Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Talk about NVIDIA G-SYNC, a variable refresh rate (VRR) technology. G-SYNC eliminates stutters, tearing, and reduces input lag. List of G-SYNC Monitors.
User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by jorimt » 11 Jan 2020, 14:50

Vleeswolf wrote:
11 Jan 2020, 11:49
Battlenonsense finds the new NVIDIA FRL perform as well as RTSS in BFV and Fortnite: https://youtu.be/W66pTe8YM2s
In the games he tested, that indeed seems to be the case. I've recently updated my Optimal G-SYNC Settings to reflect this info. That said, there may still be some kinks for Nvidia to work out depending on the game engine the new option is being used with:
https://www.blurbusters.com/gsync/gsync ... mment-9288

So, RTSS still might be the safer, more reliable bet for external limiting at this time.
poppe wrote:
11 Jan 2020, 12:00
Still unsure what to set Low Latency mode to...
Guess I'll just set it to Ultra globally with the NVCP FPS cap.
And in games I'm more serious about like CS:GO I'll just use LLM "On" and use the ingame FPS cap.

We still don't know if Nvidia are improving the Ultra mode unless someone decides to test it every driver release..
You didn't say whether you had a VRR setup, but with G-SYNC, "Ultra" is basically an automatic FPS limiter + MPRF "1."

So, say you limit your FPS to 141 @144Hz with Nvidia's new Max Frame Rate option, and then enabled "Ultra" LLM as well; the ~138 auto limit will take effect instead of your set 141 FPS limit, which is why I recommend setting LLM to "On" when using it in combo with G-SYNC and an in-game or external FPS limiter.

For non-G-SYNC though, an external or in-game limiter + "Ultra" should be fine, as it doesn't have the auto-capping behavior when used with fixed refresh rates.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

poppe
Posts: 7
Joined: 08 Feb 2019, 06:51

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by poppe » 11 Jan 2020, 15:39

jorimt wrote:
11 Jan 2020, 14:50
You didn't say whether you had a VRR setup, but with G-SYNC, "Ultra" is basically an automatic FPS limiter + MPRF "1."

So, say you limit your FPS to 141 @144Hz with Nvidia's new Max Frame Rate option, and then enabled "Ultra" LLM as well; the ~138 auto limit will take effect instead of your set 141 FPS limit, which is why I recommend setting LLM to "On" when using it in combo with G-SYNC and an in-game or external FPS limiter.

For non-G-SYNC though, an external or in-game limiter + "Ultra" should be fine, as it doesn't have the auto-capping behavior when used with fixed refresh rates.
I do use G-SYNC @ 240 Hz

Reason I'm so confused about latency mode is because of Battlenonsense testing:

https://i.imgur.com/u3ysznI.png
https://i.imgur.com/5HSeMvs.png

Ultra latency mode gives higher input lag even though it "autocaps" FPS. But when GPU load is high then Ultra is better..
So I thought setting Ultra globally for casual games with an FPS cap for DX12 games (since the autocap don't work there) and then for serious games like CS:GO I'll just use LLM "On" with ingame FPS limiter.

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by jorimt » 11 Jan 2020, 16:38

poppe wrote:
11 Jan 2020, 15:39
Reason I'm so confused about latency mode is because of Battlenonsense testing:

[...]

Ultra latency mode gives higher input lag even though it "autocaps" FPS. But when GPU load is high then Ultra is better..
So I thought setting Ultra globally for casual games with an FPS cap for DX12 games (since the autocap don't work there) and then for serious games like CS:GO I'll just use LLM "On" with ingame FPS limiter.
If we're talking only about G-SYNC, again, LLM "Ultra" is MPRF "1" + auto FPS limiting, whereas LLM "On" is only MPRF "1." For non-G-SYNC scenarios, Nvidia claims that LLM "Ultra" (on top of using MPRF "1") has a "just-in-time" frame delivery function. This isn't needed (or used with G-SYNC) as it already does this.

So, for G-SYNC, if you're already using an FPS limiter and your framerate is sustained above it, LLM "On" is doing nothing until your FPS drops below your limit, at which point it can reduce input lag by up to 1 frame. With "Ultra," it does the same thing as "On" but also sets an auto FPS cap in supported games, which isn't needed if you're already using an in-game or RTSS/Nvidia Max Frame Rate limit, thus the recommendation to use LLM "On" in these scenarios instead, so the "Ultra" auto cap doesn't interfere with your manually set FPS limit.

tldr; VRR LLM "Ultra" functions differently than fixed refresh rate LLM "Ultra."
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

Zaxeq
Posts: 4
Joined: 27 Nov 2019, 17:09

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by Zaxeq » 11 Jan 2020, 17:41

I do use G-SYNC @ 240 Hz

Reason I'm so confused about latency mode is because of Battlenonsense testing:

https://i.imgur.com/u3ysznI.png
https://i.imgur.com/5HSeMvs.png

Ultra latency mode gives higher input lag even though it "autocaps" FPS. But when GPU load is high then Ultra is better..
So I thought setting Ultra globally for casual games with an FPS cap for DX12 games (since the autocap don't work there) and then for serious games like CS:GO I'll just use LLM "On" with ingame FPS limiter.
Okay so let's do a little math real quick.. In the first picture you posted the averages of both were the following: LLM Ultra, Vsync+Gsync is 44ms avg, with Vsync+Gsync NO NULL was 32ms avg.

If you take 1000/139 you get 7.19ms delay from the FPS, now with LLM Ultra, Vsync and Gsync all enabled it does the hidden nvidia fps limiter which is 2 frames of delay, so 7.19ms*2= 14.4ms delay, now look back at the previous values, 44ms(NULL)-32ms(No NULL). and you get 12ms which is pretty close to 14ms which makes sense because it has around a 2 frame delay, hope that helps.

Now why maybe NULL ultra+auto fps cap is 12ms instead of 14.4 could either be margin of error or possibly what NULL says it does on ultra is it delivers the frame to the gpu just in time instead of having up to a frame delay with it off, but I haven't seen someone test with a high speed camera NULL ultra without gsync and vsync auto capper to see the difference with it on Ultra, On, or off.

poppe
Posts: 7
Joined: 08 Feb 2019, 06:51

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by poppe » 11 Jan 2020, 18:32

jorimt wrote:
11 Jan 2020, 16:38
poppe wrote:
11 Jan 2020, 15:39
Reason I'm so confused about latency mode is because of Battlenonsense testing:

[...]

Ultra latency mode gives higher input lag even though it "autocaps" FPS. But when GPU load is high then Ultra is better..
So I thought setting Ultra globally for casual games with an FPS cap for DX12 games (since the autocap don't work there) and then for serious games like CS:GO I'll just use LLM "On" with ingame FPS limiter.
If we're talking only about G-SYNC, again, LLM "Ultra" is MPRF "1" + auto FPS limiting, whereas LLM "On" is only MPRF "1." For non-G-SYNC scenarios, Nvidia claims that LLM "Ultra" (on top of using MPRF "1") has a "just-in-time" frame delivery function. This isn't needed (or used with G-SYNC) as it already does this.

So, for G-SYNC, if you're already using an FPS limiter and your framerate is sustained above it, LLM "On" is doing nothing until your FPS drops below your limit, at which point it can reduce input lag by up to 1 frame. With "Ultra," it does the same thing as "On" but also sets an auto FPS cap in supported games, which isn't needed if you're already using an in-game or RTSS/Nvidia Max Frame Rate limit, thus the recommendation to use LLM "On" in these scenarios instead, so the "Ultra" auto cap doesn't interfere with your manually set FPS limit.

tldr; VRR LLM "Ultra" functions differently than fixed refresh rate LLM "Ultra."
Alright, thought the "Ultra" setting did more than that in regards to system latency.
It's basically just a "lazy mode" from Nvidia which is why they recommend it instead of trying to explain to people why they should cap their framerate.

Or for those rare games that you can't cap the framerate in, haven't stumbled upon one yet though.

poppe
Posts: 7
Joined: 08 Feb 2019, 06:51

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by poppe » 11 Jan 2020, 18:37

Zaxeq wrote:
11 Jan 2020, 17:41
I do use G-SYNC @ 240 Hz

Reason I'm so confused about latency mode is because of Battlenonsense testing:

https://i.imgur.com/u3ysznI.png
https://i.imgur.com/5HSeMvs.png

Ultra latency mode gives higher input lag even though it "autocaps" FPS. But when GPU load is high then Ultra is better..
So I thought setting Ultra globally for casual games with an FPS cap for DX12 games (since the autocap don't work there) and then for serious games like CS:GO I'll just use LLM "On" with ingame FPS limiter.
Okay so let's do a little math real quick.. In the first picture you posted the averages of both were the following: LLM Ultra, Vsync+Gsync is 44ms avg, with Vsync+Gsync NO NULL was 32ms avg.

If you take 1000/139 you get 7.19ms delay from the FPS, now with LLM Ultra, Vsync and Gsync all enabled it does the hidden nvidia fps limiter which is 2 frames of delay, so 7.19ms*2= 14.4ms delay, now look back at the previous values, 44ms(NULL)-32ms(No NULL). and you get 12ms which is pretty close to 14ms which makes sense because it has around a 2 frame delay, hope that helps.

Now why maybe NULL ultra+auto fps cap is 12ms instead of 14.4 could either be margin of error or possibly what NULL says it does on ultra is it delivers the frame to the gpu just in time instead of having up to a frame delay with it off, but I haven't seen someone test with a high speed camera NULL ultra without gsync and vsync auto capper to see the difference with it on Ultra, On, or off.
Makes sense.

The max GPU load latency still makes me question how this "Ultra" mode really works though.

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by jorimt » 11 Jan 2020, 21:39

poppe wrote:
11 Jan 2020, 18:32
It's basically just a "lazy mode" from Nvidia which is why they recommend it instead of trying to explain to people why they should cap their framerate.

Or for those rare games that you can't cap the framerate in, haven't stumbled upon one yet though.
Pretty much.

So long as you can sustain the framerate at your set FPS limit, and said FPS limit is also preventing your GPU from maxing out, it's effectively already the same thing as MPRF "0," G-SYNC or no G-SYNC.

Typically, pre-rendered frames settings only kick in (at least where input lag is concerned) when your framerate isn't limited by an FPS cap.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

janos666
Posts: 16
Joined: 15 Dec 2017, 18:50

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by janos666 » 11 Jan 2020, 22:32

So Ultra doesn't (try to) reduce the lag with ~100% GPU utilization (and the fps stuck below the limit) in case of G-Sync+VSync=On?
I didn't do any scientific testing on this but Ultra seemed to cause some stutter in The Witcher 3 (I just loaded the same save file and ran a few rounds in a town after changing some settings) compared to LLM On (with game settings which provide ~100 fps for 1440p120). Could be coincidence/placebo though but I thought Ultra does the same thing it supposedly does with G-Sync=Off in GPU-limited scenarios.

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by jorimt » 11 Jan 2020, 23:16

janos666 wrote:
11 Jan 2020, 22:32
So Ultra doesn't (try to) reduce the lag with ~100% GPU utilization (and the fps stuck below the limit) in case of G-Sync+VSync=On?
[...] Could be coincidence/placebo though but I thought Ultra does the same thing it supposedly does with G-Sync=Off in GPU-limited scenarios.
As far as I'm aware, in games that support external manipulation of the pre-rendered frames queue (The Witcher 3 seeming to be one; not all do), LLM "Ultra" with G-SYNC doesn't do anything more (or different) than LLM "On" to reduce the pre-rendered frames queue in non-FPS-limited/GPU-bound scenarios. E.g. say when you're framerate is hovering in the 100's at 144Hz with 99% GPU usage.

In non-G-SYNC scenarios, according to Nvidia, it does, by delivering frames "just-in-time," which could potential cause additional stutter with fixed refresh rates (due to the occasional miscalculation in frame delivery, causing a repeated/skipped frame; something Battle(non)sense seemed to be observing in his original tests of the mode), where (beyond any additional stutter that LLM "On" might cause due to possible starvation of frames on a system with a lopsided CPU/GPU setup) it wouldn't with G-SYNC, as, again, G-SYNC is already delivering frames "just-in-time" by syncing the display to the GPU.

If "Ultra" is doing anything else beyond setting MPRF to "1" internally, and auto-capping the FPS with G-SYNC, I'm not aware of it, and Nvidia probably isn't going to volunteer that info, so it would simply take more objective testing (of many, many games), of which I don't have the current tools or time to accomplish in an efficient or effective manner.

If at any point I had either (ideally both), I would consider it though.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

janos666
Posts: 16
Joined: 15 Dec 2017, 18:50

Re: Driver 441.08: Ultra-Low Latency Now with G-SYNC Support

Post by janos666 » 12 Jan 2020, 00:02

jorimt wrote:
11 Jan 2020, 23:16
As far as I'm aware, in games that support external manipulation of the pre-rendered frames queue (The Witcher 3 seeming to be one; not all do)
I think Frostbite3 games (like DAI) override this. I see a clear difference when I set the renderaheadlimit from the dev.console from default -1 (assumed to result in 2) to 1 after launching the game with On or Ultra (which supposedly sets it to 1 or creates a scenario where it's irrelevant and effectively 0). I don't understand why nVidia didn't try to force this new setting over the game's own numbers, at least with Ultra and popular game engines.
jorimt wrote:
11 Jan 2020, 23:16
In non-G-SYNC scenarios, according to Nvidia, it does, by delivering frames "just-in-time," which could potential cause additional stutter with fixed refresh rates (due to the occasional miscalculation in frame delivery, causing a repeated/skipped frame; something Battle(non)sense seemed to be observing in his original tests of the mode), where (beyond any additional stutter that LLM "On" might cause due to possible starvation of frames on a system with a lopsided CPU/GPU setup) it wouldn't with G-SYNC, as, again, G-SYNC is already delivering frames "just-in-time" by syncing the display to the GPU.
That makes sense. There is always a known 'time' in case of fixed refresh with V-Sync (the time of the next planned refresh in the fixed period cycle) but "just-in-time" is hard to apply for G-Sync where the next refresh is planned to happen whenever the next frame becomes available.

And I guess the 'V-Sync lag' (in case of uncapped fps with G-Sync+V-Sync) would still be higher with Ultra (if that could even work since it's not regular V-Sync) than the lag of the fps limiter, so that (G-Sync+V-Sync with no fps cap + Ultra for lower lag when fps=maxrefresh) wouldn't be a recommendable option anyways (and that's probably why nVidia added an auto fps cap instead of making this kind of Ultra mode working for G-Sync + V-Sync).
jorimt wrote:
11 Jan 2020, 23:16
If "Ultra" is doing anything else beyond setting MPRF to "1" internally, and auto-capping the FPS with G-SYNC, I'm not aware of it, and Nvidia probably isn't going to volunteer that info, so it would simply take more objective testing (of many, many games), of which I don't have the current tools or time to accomplish in an efficient or effective manner.
I vaguely theorized something like Ultra might tries to "starve" the GPU a little bit (sacrificing some fps) to avoid the regular ~100% utilization scenario. But that doesn't make much sense (if it limited the utilization on the driver side to, say, 95% then 95% would simply become the new 100%, achieving nothing but slightly lower fps, hence slightly higher lag...).

Post Reply