New Nvidia driver Gsync support - Gsync Ceilling problems

[RealNC]

What I'm saying is that driver might be doing things we don't know about. Does it stutter or does it not? That's the important thing.

[jorimt]

From what user reports I've seen so far (here and elsewhere), this is my educated (but incomplete and probably very imperfect) guess to the functionality of the current driver-implemented G-SYNC functionality on FreeSync monitors...

1. The V-SYNC option enabled with G-SYNC does not appear to work like it does on a genuine G-SYNC monitor. Instead of providing frametime compensation to prevent partial or complete tearing within the G-SYNC range, it simply reverts to full V-SYNC behavior whenever the framerate drops below the physical limit of the given panel (or apparently even with tiny frametime variances in the upper range), causing repeated and continual start, stop, start, stop stutter effect as the monitor jumps from V-SYNC to VRR and back. Though the effectiveness of G-SYNC + V-SYNC "On" may depend on the given system running the given game on the given monitor (FreeSync range + unsupported/supported monitor + system frametime performance + game compatibility).

2. It's possible there is an increased polling/report rate for updates read from the GPU to the driver, causing more VRR range issues than necessary. This means it's also possible that this driver-only implementation of G-SYNC needs a lower FPS limit to stay within the VRR range when directly compared to standalone G-SYNC and FreeSync. What that number might be is up in the air until actual high speed tests are done.

3. LFC (low framerate compensation) seems non-existent on officially unsupported (especially lower refresh rate) FreeSync monitors, and highly limited on officially supported FreeSync monitors, exacerbating issue #1 and #2 above.

So, to sum up, this appears to be basic, bare-bones, driver-level adaptive sync, which means beyond adjusting the refresh rate to the GPU render rate (and only within each monitor's reported physical VRR refresh rate range), there is no working frametime compensation (V-SYNC option for 100% tear-free VRR) and little to no LFC support.

In other words, on average (and especially with officially unsupported monitors) it seems G-SYNC + V-SYNC "Off" with moments of tearing is the best that can be expected of current functionality. Whether they improve this over time, or "officially" support more monitors with this mode remains to be seen.

[phatty]

What I'm saying is that driver might be doing things we don't know about. Does it stutter or does it not? That's the important thing.

I get what you are saying about the stuttering.

My question still stands.

if I put a FPS cap on overwatch at 220, but I see frame variance on the hardware fps at 240 routinely. Are we looking at a hardware imperfection or a driver problem with VRR?

[Malinkadink]

What I'm saying is that driver might be doing things we don't know about. Does it stutter or does it not? That's the important thing.

I get what you are saying about the stuttering.

My question still stands.

if I put a FPS cap on overwatch at 220, but I see frame variance on the hardware fps at 240 routinely. Are we looking at a hardware imperfection or a driver problem with VRR?

I think this would be best answered if we could compare the behaviors of said monitor with an AMD GPU and then with an Nvidia GPU and the latest driver. If you lock your fps at 237 (3 below cap) on the AMD GPU and the game is reporting 237fps and the monitor is reporting 237hz then everything looks well and good. Trying the same setup but with an Nvidia GPU and you lock your fps at 237 but the monitor is reporting 239/240 which is evidence of being up against the limit and therefore inducing vsync input lag or tearing if vsync is disabled.

That test would indicate a driver problem on Nvidia's side, and i actually believe at this point that it is indeed a driver issue and not a hardware issue. If you have a perfect locked fps at 220 and the monitor is reporting 240hz when the actual FPS is 220, then the driver is sending the monitor the wrong value, in fact if that were the case the monitor would then be refreshing at 240hz while your fps is actually 220 and that mismatch in fps and hz would cause screen tearing.

The GPU is responsible for telling the monitor when it is ready to display frames and what the fps value is so the monitor can match that number and be in sync to make VRR work proprely.

[jorimt]

On this subject, while it doesn't mean the G-SYNC compatible FreeSync monitors aren't jumping out of the VRR range, the built-in refresh rate meters exhibited this spiking to max refresh rate periodically during my G-SYNC 101 tests, and on both a genuine 144Hz and 240Hz G-SYNC monitor paired with an Nvidia GPU.

With a -2 FPS limit, the input lag results showed this behavior didn't affect actual functionality, and the monitors remained in the VRR range regardless.

So what I do know is that this behavior on a G-SYNC monitor is nothing more than a benign, built-in readout issue (be it a glitch, or an OS or driver specific readout interaction difference; what I do recall, is it wasn't doing this before a certain Windows 10 update a couple years back).

As for the FreeSync monitors using the G-SYNC driver update, and their relation to this readout issue, not sure. Would have to be tested.

[Chief Blur Buster]

There are actually over 100 different instantaneous framerates during 100fps, if each frametime are slightly different.

For a 1/100sec time period, instantaneous framerate might be 99.391fps, or 103.155fps. But the next frametime varies, and then that changes the instantaneous framerate.

Plus, repeat-refreshing behaviours can affect things, e.g. framerates going below 30fps can cause doubling a 28fps = 56Hz = "56fps" readout.

So the monitor framerate display algorithm is subject to certain kinds of occasional inaccuracies
(1) Divergences from instantaneous framerate versus average framerate
(2) Divergences caused by automatic repeat-refresh cycles (due to frametimes exceeding VRR min Hz).

[Malinkadink]

On this subject, while it doesn't mean the G-SYNC compatible FreeSync monitors aren't jumping out of the VRR range, the built-in refresh rate meters exhibited this spiking to max refresh rate periodically during my G-SYNC 101 tests, and on both a genuine 144Hz and 240Hz G-SYNC monitor paired with an Nvidia GPU.

With a -2 FPS limit, the input lag results showed this behavior didn't affect actual functionality, and the monitors remained in the VRR range regardless.

So what I do know is that this behavior on a G-SYNC monitor is nothing more than a benign, built-in readout issue (be it a glitch, or an OS or driver specific readout interaction difference; what I do recall, is it wasn't doing this before a certain Windows 10 update a couple years back).

As for the FreeSync monitors using the G-SYNC driver update, and their relation to this readout issue, not sure. Would have to be tested.

Would have been nice if Dell included a refresh meter for the S2417DG, unfortunately i have no way to see what the monitor is actively switching to. I will say that my experience using gsync has never been a bad one though, so long as i have the limit set to avoid vsync lag there are no issues.

However, i have noticed stuttering on my gtx 1080 in the form of significant frametime spikes when playing overwatch on the latest drivers. I play at a locked 162fps and never dip my gpu usage is around 70% at the most but i've lately been noticing slight dips to 150s for a split second and frametimes spike to 30+ms. I went back to the 399.24 driver before the RTX drivers and its much better on those drivers, but still not perfect. At this point its either the game or windows 10 doing something to add to it. I recently reinstalled W10 and updated to 1809, who knows what might have been broken transitioning from 1803. I'll have to do more testing, starting with disabling fullscreen optimizations.

Now that i think about it i've been noticing how much faster alt tabbing has felt recently from running the game in fullscreen, yeah, i'll definitely be disabling fullscreen optmizations globally and see if i notice any changes.

[Notty_PT]

All these inconsistencies are the reason I said the best experience with this driver enabled gsync are:

- capping fps to 120 on a 144hz monitor
- capping fps to 50 on a 60hz

Otherwise I assure everyone will have tearing. Even at 130fps. And if you use Vsync you will have input lag.

Another thing I found: this software Gsync adds input lag.. Just by toggling it OFF and ON on the desktop I immediatly notice it in my mouse. More noticeable in game.

The sad part is that I spent hours testing this with several monitor setups, but I have no way of using science to proof it to other users as I have no tools to measure. All I know is that Im right, but others will not believe in some random dude without proof, wich is completly normal. At least on other forums and youtube already seen people mentioning the 120fps cap and the added input lag. Now need official tests.

[jorimt]

All these inconsistencies are the reason I said the best experience with this driver enabled gsync are:

- capping fps to 120 on a 144hz monitor
- capping fps to 50 on a 60hz

Otherwise I assure everyone will have tearing. Even at 130fps. And if you use Vsync you will have input lag.

FYI, this is normal even on a genuine G-SYNC monitor with G-SYNC + V-SYNC "Off"; if the framerate can be sustained at the max refresh rate, say 144Hz with a 141 FPS cap, for instance, bottom screen tearing (due to frametime variances in the upper range) can only be (in my testing) removed with about a 120 FPS cap. Anything above that will make the tearing in that area return.

So we know that even if the G-SYNC driver implementation on FreeSync monitors was perfected, it will still retain this issue in that scenario.

What is wrong with this driver implementation (at least according to user reports thus far), is that the V-SYNC option appears to be functioning as plain old V-SYNC (instead of a frametime compensation mechanism), and LFC isn't working, or isn't working well, even on officially supported monitors.

Granted, software LFC is hard to get right. In fact, all of this is hard to get right solely through software, especially with the nearly infinite variations of setup and monitor combinations this feature has to run with.

Another thing I found: this software Gsync adds input lag.. Just by toggling it OFF and ON on the desktop I immediatly notice it in my mouse. More noticeable in game.

Very possible, but you're right, it would have to be tested.

Again, I don't have access to a FreeSync display, but the Chief did say he may be doing some basic high speed tests with this driver feature in the coming months.

Suffice to say, while it's a fun little novelty to play with if you already have an Nvidia GPU paired with a FreeSync monitor, this driver feature isn't currently something that should be the motivation to buy a FreeSync monitor to pair with your existing Nvidia GPU to save on the cost of a genuine, standalone G-SYNC or FreeSync setup...at least yet; time will tell...

[lexebidar]

I can share some nvidia freesync experience with new Acer XV272U. Unfortunately I had to return the monitor because 1440p is a bit too demanding for me and does not look that much better than 1080p... And the there was build quality issue(frame gap) The pixel responsiveness on this unit was bad too. A lot of ghosting.
About it's freesync implementation. It worked very well with 140fps rtss lock. No noticeable issues whatsoever. Frametimes seemed consistent and LFC indeed worked but the range was WRONG. The vsync behaviour was not the same as with gsync indeed but I could not put my finger on it. It behaved different on ag251fg when I had it

The frame doubling started below 60fps and doubling stayed enabled until 70fps at least. I felt that the monitor stayed in LFC doubling mode for far too much than it was necessary.
Other than that, the experience was fine.

Anyone tried these monitors:
LG 34UC79 (esoecially how it compares to gsync version 34UC89)
Samsungs cfg73, chg73 (seens some reports of bad range and/or flickering at certain fps)
AOC C27G1
Asus VG279Q (very interesting to see this one. 1080p 144hz IPS!)