vsync ?

 

[Jim Liak]

No. 11 settings are more than enough though.

 

[Jim Liak]

Hi again, i upgraded my pc from Nvidia 660GTX to an AMD 470.
Also bought a new TV (LG 43LH5100).
So let's do Part 2 of minimizing input lag project.

Tests as follows ( i was unable to force vsync through crimson drivers )

1. latency = 0, fullscreen off, vsync=off, lock off (166 fps )
2. latency = 0, fullscreen on, vsync=off , lock off (171 fps )
3. latency = 0, fullscreen on, vsync=on ( 60 fps )
4. latency = 1, fullscreen on, vsync=on ( 60 fps )
5. latency = 1, fullscreen on, vsync=off, lock at 63fps
6. latency = 0, fullscreen on, vsync=off, lock at 60fps
7. latency = 1, fullscreen on, vsync=off, lock at 60fps
8. latency = 1, fullscreen on, vsync=on ( 60 fps ), VSR 1440p

So here is the video :

Also if someone wants to see it in his/her pc , here is the file https://drive.google.com/file/d/0B1UNl21-h5iHUUd2dTlrcXpMN2M/view?usp=sharing

As always, i watched the video and counted the frames to see when the virtual wheel will move .
Results :

1. latency = 0, fullscreen off, vsync=off, lock off (166 fps ) - > 17 frames
2. latency = 0, fullscreen on, vsync=off , lock off (171 fps ) - > 13 frames
3. latency = 0, fullscreen on, vsync=on ( 60 fps ) - > 17 frames
4. latency = 1, fullscreen on, vsync=on ( 60 fps ) - > 13 frames
5. latency = 1, fullscreen on, vsync=off, lock at 63fps - > 12 frames
6. latency = 0, fullscreen on, vsync=off, lock at 60fps - > 13 frames
7. latency = 1, fullscreen on, vsync=off, lock at 60fps - > 12 frames
8. latency = 1, fullscreen on, vsync=on ( 60 fps ), VSR 1440p - > 13 frames