NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Everything about latency. Tips, testing methods, mouse lag, display lag, game engine lag, network lag, whole input lag chain, VSYNC OFF vs VSYNC ON, and more! Input Lag Articles on Blur Busters.
ELK
Posts: 125
Joined: 18 Dec 2015, 02:51

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by ELK » 16 May 2021, 03:10

jormit have you got the input latency testing equipment yet? I'd like to see the testing.

I think it'd be a good idea to start a thread to compile every input lag trick to help speed up the testing and have all the results in 1 spot

User avatar
jorimt
Posts: 1513
Joined: 04 Nov 2016, 10:44
Location: USA

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by jorimt » 16 May 2021, 09:31

ELK wrote:
16 May 2021, 03:10
jormit have you got the input latency testing equipment yet? I'd like to see the testing.

I think it'd be a good idea to start a thread to compile every input lag trick to help speed up the testing and have all the results in 1 spot
Once you ensure you have a high refresh rate monitor and input devices with low native delay, and you prevent both V-SYNC and render queue input lag (which I and others have already extensively tested and shared the results of), you start hitting severely diminishing returns that are hard to prove/show one way or the other, even with good test equipment, at which point it tends to get into the subjective (sometimes even the placebo and/or OCD) territory very quickly.

If you have a specific question that you've potentially missed my answer on, however, I'll try to repeat it or direct you toward it.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: Acer Predator XB271HU / LG 48CX OS: Windows 10 MB: ASUS ROG Maximus X Hero CPU: i7-8700k GPU: EVGA RTX 3080 FTW3 UG RAM: 32GB G.SKILL TridentZ @3200MHz

teo
Posts: 2
Joined: 15 Apr 2021, 19:28

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by teo » 16 May 2021, 16:18

has anyone done any testing (or know from theory) why reflex implements its frame cap at the chosen values? I understand that it's relative (~116 for 120hz, 138 for 144hz, 158 for 165hz, 225 for 240hz, etc.), but it's not obvious to me why those are the "right" choices. obviously it's not a simple proportion of the max refresh rate. the only thing that's apparent to me is that each frame cap provides for at least a ~0.25ms "buffer" of frametime between the cap and the max (e.g. the 116 fps frametime is ~.287ms slower than 120hz, while 117 is ~.213ms). the known caps above result in a buffer of 0.268-0.3019ms, and if most* of those caps were even 1 frame higher that buffer would shrink to <0.25ms (i.e. a 139 cap has a frametime 0.2498ms slower than 144, vs a 138 cap's 0.3019). does that suggest that nvidia is confident that the 0.25ms buffer is sufficient to ensure that frame rate stays within the g-sync window no matter the (reflex supported) game engine?

as an aside, I remember battlenonsense mentioning that he'd cap at 232 to stay in range for a 240hz display. assumedly this is for the same reason but I haven't found an explicit discussion on it.

*a 240hz monitor could actually support a 226 cap and maintain a buffer over 0.25ms (at 0.258), but a 227 cap drops it down to 0.238ms

User avatar
Chief Blur Buster
Site Admin
Posts: 9193
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by Chief Blur Buster » 16 May 2021, 20:31

Error margin.

VRR has lowest lag when frametimes are never faster than refreshtimes.

Fluctuating frame rates at 240fps could mean some frames are 1/235sec and others are 1/245sec. The frames that are faster than refreshtimes will be lagged by the Hz limitation of the display for some sync technologies (especially VRR modes).

So capping slightly below max Hz, gives varying frametimes some breathing room below max Hz.

Blur Busters was the world's first website that discovered this principle in year 2014 with www.blurbusters.com/gsync/preview2
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

       To support Blur Busters:
       • Official List of Best Gaming Monitors
       • List of G-SYNC Monitors
       • List of FreeSync Monitors
       • List of Ultrawide Monitors

teo
Posts: 2
Joined: 15 Apr 2021, 19:28

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by teo » 16 May 2021, 21:55

right, I get why it's important to cap the framerate to stay within the g-sync operating window and avoid the ceiling. my question is why nvidia specifically chose those error margins when implementing reflex.

my understanding of your investigations at blur busters is that once you're sufficiently far enough away from the g-sync ceiling, there's no benefit (and eventually, some detriment) to cap fps lower than necessary. that is, if a -2 frame rate cap yields the same input lag measurements as a -10 cap, you can be confident that the -2 cap is sufficient to avoid the ceiling (at least for that game/frame rate limiter).

as refresh/frame rates rise, the absolute frame time is going down. thus, the frame cap necessary to avoid the g-sync ceiling for a 240hz display will be a larger than the cap necessary for a 144hz display (i.e. why reflex chooses a 225 [-15] fps cap for 240hz and a 138 [-6] fps cap for 144hz). I think we're on the same page.

the blur busters approved fps cap to avoid the g-sync ceiling is -3 for 144hz, right? I assume this is because you guys measured that a -2 cap seemed to work fine, and an extra frame should be enough to cover most game engines/rate limiters. what I haven't seen is any input lag measurement data for what frame cap is appropriate for refresh rates >144hz. obviously, nvidia thinks this should be a -6 cap for 144hz and -15 cap for 240hz. chris is seemingly comfortable with a -8 cap for 240hz. I recognize that the frametime difference of 225 fps and 232 fps is a whopping 0.134ms; I'm not worried about that impacting my experience if the nvidia cap is more conservative than it needs to be. I'm curious if anyone in the community has done the measurements to determine "yeah, about 0.25ms of frametime buffer is what you need to be safe" or had any insight as to why nvidia came to the conclusion that their chosen caps are the most appropriate.

User avatar
jorimt
Posts: 1513
Joined: 04 Nov 2016, 10:44
Location: USA

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by jorimt » 17 May 2021, 08:49

teo wrote:
16 May 2021, 21:55
my question is why nvidia specifically chose those error margins when implementing reflex.
Only Nvidia knows.

These offsets pre-date Reflex, and existed with G-SYNC + NVCP V-SYNC + LLM Ultra as well.

My guess is they're being super safe with the margins as the refresh rate increases. That, and the lower the framerate, the lower the chance the GPU usage becomes maxed (render queue, not G-SYNC range related), but that reason seems highly unlikely.

It's also possible that's just how their automated limiting algorithm scales as the refresh rate increases, and they haven't given the higher offsets that occur at higher refresh rates much thought, especially since, as you mentioned, the frametime difference between, say, 237 and 225 FPS at 240Hz is relatively minuscule from an additional input lag standpoint (and most would be hard pressed to notice a fluidity difference between the two either).
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: Acer Predator XB271HU / LG 48CX OS: Windows 10 MB: ASUS ROG Maximus X Hero CPU: i7-8700k GPU: EVGA RTX 3080 FTW3 UG RAM: 32GB G.SKILL TridentZ @3200MHz

User avatar
Chief Blur Buster
Site Admin
Posts: 9193
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by Chief Blur Buster » 17 May 2021, 17:24

teo wrote:
16 May 2021, 21:55
my understanding of your investigations at blur busters is that once you're sufficiently far enough away from the g-sync ceiling, there's no benefit (and eventually, some detriment) to cap fps lower than necessary. that is, if a -2 frame rate cap yields the same input lag measurements as a -10 cap, you can be confident that the -2 cap is sufficient to avoid the ceiling (at least for that game/frame rate limiter).
The common VRR 3fps-below is actually an artificial boilerplate recommendation that is given out frequently for simplicity.

However, in reality, sometimes you need 0.5fps-below and sometimes you need 15fps-below. It depends on the Hz and the erraticness of frametimes. As a general rule of thumb, for GSYNC users, I tend to approximately recommend various number such as:

60Jz = use about 0.5fps-2fps below
144Hz = use about 2-4fps below
240Hz = use about 3-5fps below
360Hz = use about 5-10fps below

These are only examples and can deviate (above/below) for specific games, potentially.

The tighter 0.5fps below is good for 4K 60Hz FreeSync LCDs overclocked to 60.5Hz and capped to 60fps for emulator use.

But yes, the choosing of a bigger cap differential is a "just in case" manoever.

Fortunately, the latency differential of 225fps and 230fps is extremely tiny:
((1/225) - (1/230) = 0.097ms frametime difference = 97 microseconds latency difference.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

       To support Blur Busters:
       • Official List of Best Gaming Monitors
       • List of G-SYNC Monitors
       • List of FreeSync Monitors
       • List of Ultrawide Monitors

Post Reply