Page 22 of 33

Re: G-Sync 101 w/Chart (WIP)

Posted: 15 Apr 2017, 15:51
by MT_
RealNC wrote:
MT_ wrote:Doubt this will take me anywhere near Windows 10 though, I think ill avoid an experimental OS for at least until support runs out for W7 in 2020 :o
I'm using it and have the "Defer Feature Updates" enabled in Windows Update. This puts you on the stable channel of updates, and blocks all experimental stuff from being installed (like this "Creators Update"), and only gets me security updates. The feature updates will then only be delivered months down the line (about 4 or 5 months later), once they're stable. You need at least W10 Pro though for that option to be available...

Never had an issue yet :-)
I actually used Windows 10 Anniversary LTSB for a while. The added input delay in Windows just felt weird to me due to forced DWM, I'd much prefer 'classic' theme where its old skool drawn instantly.

Furthermore my fps felt more fluctuating compared to W7, meaning I had to drop in-game fps lower to avoid G-sync tearing. I'm not sure how or why, but perhaps is is due to the 'tickless' kernel in W10...

Guess one day I will try it again, perhaps when MS decides to just throw in a huge 'bug fixes and system wide improvement patch' without any additional junk and features. When that happens MS probably finished Windows 11/12 even when they said there will never be any new windows version ;)

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 13:11
by Chief Blur Buster
RealNC wrote:I'm using it and have the "Defer Feature Updates" enabled in Windows Update.
One note.

Make sure you get Windows 10 Professional [extra $$$] to be able to do this.

Windows 10 Home has forced feature updates.

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 13:20
by Haste
New video from Battle(non)sense about G-Sync on laptops:

phpBB [video]

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 16:15
by Bouttime
Haste wrote:New video from Battle(non)sense about G-Sync on laptops:

phpBB [video]
Thanks......Very good video. Pretty much the same findings as Jorimt except the frame rate cap is slight lower and he suggests turning off V-Sync. I figure there's no need as it won't kick in if the FPS is capped below the displays max refresh rate.

Sent from my Nexus 6 using Tapatalk

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 16:17
by RealNC
Good video. But he got the vsync off/on case wrong. The best default when using g-sync is vsync on. It doesn't cost you latency, and it makes sure there's no tearing when there's frame time spikes, and it should allow you to cap more near the refresh rate (73 or 72) without starting to tear.

Unless it works differently on laptops...

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 17:14
by jorimt
Overall, a nice video, very thorough. And yes, I don't doubt there are subtle differences between laptop and desktop G-SYNC implementations. I will be including a disclaimer in my final G-SYNC 101 article, stating that all of my testing applies to desktop G-SYNC implementation only, and that laptop performance and functionality may vary.

That said however, most of the differences I saw were mainly refresh rate and specs-based. Frametime delivery is what dictates how much tearing is see in the lower part of the screen with G-SYNC + v-sync off. It is interesting that the G-SYNC laptops required a lower than -2 framerate limit. It appears he only decreased the framerate more then -2 via the limiter to reduce the bottom tearing seen with G-SYNC + v-sync off, which is due to frametime fluctuations, the reason you need to enable v-sync with G-SYNC in the first place. I'm wondering what the input latency numbers would have been if he had tested G-SYNC + v-sync on + -2 frames below max refresh only.

Finally, his tests showed there was virtually no input latency difference between G-SYNC + v-sync off and G-SYNC + v-sync on with the same framerate limit on the laptops, so I'm also not sure why he ultimately recommended G-SYNC + v-sync off + framerate limit.

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 17:22
by Bouttime
It's the Framerate cap for 75hz IPS G-SYNC laptops I'm interested in. I've capped mine using RTSS at 73 FPS as recommended here in this thread. But this video suggests the game would speed up and slow down as G-SYNC was engaged/disengaged. So tomorrow I'll lower my cap to 70 FPS to see if it's there's a difference.

Sent from my Nexus 6 using Tapatalk

Re: G-Sync 101 w/Chart (WIP)

Posted: 16 Apr 2017, 19:20
by RealNC
jorimt wrote:It is interesting that the G-SYNC laptops required a lower than -2 framerate limit.
I'm on desktop, and I get bottom tearing even at 120FPS on 144Hz in CS:GO using fps_max. With RTSS it works fine. No tearing up to 142Hz.

I said in the past (before I had a G-Sync monitor, even) that CS:GO's frame limiter isn't accurate. Now this seems to confirm it :mrgreen:

RTSS is accurate down to a small fraction of a millisecond. Without G-Sync, it can keep the tearline completely fixed in place if the cap matches the refresh perfectly.

Re: G-Sync 101 w/Chart (WIP)

Posted: 17 Apr 2017, 09:38
by Chief Blur Buster
jorimt wrote:Finally, his tests showed there was virtually no input latency difference between G-SYNC + v-sync off and G-SYNC + v-sync on with the same framerate limit on the laptops, so I'm also not sure why he ultimately recommended G-SYNC + v-sync off + framerate limit.
Probably doesn't matter due to the frame cap preventing it from mattering -- but most people trust VSYNC OFF not to have input lag.

BTW, for a completely stationary tearline (not a single pixel above/below), you need an accuracy of 1/(horizontal scanrate)th of a second. At 1920x1080 120Hz, that would be 1/135,000th of a second accuracy! Even if the tearline vibrates up and down in an amplitude of 5 pixels, that's still 5/135,000th of a second accuracy. Damn impressive.
RealNC wrote:Good video. But he got the vsync off/on case wrong. The best default when using g-sync is vsync on. It doesn't cost you latency, and it makes sure there's no tearing when there's frame time spikes, and it should allow you to cap more near the refresh rate (73 or 72) without starting to tear.
Unless microseconds matter. (Which it does not...even Olympics races aren't that close!)

The position of tearline moves downwards by 1 pixel during every 1/(horizontal scanrate)th of a second. The closer the tearline to the bottom edge of the screen, the less lag penalty will occur if you had waited before the tearline went off (that you had VSYNC ON ).

So from this, here's an example: Say, a tearline occurs 27 pixels up from the bottom edge of the screen. The lag difference between a bottom-edge tearline 27 pixels from the bottom of 1920x1080@120Hz (135KHz scanrate) would mean using VSYNC ON instead would add an input lag delay of 27/135,000th of a second if you chose to use GSYNC+VSYNC-ON instead of GSYNC+VSYNC-OFF.

Maybe a little bit more, given the blanking interval consists of front/back porches and sync, but with a common Vertical Total of 1125 (1080 visible, 45 blanking interval), that's only another 45/135,000th of a second extra at the most (on top of 27/135,000th). Still tiny. But even VRR also has a mandatory minimum pause between refresh cycles, so all things considered, let's exclude this for simplicity's sake. Even if it was included, it's a tiny fraction of one millisecond.

Yes, nitpicking. But of interest for those Advanced Users who know more details behind the scenes of "what-if-VSYNC-was-ON-instead" during tearline-occurence situations. Mathematically, it's possible to determine how much input lag you saved in this "WHAT-IF" exercise, simply by knowing where the position of the tearline was.

TL;DR: If tearline occurs near the bottom edge of the screen, the lag difference of going VSYNC ON is less than a millisecond in this particular situation.

Re: G-Sync 101 w/Chart (WIP)

Posted: 17 Apr 2017, 10:22
by Chief Blur Buster
Geek to geek, mentioning the case of 240Hz GSYNC -- and the need to cap below GSYNC limit...

...I successfully convinced the RTSS author to extend GUI frame limit to 300 in his next version. ;)

http://forums.guru3d.com/showthread.php ... ost5421412