Adaptive VSYNC VS VSYNC input lag

[petrakeas]

I have been using RTSS + VSYNC to simulate double buffered VSYNC as explained in this Blur Busters article. Even though I can't take measurements, I can feel the difference in input lag versus using normal VSYNC.

I tried using the "adaptive" setting in Nvidia Control panel and I removed framerate capping from RTSS. I noticed that the input lag felt the same as the above method and it was much better than normal VSYNC. So, I think that "adaptive vsync" uses 2 buffers and not 3 buffers as the normal VSYNC in Nvidia cards. If this is the case, it makes the whole RTSS + VSYNC workaround unnecessary.

The games I tested were Overwatch and Just Cause 3 in fullscreen mode in Windows 10 using GTX1080.

I couldn't find any input lag tests that use "adaptive VSYNC". Are there any input lag tests that can confirm my hypothesis?

[RealNC]

I don't think anyone did any tests.

With that being said, the driver, at least for DX games, cannot decided whether to use double buffer or triple buffer vsync. That's up to the game.

If you have a 24Hz capable monitor, you might be able to tell with more certainty. A single frame makes a huge latency difference at 24Hz (41.6ms.)

[petrakeas]

With that being said, the driver, at least for DX games, cannot decided whether to use double buffer or triple buffer vsync. That's up to the game.

If the driver has no control over it at DX games, how it fast sync, adaptive vsync, vsync implemented at the driver level? For example Fast sync theoretically requires 3 buffers.

[RealNC]

If the driver has no control over it at DX games, how it fast sync, adaptive vsync, vsync implemented at the driver level? For example Fast sync theoretically requires 3 buffers.

Fast sync runs the game with vsync off. Also, DX triple buffer is completely different from fast sync triple buffer. Interestingly, a video came out today from Digital Foundry that is somewhat relevant:

https://www.youtube.com/watch?v=seyAzw9zEoY

As for driver vsync, it will just force vsync in whatever buffer mode the game uses. If in-game vsync is disabled, that means driver vsync will result in double buffering. If in-game vsync is enabled, and the game has a triple buffer vsync option, driver vsync will also be triple buffered. At least that was the case in the past. Last time I tested this was a few years ago. But I don't think anything has changed since then though.

[petrakeas]

That's the video I saw and had the idea to check the input lag of adaptive vsync. I think that the video has an error though. It mentions that fullscreen borderless windows mode uses 3 buffer vsync, while in reality it uses what the video calls true triple buffer vsync (similar to fast sync).

Yes, I know the difference between fast sync and 3 buffer vsync.

So, what you are saying is that adaptive vsync can't have different input lag to regular vsync?

[jorimt]

So, what you are saying is that adaptive vsync can't have different input lag to regular vsync?

NVIDIA's Adaptive V-SYNC is automatic "V-SYNC ON" whenever the framerate is above the monitor's max refresh rate, and is automatic "V-SYNC OFF" whenever the framerate is below the monitor's max refresh rate.

So yes, NVIDIA's Adaptive V-SYNC has less input lag than other forms of V-SYNC, but only when the framerate is below the monitor's max refresh rate, and then you get full tearing, since it's the same thing as V-SYNC OFF in that instance.

[petrakeas]

So yes, NVIDIA's Adaptive V-SYNC has less input lag than other forms of V-SYNC, but only when the framerate is below the monitor's max refresh rate, and then you get full tearing, since it's the same thing as V-SYNC OFF in that instance.

In the games I played, the frame got rendered fast enough that Vsync was never disabled in adaptive vsync. However, I did experience significantly less input lag without having screen tearing than using regular vsync . My hypothesis, is that the adaptive v-sync implementation is different to normal vsync in a way that it results in less input lag. Perhaps less buffers are queued for the display.

I am surprised that adaptive vsync hasn't been tested for input lag with a high speed camera or that no-one has even tried it before choosing the framecap + vsync method.

[RealNC]

Well, it could be. We don't know what tricks nvidia implements in their drivers.

But nobody tested it. Again, if your display is 24Hz capable, you could try adaptive vs full vsync and see if there's a clear difference.

[jorimt]

My hypothesis, is that the adaptive v-sync implementation is different to normal vsync in a way that it results in less input lag. Perhaps less buffers are queued for the display.

As far as I'm aware, Adaptive V-SYNC is identical to standard V-SYNC when the framerate is above the refresh rate, and was created to remedy double buffer FPS lock when the framerate went below the refresh rate by automatically disabling V-SYNC.

[H]ardOCP tested Adaptive V-SYNC when it originally released, and the article seems to back my assumption up, specifically in this paragraph:

This game scenario case [Mass Effect 3] where the framerate is always higher than the refresh rate indicates that Adaptive VSync has the same result as VSync on. There is really no benefit to it in this case because the framerate was faster than the refresh rate to begin with. Therefore, in any game where the framerate is faster than the refresh rate, Adaptive VSync has little benefit. The benefit comes in when the framerate is under the refresh rate as you will see below.

Full article link:
https://www.hardocp.com/article/2012/04 ... y_review/2

And yes, they aren't testing for input lag, but framerate, so it doesn't 100% rule out your "hypothesis," but it's safe to also assume that if Adaptive V-SYNC had a notable input lag reduction vs. standard V-SYNC above the refresh rate, NVIDIA would have promoted the heck out of that fact when it originally released.

If there is truly a reduction of input lag with Adaptive V-SYNC vs. standard V-SYNC above the refresh rate, the only thing I could think is that NVIDIA may have tweaked the driver since Adaptive V-SYNC's original release for it to use Fast Sync as a base (a form of true triple buffer implementation above the refresh rate) instead of double buffer V-SYNC, though that would be pretty easy to tell, as you'd get recurring stutter (more or less, depending on how high the framerate is above the refresh rate) when directly compared to double buffer V-SYNC in that scenario.

[petrakeas]

@RealNC

I'll test in on my TV at 24 Hz and come back with results.

If there is truly a reduction of input lag with Adaptive V-SYNC vs. standard V-SYNC above the refresh rate, the only thing I could think is that NVIDIA may have tweaked the driver since Adaptive V-SYNC's original release for it to use Fast Sync as a base

I don't think that this is necessary. Using just 2 buffers instead of 3 (as most modern games do in regular Vsync) could result in less input lag. However, RealNC states that this can't be controlled by the driver. I am not familiar with DX, so I can't argue with that.

Anyway, you can also give it a shot and see how it feels.