Should you cap more frames beneath monitor refresh with G-Sync when the monitor is high refresh?

For example, I noticed Nvidia defaults their FPS limiter in the driver to 58. I expect because with G-Sync you want to cap 2 FPS beneath monitor refresh as shown here by BlurBusters: https://blurbusters.com/gsync/gsync101-input-lag-tests-and-settings/5/
To be on the safe side they recommend -3 FPS but note they're using a 60 Hz G-Sync panel for this test.

Now I also noticed Nvidia caps FPS to about 138 or 139 when using Ultra Low Latency mode in conjunction with G-Sync on my 144 Hz monitor -- they do this automatically. I don't own a 60 Hz G-Sync panel but I suspect they'd cap that one automatically to the 58 value so they seem to be increasing the number of frames beneath monitor refresh as the monitor refresh value gets higher.

So, notice 2 frames out of 60 is about 3.333% and 6 frames over 144 is 4.17%. 4.17% would be equivalent percentage-wise to capping 2.5 frames beneath 60 which would fall inline with BlurBuster's recommendation above seems to me. Could this be because the higher refresh you go the smaller frametimes become so at 60 Hz 2 frames is quite a bit frametime wise compared with 2 frames at 144 Hz? Could the G-Sync module need a bit of frametime buffer time to work right? I could be misremembering but I thought I remembered BattleNonSense at one point saying his 165 Hz G-Sync monitor he had to cap 5 frames beneath or some such in his own testing to avoid the g-sync ceiling, something like that, but I'd have to go dig that video up again, I could be misremembering precisely what he'd said.

Basically I'm wondering if with higher and higher refresh rate G-Sync monitors if you need to progressively cap more and more frames beneath refresh to avoid additional input latency. Thanks!

 

@RealNC Sorry to bother you, but I wondered if you'd know about this since it's G-Sync related. No worries if not of course, thanks!

 

Here: https://forums.blurbusters.com/viewtopic.php?f=10&t=7522&p=67283&hilit=cap#p67283

you have an answer to your question by the Chief himself

 

Thanks mate, readin it now

EDIT: Awesome, so for 144 Hz G-Sync -4 you should be in the clear most likely/most of the time.

I imagine Nvidia’s auto capping for g-sync users with ULLM is on the conservative side and goes a bit beyond what’s necessary then (-2 for 60 and -6 for 144).

 

If I enable V-Sync on games that have Nvidia Reflex, my FPS is capped at 171 FPS on my 180hz monitor (LG 27GP850-B)

 

No idea how those capping values were chosen. Maybe to have some tolerance in order to ensure they work across all setups out there.

When I look at the time headroom of each cap, it gets larger the higher the display Hz gets. 58FPS on 60Hz is 33ms headroom. 138FPS on 144Hz is 42ms. 171FPS on 180Hz is 50ms. Looks like the higher the refresh rate gets, the higher the headroom nvidia has chosen gets.

 

Thanks, yeah, it seems that way -- at first I figured they were perhaps doing a percentage approximation to scale with higher refreshes or some such (2 frames at 60 is 3.33% of the refresh, but 6 frames at 144 is 4.17% and so on), but doesn't look that way as the values don't really line up/the number of frames capped below increases with higher refresh panels as you say.

 

@RealNC
If cap is set to minimum fps for a game, lets say 65fps, would it be best to set a G-sync monitor to 120, 144 or 240Hz?

 

Highest Hz is usually best. You get the lowest possible input lag that way. Even at 65FPS, you still get the frame scanout speed of 240Hz (4.2ms). At 120Hz you'd get a scanout speed of 8.3ms instead.

So even if you ever want to play 60FPS-locked games, a high-Hz g-sync monitor is still better than 60Hz.

 

Thanks, so I understood an old discussion you had with Chief correct then, question was in regards to something else but the part of what was written regarding scanouts indicated that more Hz is less lag even with low frame caps.
Just wanted to be sure that what it seems to me like in games isn't placebo. 240Hz with G-sync + V-Sync it is.

 

Just something additional I noticed here in Jorimt's 60 Hz G-Sync panel tests -- you can see that the latency difference between 59 and 58 is "almost" identical with no meaningful observable change below that.

Given that's so, perhaps the "optimal" spot might actually be around 1.5 fps beneath 60. 1.5/60 = 0.025 so if we're scaling by percentage that would be 3.6 fps beneath refresh on a 144 Hz panel which would round to 4 frames.

Maybe not the best way to go about it (and again i'm not really sure how Nvidia is making their FPS capping decisions for their automated approach in conjunction with ULLM since they seem to do 2 fps beneath at 60 hz and 6 fps beneath at 144 hz which going by percentage would be 5 frames not 6 at 144), but that would fall more inline with blur buter's chief's own recommendations for doing ~4 with 144 Hz.

I dunno I'm just trying to find a standard rule/pattern for all this. Might be going down the wrong paths ¯\_(ツ)_/¯ In any case it seems to me for 144 Hz, 4 fps beneath would be a reasonable value or maybe 6 if you really want to be on the safe side (at that point you're matching what Nvidia does automatically with ULLM ON and 6 would be percentage-wise the same as doing 2.5 fps beneath at 60 converted to 144). Percentage matching might be the wrong way to go for this though since that's not what Nvidia seems to be doing for their auto-capping in conjunction with G-Sync + ULLM.

My G-Sync 144 Hz panel isn't the nicest and is somewhat old by this point (Dell branded 1440p monitor) so I might opt to match Nvidia's recommended cap to be on the safe side. A larger cap might also help somewhat if you're using a less accurate/more erratic fps limiter for sake of further reduce input latency such as Overwatch's built in limiter (though in fairness I'm not certain how much less accurate than RTSS / Nvidia's it would be -- it does seem to spike upwards by 1-2 frames occasionally when dips occur has been my observation).

 

The "optimal spot" depends on the number of buffers a game creates in its swap chain when running vsynced. I don't think that it changes much between refresh rates but it is likely that going with more frames between a cap and a refresh takes care of cases where a game creates some big number of buffers for some reason - and when you're running at 144+ you can spend a dozen of frames for this since the refresh is high enough for that to not matter much. At 60 though losing ~10 fps will be noticeable in performance even if it would take care of these weird cases.

 

Thanks, that's good to know

 

So what if max ref is 144 put it on 140?I always did -3 max refresh rate.

 

Depends on what you want.

For the least amount of input lag, disable V-Sync and G-Sync and just push it up to 300 (or at least double the amount of Hz in fps), this is for competitive fps gaming, take note that some games might have an internal cap, having max at 300 should be fine regardless though even for those games.
If you don't have a mouse that can tag along and if you're not even close to a pro level player it might be pointless though.
Downside of this is that your GPU will be working harder when it can push all those frames, which means more heat.

For the least amount of stutter and tearing, enable V-Sync and G-sync and set the cap to what works best for the game, 142fps and downwards, some older games run best with a cap as low as 60fps.
If you dislike large jumps or want to keep the card cool you can also cap it to minimum fps besides having both V-Sync as G-Sync enabled and keep Hz at max.

If you dont have a G-Sync / G-Sync compatible monitor, take your time and test with values from 142fps and 1 below at a time, if you got some games that run best at way lower fps consider to go down in Hz just for those games to avoid tearing.

 

Id rather just slap 141 fps limit with g-sync/v--sync(off ingame) and be done with it for every game.Cant be arsed to manually adjust per game.

Something you said made me think of this issue I have mainly on 1440 165hz :/

Gets greatly reduced or straight up goes away on 1080p or if I wanted to reduce the issue on 1440p putting an fps cap of 120 helped on my 1440p 165hz monitor.
The only thing I can think off is I should probably get a better cpu then my 9900k to increase my average lower frames and 1% frametime.
The higher the refresh rate the better your cpu needs to be right?and from what I'm seeing my 9900k creeps into almost not being enough territory in some games when playing at 1440p/165hz

 

Well, an old game like Skyrim as example doesn't play nice with to much over 60fps, so just keep in mind that some of them older games might introduce issues with high fps.

Don't think that the CPU cares about refresh rates, I don't see mine working any harder @1440p going between 60, 120, 144 and 240.
There might be some parts of them game effects that uses the CPU for calculation, so adjusting them down will most probably help, still doubt that a 9900k struggles with it though.
Would help if you had something showing CPU usage in the video, although I still doubt that there is any significant performance drop with any of those games caused by the CPU.
Only difference I can find is PCIE3 and PCIE4 and minimum fps, nothing that should make it stutter like in your clip.

Have you tried 3080 FE reference values for frequency and power limit instead of dropping the voltage down?
Had an issue with my ASUS 2080S some drivers ago, dropping down to FE values did the trick, I could probably have been tweaking up one at a time from there but cba.

EDIT: GPU usage gets close to 100% sometimes, maybe try to tone some effects down a notch, like shadows and ambient occlusion and see if that makes a difference, although I would think that a 3080 should be able to handle it with ease.

 

I tried at one point reducing to 1900mhz and unvervolting a bit and barely made a difference.Even increasing power limit to 120%

https://drive.google.com/file/d/1H0uWRhjdVFRiamwPv_o5Q9vMjWC2N71j/view?usp=sharing
Did some logging while the issue was going on on warhammer totalwar2.I see nothing wierd cpu usage wise either.

I tried running the game just now with cpu usage showing in msi afterburner and 1 core is like at 80-90% and the rest are all over the place so nothing "odd".

Also yes toning down some settings did reduce the issue on 1440p.(3d portaits and effects to medium instead oh high) But like you said a 3080 should handle it no prob.
On 1080p I don't have to mess with anything :/

Think its just some games/the driver that could be the cause if by reducing some settings the issue gets better?Still tho its a 3080 goddamn.Maybe its an issue with 3080's not scaling properly on 1440p?Over the months I've had a couple of ppl telling me 3080 don't scale well on 1440p and that could be the cause.

 

Not quite; 59 FPS had a higher enough maximum to signify -1 FPS was not sufficient in keeping the framerate within the G-SYNC range at all times. I.E. there were frametime variances with that limit that caused the framerate to exceed the refresh rate when compared directly to -2 and -10 FPS, where the difference in maximums between the latter two could be accounted for by margin-of-error and test-to-test variance.

As for the recommended limit across refresh rates, there technically isn't "one;" it can depend.

That said, the blanket recommended minimum starting point is -3 FPS, which is usually enough for your typical in-game or external limiter to keep the framerate within the refresh rate during VRR operation, exceptions notwithstanding (the newer CoDs' IW engine-level limiter, for instance; it's sporadic, unreliable, and doesn't actually limit itself to what you set it to, at least when I last tested it).

It should also be made clear that this minimum -3 FPS recommended limit is only in place to keep the framerate within the refresh rate to keep G-SYNC engaged, and does not factor in things such as GPU-limited scenarios (where the average framerate can't always reach the selected FPS limit, which can cause render queue-induced input lag separate of V-SYNC-induced input lag), achievable average framerate in non-GPU-limited scenarios, general frametime consistency, or engine limitations (where things such as the physics system breaks above certain framerates), which must be addressed on a per system, per game basis.

 

Something similar with path of righteus game.Gpu isnt even maxed.Gets better when going away from the city...Could be the unity engine tbh.Damn alot of games run like trash these days.
Update:yea my bro has some similar camera micro hitching too.Has to be the game.