V-Sync: Driver or Game?

[Glide]

I was wondering what downsides there are to using using driver-level V-Sync compared to in-game V-Sync options.
Is this potentially going to add a lot of input lag?

Disabling in-game v-sync and forcing v-sync in the NVIDIA drivers often seems to fix or greatly improve frame-pacing issues.

Playing through the second episode of Life is Strange really emphasized that. The default frame-pacing is terrible, which seems to be a common issue with games running on Unreal Engine 3.


If I set the maximum number of pre-rendered frames to 1, frame pacing is greatly improved:


If I do this and also disable the in-game v-sync, forcing v-sync in the driver instead, I get excellent frame-pacing:


The difference in fluidity is staggering on this 60Hz display.

It proved to be a perfect demonstration of how your framerate can appear to be great, but the game itself is a stuttering mess.
I should point out that these results were with a single card - an old 570 - though I had similar problems with the first episode when I still had a 970.

Input lag does not matter for this game because I'm playing it on a TV with a controller, but I wonder whether it would be a problem in other games if I'm using the driver-level v-sync instead of the in-game v-sync.
To have such smooth frame pacing, I have to imagine that NVIDIA are buffering the frames.

This also raises other questions for me:
Is frame-pacing an issue with G-Sync at all, or does it bypass that altogether?
Do AMD have similar options that will improve frame pacing in games?

[lexlazootin]

What software are you using to record frame times? i got a amd and might check it out.

[Glide]

What software are you using to record frame times? i got a amd and might check it out.

I had MSI Afterburner polling at 100ms. (its minimum interval)
I know that is not ideal, since the frame time is 16.7ms, but the results were consistent and very obvious in person.

[flood]

could be that the game is manually sleeping for a period of time.

[lexlazootin]

Do AMD have similar options that will improve frame pacing in games?

My driver hates me... i can't get my 280x driver to force VSync anything. Idk if it's something I've done or just that's just normal AMD behavior. Well i'm getting a 970 soon so I've deffs be trying the driver VSync because that's awesome performance difference!

[masterotaku]

I have this same experience. I found that I have to force vsync in the Nvidia drivers in lots of games. In some of them, I even have to limit fps to the refresh rate with RTSS (F.E.A.R., Tomb Raider 2013 and some more games) or else I get microstuttering.

There are also rare cases like Castlevania Mirror of Fate, which is totally smooth only at 50fps or 100fps (this game is also capped at 100fps, that may be a reason). Some unity engine games like Anna or Shelter only work perfectly at 50fps in 2D or 3D. More fps than that are repeated. I play the games mentioned in this paragraph in 3D at 100Hz (50fps per eye).

Microstuttering is really easy to perceive when you don't have motion blur .

[Glide]

I have this same experience. I found that I have to force vsync in the Nvidia drivers in lots of games. In some of them, I even have to limit fps to the refresh rate with RTSS (F.E.A.R., Tomb Raider 2013 and some more games) or else I get microstuttering.

There are also rare cases like Castlevania Mirror of Fate, which is totally smooth only at 50fps or 100fps (this game is also capped at 100fps, that may be a reason). Some unity engine games like Anna or Shelter only work perfectly at 50fps in 2D or 3D. More fps than that are repeated. I play the games mentioned in this paragraph in 3D at 100Hz (50fps per eye).

Microstuttering is really easy to perceive when you don't have motion blur .

It was interesting because I rarely ever experience this sort of microstuttering where the game is running at 60 but it stutters very badly, since I have only ever used single GPU setups.
Normally if a game is stuttering, the performance is bad as well.

I tried all of the typical solutions: Driver frame limiting, RTSS frame limiting, increasing the maximum number of pre-rendered frames (which usually results in smoother gameplay at the cost of higher latency), even changing the game to its lowest quality settings, and combinations of all the above, but forcing the game to 1 pre-rendered frame and using driver v-sync was the only thing which fixed it in this case.

Do AMD have similar options that will improve frame pacing in games?

My driver hates me... i can't get my 280x driver to force VSync anything. Idk if it's something I've done or just that's just normal AMD behavior. Well i'm getting a 970 soon so I've deffs be trying the driver VSync because that's awesome performance difference!

That concerns me a bit. I have been waiting to see how the 390X turns out, but AMD have always had a bad driver reputation, and it looks like FreeSync is not offering the same experience as G-Sync.

I would prefer to switch after that 970 mess, but maybe I will end up waiting for the next Titan.
The current Titan X looks great, but I don't plan on upgrading yearly, and the next cards will be moving to a smaller process, have HBMv2, and hopefully Pascal will have DX12 Tier 3 support. (Maxwell is Tier 2)

[Chief Blur Buster]

Glide,

Excellent observation of frame pacing differences!

However, different methods framepacing could also lead to more input lag. Disabling VSYNC in both driver/game and using a fps_max limiter in a Source Engine game (1Hz below refresh rate), produces worse frame pacing but much less input lag. I am especially curious about the input lag differences of the various different methods of frame pacing, as better frame pacing sometimes (not always) produces more input lag.

For motion quality, better frame pacing is far more important, and improves enjoyment of many video games that has lots of motion, especially when a single millisecond or two doesn't matter (e.g. non-professional-competitive environments)

This thread is also of great interest to strobed/LightBoost display users, as superior framepacing is extremely important to witnessing a very good "LightBoost" effect (as microstutters are more visible during strobing blur reduction).

[Glide]

I don't have the hardware required to measure input lag, so I was hoping that someone who does might be able to test this and see whether there is much difference.

I expect that using driver-level v-sync probably does add some latency over in-game v-sync, however reducing the maximum number of pre-rendered frames to 1 seems to counteract that.
When set to 0 (the default) it is up to the application to decide, and 3 is used if it does not specify a value.

I have tried this with a couple of other games/engines now and have seen similar improvements, though nothing was quite as bad as this to begin with, and only some of them need driver-level v-sync to deliver good frame pacing.
In some games, overriding the maximum number of pre-rendered frames seems to be enough.

I want to do more testing though, because I had been under the impression that reducing the maximum number of pre-rendered frames was supposed to reduce performance.
In all of my testing so far, reducing this value to 1 has always improved frame-pacing.

What I suspect - though I don't have any data to back this up yet - is that increasing the maximum number of pre-rendered frames may help keep the framerate high if you are pushing the limits of your GPU, but it does this at the cost of good frame-pacing.

As long as you tune the graphics settings appropriately for your GPU so that you can maintain a constant 60, it seems like you're better off using this reduced value.

For motion quality, better frame pacing is far more important, and improves enjoyment of many video games that has lots of motion, especially when a single millisecond or two doesn't matter (e.g. non-professional-competitive environments)

That is generally my feelings on the matter as well. I don't play games competitively any more, everything is single-player.
So I am not too concerned if it is increasing latency, because if it is, it seems to be minimal and the improved frame-pacing is far more noticeable.

I will also mention that the TV I'm using right now is only a 60Hz panel without any flicker (no PWM or backlight scanning) and the frame-pacing issues were still very obvious. I'm sure it would have been much worse on a low persistence display though.
Unfortunately my TV combines its backlight scanning options with interpolation so it works well for movies, but is useless with games.

[MrBonk]

I don't think the differences would be huge in terms of latency (For example, frame pacing in a competitive game wouldn't be as important as low latency. Vsync would give you latency increases either way short of Gsync)


But the rule of the thumb for me in all of my experiences is that most games built in Vsync is subpar. Especially UE3 games.
The best frame pacing will always be a combination of>

If 60+hz| In game vsync disabled. Frame cap set with RTSS to your desired refresh. Vsync forced through drivers.
If 30hz on 60hz monitors| In game vsync disabled. Frame cap set with RTSS to 30. Maximum pre-rendered frames set to 1 in drivers and 1/2 refresh vsync as well.