Minimum Visible Frametime Variance

[brogers]

While playing Pragmata, I noticed that in certain areas there was very slight occasional jitter whenever I moved the camera to reveal whatever geometry/reflection was causing it. Using capframeX, I saw that while my game was capped to 100fps with RTSS, there was still variance (spiking to 13ms). And it got me thinking, are my eyes really sensitive enough to see a 3ms spike?

I have G-SYNC enabled, and with all the recommended settings. 3ms seems like such an insanely small amount of time that I almost don't believe that my eyes can register it. Granted, when testing this I was specifically looking for it, and using my controller to very slowly pan around in circles. Has anyone else gotten sensitive enough to see spikes as small as this? Or is there something with VRR/my monitor that I need to look in to?

[Chief Blur Buster]

While playing Pragmata, I noticed that in certain areas there was very slight occasional jitter whenever I moved the camera to reveal whatever geometry/reflection was causing it. Using capframeX, I saw that while my game was capped to 100fps with RTSS, there was still variance (spiking to 13ms). And it got me thinking, are my eyes really sensitive enough to see a 3ms spike?

Yes, it is possible to see 3ms frametime spikes when it's a large % of refreshtime (unstrobed) or a large % of pulsewidth (strobed).

A 3ms spike is way easier to see at 500fps OLED (2.5ms frametimes) than 100fps LCD (10ms frametimes + additional GtG pixel response ghosting).

I've seen sub-3ms spikes with my eyes before on my 720 Hz OLED, it's a real thing. Very faint, but when framerate=Hz perfectly framepaced, and you're paying close attention to it. As a general rule of thumb, a frametime spike that briefly doubles MPRT, is like a camera shutter suddenly going from 1/720sec to ~3/720sec and then back to 1/720sec. So it manifests itself as a very faint, very brief flicker of extra motion blur in an almost-invisible jitter.

But if it's low frequency enough (e.g. 2 or 3 times per second), and it's been glassfloor framepacing otherwise (like perfect framerate=Hz VSYNC ON), then you can tell that something wasn't perfectly smooth.

In short, frametime spikes near and beyond MPRT is the rough threshold, combined with FAST smooth motion speed (like a strafe in front of a high-detail wall). During VRR, a 3ms spike at 16ms MPRT almost always is never noticeable, but a 3ms spike at 1.4ms MPRT is more noticeable.

Also, mouse jitter will hide 3ms frametime spikes unless you're using high DPI (1600-3200dpi) and low in-game sensitivity. I find that playing FPS games feels much smoother at 1600-3200dpi with Windows Control Panel mouse sensitivity set way low (to slow down mouse cursor) and H/V sensitivity set low. That eliminates mouse jitter, amplifying the visibility of 3-to-10ms frametime spikes on a fast-GtG display like a 500+ Hz OLED.

What's your panel and panel settings?

[RealNC]

While playing Pragmata, I noticed that in certain areas there was very slight occasional jitter whenever I moved the camera to reveal whatever geometry/reflection was causing it. Using capframeX, I saw that while my game was capped to 100fps with RTSS, there was still variance (spiking to 13ms). And it got me thinking, are my eyes really sensitive enough to see a 3ms spike?

A thing I noticed in the last few years is that capping FPS can produce these spikes that result in microstutters/jitters because render latency is 0. Just as a test, what happens if you force fractional vsync in nvidia profile inspector for this game to a value that gives you sub-100FPS? On 240Hz, that would be 1/3 vsync (80FPS) for example. Don't cap your FPS, let it hit the vsync ceiling. (Fractional vsync will also prevent g-sync from working, so no need to disable g-sync manually.)

Quite a few games become perfectly smooth when I do that, because with vsync and no cap, render latency is higher than the frame time spikes, and as a result they don't become visible on screen (they still show up on the frame time graph, but what you see on screen is perfectly smooth.) Latency is going to increase quite a bit when doing this, obviously.

[brogers]

What's your panel and panel settings?

AW2724DM, the recommended G-Sync settings: G-sync on, v-sync on in nvcp, off in game. My monitor's HZ is 165 but I have a global nvcp cap of 144. Like I said, I was getting pretty distracted so I was just standing still and spinning the camera and I could see it. But during normal gameplay, like turning around with the analog stick, I occasionally catch it out of the corner of my eye, the edges of background geometry jump ahead very slightly. As far as Pragmata goes, it seems like the frametime variance is unavoidable unless lowering a bunch of settings and disabling RT/PT. Which just irked me because I hadn't seen any reviewer or benchmark video notice or comment on it, even when it's plainly visible on a graph. I guess I should just stop expecting perfect frametimes from modern DX12 games, haha. Hopefully the longer I play I'll learn to ignore it.

[Chief Blur Buster]

What's your panel and panel settings?

AW2724DM, the recommended G-Sync settings: G-sync on, v-sync on in nvcp, off in game. My monitor's HZ is 165 but I have a global nvcp cap of 144.

At 144fps + GSYNC, a 3ms frametime spike is fairly hard to see. Mouse jitter at 800dpi will usually be more dominant than a 3ms frametime spike.

Being that said, if you're running 3200dpi + low ingame sensitivity + good mousepad, and you have very fast GtG, it is still possible to detect a resident "unsmoothness" in half-MPRT frametime spikes. But it's very hard/subtle. In some games, gametime spikes (game frametime timestamp & world motion calculations) may be bigger than frametime spikes, creating motion divergences that are bigger than frametime spikes. So you may be seeing more than 3ms divergences, despite RTSS claiming 3ms.

Also power management can jitter frame presentation to photons a bit, so if you're using default settings (e.g. Balanced), a 144fps cap might push GPU utilization low enough to cause power management jittering worse than 3ms frametime jittering. This is fixed by adjusting GPU/CPU to stay at maximum performance without power management.

Also, optimizing mouse DPI will help fix the jitterfeel of mouselooks/pans/scrolls/etc. (Example: 3200dpi + ultra low sensitivity in game + lowered sensitivity in Windows Control Panel, etc) since low mouse DPI adds continuous jitterfeel to mouselooks on high-Hz displays (mathematically can be perpetually more than 3ms worth of randomized position:position jitter than the frametime graph suggests).

Now, that said I noticed one thing, 144fps cap = 1/144sec = 6.94ms. And you mentioned a 13ms frametime. So, that would be (~13ms - ~7ms) = 6ms frametime spike if you were running flat out at 144fps and got the spike. More noticeable - it's literally a full framedrop.

Still, at only 144fps on 165Hz, a 3ms frametime spike is barely half MPRT (at best) and should not easily be too visible (still visible, but not very visible). Given motion blur *and* LCD GtG will hide jitter much smaller than the resulting display motion blur.

If you're seeing noticeable stuttering/jittering while utilities show only 3ms frametime spikes -- then I have to suspect other factors like gametime timestamp variances that significantly diverge from frametime (I consider this to be a bug, and suboptimal game optimization).

Trying a different cap utility like RTSS might produce noticeably smoother results than an NVCP cap, so you might want to test that too. Also verify that VRR indeed, is of course, turned on, in your display.

[brogers]

Now, that said I noticed one thing, 144fps cap = 1/144sec = 6.94ms. And you mentioned a 13ms frametime. So, that would be (~13ms - ~7ms) = 6ms frametime spike if you were running flat out at 144fps and got the spike. More noticeable - it's literally a full framedrop.

In the specific situation I was getting the 3ms spike I was capped to 100fps with RTSS Front Edge Sync, and I was able to see jitters when I was testing by slowly spinning my camera with the analog stick. Which probably is what made them so noticeable. I guess RTSS's limiter might be interacting poorly with the game's timing and causing it to look larger than it really was.

I tested a similar scenario in Cyberpunk 2077, and if I spun the camera slowly and really focused, I could see extremely subtle jitters, but not as much as Pragmata's. Honestly, I think controller analog movement might also contribute to it, since I'm using a 2.4ghz wireless controller and probably not keeping the same acceleration exactly as I pan. Probably nothing I would notice in normal gameplay unless it was during a similar slow pan like a cutscene.

I think I should just stop worrying about and testing for perfect framepacing until I actually notice it in gameplay lol. The spikes I have noticed in gameplay are in the game's hub area when the robot girl's hair enters and leaves the frame, that causes a noticeable spike to around 15ms. The spike shows up if I use the ingame 120 fps limiter, but strangely, if I use RTSS Front Edge Sync to cap to 100, the frametime graph remains completely flat although I do still feel a minor stutter (more subtle than at 120fps). That kind of thing made me worry about something with G-SYNC being off, but I tested with my second monitor disabled and noticed no difference.

[Chief Blur Buster]

There's some black box jitters that can happen. I have been seeing more and more of it happen over time, and I may need to invent new tests to measure this. Work is ongoing to measure jitter not detectable by software utilities.

[brogers]

Since it only seems to happen in games that use ray tracing or very heavy GPU utilization and goes away when I lower my gpu usage, my theory is that the jitter occurs in whatever part of the pipeline the frametime graph isn't measuring, but I'm not knowledgeable enough about that stuff.

The attached image is the frametime graph from CapFrameX during a test where I capped at 60 and tried to max out my GPU with ray tracing and DLAA. Obviously, some spikes show up and those were visible, but the yellow line is the GPUBusy time as I repeatedly rounded a corner into a zone with a lot of reflections, and I feel a slight jutter there with no frametime spike.

But generally, is a spike from ~10ms to ~13/14ms human detectable? It's very subtle and I probably wouldn't notice it if I wasn't actively looking and testing for it, but I just wanted to make sure my monitor isn't somehow exacerbating them.

[brogers]

Did some more tests on a different game, Dragon Quest 7 Reimagined because I remembered that game had jitter that didn't go away when capping. If I cap to 60fps and then run around, certain spots feel very jittery. When capped to 144, its smoother but there are traversal spikes to 12ms. I enabled "msBetweenDisplayChange" in CapFrameX, and the graph looks crazy, with those values bouncing between 9 and 26ms. Is this normal or does this indicate something is wrong with my monitor setup that is delaying frames? I might test this more tomorrow with G-SYNC off.

[brogers]

So, I tested with G-SYNC off, and the stutters are still there (and worse because of VSYNC), and surprisingly, they show up on the frametime graph. So I don't know, I guess when I cap with RTSS something about how it measures frametimes is hiding them. It's at least good to know nothing is wrong with my setup and it's just the game.