PC Latency - Vsync with/without Framerate cap

[RealNC]

I edited my post to reflect this.

So it's a bit of a black box situation?
Meaning users should always check if setting in-game limiters to FPS=Hz has a beneficial effect on latency?

Probably. Some games have other settings tied to their frame limiter setting that they don't tell you about. For example, in Forza Horizon games, if you enable the in-game limiter, it will also secretly engage fractional vsync. At 240Hz, if you set the limiter to 120FPS and have the in-game vsync setting also enabled, it will enable 1/2 vsync and thus the latency will skyrocket.

Most games don't do weird things like that though. Usually, if you set an in-game limiter to a value that's above your refresh rate and you're using vsync, it will simply have no effect.

[2mg]

Probably. Some games have other settings tied to their frame limiter setting that they don't tell you about. For example, in Forza Horizon games, if you enable the in-game limiter, it will also secretly engage fractional vsync. At 240Hz, if you set the limiter to 120FPS and have the in-game vsync setting also enabled, it will enable 1/2 vsync and thus the latency will skyrocket.

Most games don't do weird things like that though. Usually, if you set an in-game limiter to a value that's above your refresh rate and you're using vsync, it will simply have no effect.

So I just booted Talos Principle 2.
*note that this is the second UE5 game I'm testing, maybe that plays a role

Monitor @ 60Hz, Vsync ON (no option for what kind), GPU utilization below 60%.
PCL (av) @ 59FPS limit = ~36ms
PCL (av) @ 60FPS limit = ~70ms
PCL (av) @ 120FPS limit = ~96ms

59 and 60 make sense, 120 again ramps up the latency.

PS: what is fractional vsync?

[RealNC]

Monitor @ 60Hz, Vsync ON (no option for what kind), GPU utilization below 60%.
PCL (av) @ 59FPS limit = ~36ms
PCL (av) @ 60FPS limit = ~70ms
PCL (av) @ 120FPS limit = ~96ms

You forgot to test without any FPS limit.

PS: what is fractional vsync?

It's an operation mode of the GPU that syncs every two vsync periods, not one. When you enable 1/2 vsync at 240Hz, vsync behaves like its 120Hz even though the display is running 240Hz. Vsync lag is the same as 120Hz vsync. At 1/4 vsync (which is the highest possible setting) it's 60Hz vsync (it syncs every 4th vsync) with the display still running at 240Hz, and latency feels like 60Hz vsync (just a bit lower due to the faster scanout rate of 240Hz.)

Fractional vsync can be set by the game by using a higher value for the "present interval" setting of DirectX. Or it can be forced in the driver using Nvidia Profile Inspector. (The nvidia control panel and Nvidia App only allow you to force adaptive fractional vsync, not normal fractional vsync.)

[2mg]

You forgot to test without any FPS limit.

I switched to TP2 because BM Wukong offers only 30/60/120 steps.

In TP2, basically 61FPS or more, up to unlimited, is always ~90ms.

So 3 thresholds, 59 or less, 60, 61 or more.

Unfortunately I've no any external way to test latency further this other than "record keypress with my phone and guesstimate how right I am".

Question is, why is it working like that when 61 or more?

I think it's worthwhile to pursuit this problem, as there might be people playing with 60Hz (or even 120Hz or more, same issue would probably happen) panels and unknowingly having ingame limiters set to FPS>Hz and possibly having even more lag.

It's an operation mode of the GPU that syncs every two vsync periods, not one. When you enable 1/2 vsync at 240Hz, vsync behaves like its 120Hz even though the display is running 240Hz. Vsync lag is the same as 120Hz vsync. At 1/4 vsync (which is the highest possible setting) it's 60Hz vsync (it syncs every 4th vsync) with the display still running at 240Hz, and latency feels like 60Hz vsync (just a bit lower due to the faster scanout rate of 240Hz.)

Fractional vsync can be set by the game by using a higher value for the "present interval" setting of DirectX. Or it can be forced in the driver using Nvidia Profile Inspector. (The nvidia control panel and Nvidia App only allow you to force adaptive fractional vsync, not normal fractional vsync.)

Yeah I know NVCpl/app has that 1/2 Vsync.

While I understand it's adaptive, I've never been a fond user of adaptive sync, you kinda get worst of both worlds, vsync input lag when at threshold, and tearing when below it, and if FPS oscillates a lot Vsync is rapidly engaging/disengaging and it can look/feel gnarly and inconsistent.

Regarding adaptive fractional vs normal fractional Vsync:
1. What's the difference between them?
2. What is the purpose of that normal fractional sync?
3. Why not just limit to 1/2 or 1/4 FPS with a limiter and use normal Vsync?
3a. Same as above, but use VRR instead of normal Vsync?

[RealNC]

I switched to TP2 because BM Wukong offers only 30/60/120 steps.

In TP2, basically 61FPS or more, up to unlimited, is always ~90ms.

Question is, why is it working like that when 61 or more?

Yeah. That means vsync backpressure kicks-in since there's no FPS limit to prevent it. There's no mystery there. This has been known since forever. If you want to prevent vsync backpressure lag, cap your FPS to something that's at least a little lower than the vsync rate.

Regarding adaptive fractional vs normal fractional Vsync:
1. What's the difference between them?

An example is adaptive 1/2 vsync with the display set to 240Hz . This will vsync at a rate of 120Hz, unless FPS falls below 120, in which case it disables vsync. The result is the same as if you uses non-fractional adaptive vsync with the display set to 120Hz.

Non-adaptive 1/2 vsync on the other hand will always vsync at a rate of 120Hz. So the game will behave as if was running at 120Hz with normal vsync.

The benefit of fractional vsync is that you still get the faster scanout rate by leaving the display at its highest refresh rate.

2. What is the purpose of that normal fractional sync?

Hopefully explained above.

3. Why not just limit to 1/2 or 1/4 FPS with a limiter and use normal Vsync?

Indeed. Using 1/2 vsync at 240Hz is just the same as using full vsync at 120Hz. The lag is still there. I only use it sometimes when a game really doesn't like to have all the frame buffers empty. Some games will produce random motion hiccups every so often. For those I use the SK config I posted in a screenshot previously.

3a. Same as above, but use VRR instead of normal Vsync?

Fractional vsync doesn't work with VRR. If you enable fractional vsync, VRR will deactivate. Not that there's ever any need to use them together to begin with.

[2mg]

Yeah. That means vsync backpressure kicks-in since there's no FPS limit to prevent it. There's no mystery there. This has been known since forever. If you want to prevent vsync backpressure lag, cap your FPS to something that's at least a little lower than the vsync rate.

I mean, it was known about <60FPS@60Hz and 60@60Hz.

The 60>@60Hz is new (if you discount other possible queue buffers due to (U)LLM/Reflex) to me at least?

I've never seen anybody say "you should always at the very least cap FPS to Hz when using Vsync".

An example is adaptive 1/2 vsync with the display set to 240Hz . This will vsync at a rate of 120Hz, unless FPS falls below 120, in which case it disables vsync. The result is the same as if you uses non-fractional adaptive vsync with the display set to 120Hz.

Non-adaptive 1/2 vsync on the other hand will always vsync at a rate of 120Hz. So the game will behave as if was running at 120Hz with normal vsync.

The benefit of fractional vsync is that you still get the faster scanout rate by leaving the display at its highest refresh rate.

Thanks, makes sense, it's just about whether you'll get tearing or no when below threshold, and both fractional types have the benefit of the faster scanout, right?

[RealNC]

Yeah. That means vsync backpressure kicks-in since there's no FPS limit to prevent it. There's no mystery there. This has been known since forever. If you want to prevent vsync backpressure lag, cap your FPS to something that's at least a little lower than the vsync rate.

I mean, it was known about <60FPS@60Hz and 60@60Hz.

The 60>@60Hz is new (if you discount other possible queue buffers due to (U)LLM/Reflex) to me at least?

I've never seen anybody say "you should always at the very least cap FPS to Hz when using Vsync".

It's better to be slightly below, yes. It's not guaranteed to reduce lag if it's exactly the same as the refresh rate.

In your case, you'd need to look at the exact refresh rate your monitor is using. Try this:

https://testufo.com/refreshrate

What does it detect? Most displays are around 59.94Hz. So capping to 60 is not optimal. It can work, but the frame buffers will start to fill up over time until a few minutes later you reach the backpressure point. This can get cleared again if there's a frame time spike, which will clear the buffers and then the cycle begins anew where at first there's no lag but it comes back slowly over time.

An example is adaptive 1/2 vsync with the display set to 240Hz . This will vsync at a rate of 120Hz, unless FPS falls below 120, in which case it disables vsync. The result is the same as if you uses non-fractional adaptive vsync with the display set to 120Hz.

Non-adaptive 1/2 vsync on the other hand will always vsync at a rate of 120Hz. So the game will behave as if was running at 120Hz with normal vsync.

The benefit of fractional vsync is that you still get the faster scanout rate by leaving the display at its highest refresh rate.

Thanks, makes sense, it's just about whether you'll get tearing or no when below threshold, and both fractional types have the benefit of the faster scanout, right?

Yes. 1/4 vsync at 240Hz will have a bit less lag overall than non-fractional vsync at 60Hz. Both will have the typical vsync backpressure lag when not FPS capped. The only difference is the faster scanout speed with 1/4 vsync at 240Hz. The latency difference is relatively small though. In this case it's an improvement of 6.25ms at the center of the screen, and 12.5ms at the bottom.

[2mg]

It's better to be slightly below, yes. It's not guaranteed to reduce lag if it's exactly the same as the refresh rate.

In your case, you'd need to look at the exact refresh rate your monitor is using. Try this:

https://testufo.com/refreshrate

What does it detect? Most displays are around 59.94Hz. So capping to 60 is not optimal. It can work, but the frame buffers will start to fill up over time until a few minutes later you reach the backpressure point. This can get cleared again if there's a frame time spike, which will clear the buffers and then the cycle begins anew where at first there's no lag but it comes back slowly over time.

My old 60hz screen is retired, but I've tried my current one, it's exactly 60.000Hz in 60Hz mode.
Oddly, it's 239.995 in 240Hz mode (should I "fix" this with a custom resolution, or with CRU?).
I've been using this one to test Talos Principle 2, to check <60, 60, 60> limiter + vsync, it's exactly like the same on the old monitor regarding PCL (av).

I get what you're insinuating, but this is a much simpler test which can show PCL (av) to almost double if you don't at least limit FPS=Hz.


People usually associate Vsync ON to mean "also turn FPS limiter to Hz".

That's kinda what I'm trying to demistify here, because both BM Wukong nor TP2 don't actually automatically set their limiters to FPS=Hz, nor do they do it in Windowed mode that is usually automatically Vsync'd (sure, these days you can have tearing even in Windowed mode, but that's another story about DWM).

And that then causes PCL (av) to spike, which I don't get really why is happening in the first place, if you don't set the manually.

And I believe it's worthy of Blurbusters' attention, I know there are ways of low lag Vsync, but maybe having a disclaimer about some games potentially having even more lag if you don't at the very least limit FPS=Hz when using Vsync.

[RealNC]

My old 60hz screen is retired, but I've tried my current one, it's exactly 60.000Hz in 60Hz mode.
Oddly, it's 239.995 in 240Hz mode (should I "fix" this with a custom resolution, or with CRU?).

There's nothing to fix. 240Hz means 4.16667ms display interval, while 239.995 means 4.16675ms. One number is not better than the other.

I get what you're insinuating, but this is a much simpler test which can show PCL (av) to almost double if you don't at least limit FPS=Hz.

We already knew that. It's just that in your case, a 60FPS limit helps because your screen happens to be 60.0Hz, not 59.94Hz.

That's kinda what I'm trying to demistify here, because both BM Wukong nor TP2 don't actually automatically set their limiters to FPS=Hz, nor do they do it in Windowed mode that is usually automatically Vsync'd (sure, these days you can have tearing even in Windowed mode, but that's another story about DWM).

And that then causes PCL (av) to spike, which I don't get really why is happening in the first place, if you don't set the manually.

And I believe it's worthy of Blurbusters' attention, I know there are ways of low lag Vsync, but maybe having a disclaimer about some games potentially having even more lag if you don't at the very least limit FPS=Hz when using Vsync.

This is all known already. There's nothing to demystify here. Vsync has a large latency penalty unless you cap your FPS to something that prevents vsync from filling the frame buffers. That's all there is to it. Without VRR, the best tools for this job are Special K or RTSS. In-game limiters have jitter and might not always work for exact caps.

[2mg]

There's nothing to fix. 240Hz means 4.16667ms display interval, while 239.995 means 4.16675ms. One number is not better than the other.

We already knew that. It's just that in your case, a 60FPS limit helps because your screen happens to be 60.0Hz, not 59.94Hz.

This is all known already. There's nothing to demystify here. Vsync has a large latency penalty unless you cap your FPS to something that prevents vsync from filling the frame buffers. That's all there is to it. Without VRR, the best tools for this job are Special K or RTSS. In-game limiters have jitter and might not always work for exact caps.

Guess it's my bad I didn't know about limiting FPS to refresh rate is essential - at least if you are not doing advanced Vsync lag reduction techniques when using non-VRR Vsync.

Still, that means if FPS>Hz with Vsync, there's an extra buffer somewhere that gets filled?