Is there any point to enabling "Ultra Low Latency" mode in conjunction with V-Sync and an FPS cap?

In BattleNonSense's tests in Overwatch, he found that while using the in-game FPS limiter enabling ULLM mode externally actually increased input delay very very slightly (it was about the same though -- almost identical really seems to me).

However, he did NOT test ULLM mode in conjunction with RTSS's FPS limit which is what I'd expect many will be using in conjunction with V-Sync for "Low Lag" V-Sync such as is described in BlurBuster's guide (where you use V-Sync then cap the framerate to reduce input delay) given not all games have an internal limiter and some that do are still worse than RTSS's limiter with respect to frametimes while not always being better with respect to input lag.

What I want to know is, if I use RTSS to cap the framerate in conjunction with V-Sync will enabling ULLM mode even do anything/reduce input lag at all? Interestingly when I used the in-game reduce buffering setting in conjunction with V-Sync and Overwatch's in-game FPS cap (at 30 fps +half-refresh v-sync to exaggerate the effect as much as possible) it felt to me as though there was a bit less input delay than without the in-game reduce buffering setting ON. Once again though, BattleNonSense did not test this case as he only tested the in-game limiter with the external ULLM mode, not the internal reduce buffering one so all i've really got to go off of is my own "feeling" there.

I had read somewhere previously that limiting the framerate would mean that you were CPU bound rather than GPU bound all the time so frames wouldn't really be pre-rendered anyway so long as you were reaching your FPS limit at all times. That said, I'm not really sure how that works or if V-Sync ON + RTSS (or internal limiter) limit + ULLM has less input lag VS just using V-Sync ON + RTSS (or internal limiter) limit.

Has anyone ever done testing for this? Or, just happen to know how it works? I haven't really been able to find any concrete testing beyond BattleNonSense's which left out a lot of cases. Thanks for your time,

 

It has an effect whenever the FPS cap cannot be reached due to being GPU-bound. That's what this setting does to begin with, it reduces input lag when your GPU gets maxed out.

 

Yes it makes sense. ULL delays rendering the frame until the last possible moment, while having only a small impact on performance. Rendering the frame later helps reduce lag between your input and the result being displayed on your screen.

ULL will only help if the GPU is not taking the full frame time to render the frame, as if this is the case then the frame rendering cannot be delayed until later. So, it's unlikely to help noticably if you're GPU limited, based on this.

 

As far as I remember, LLM should be set to 'On' as opposed to 'Ultra Low Latency' when using G/F-sync+VSync+FRTC or just VSync+FRTC. What exactly the performance difference between these two settings is I do not know but the two setting options obviously dictate exactly how & when the function is implemented.

maybe @Astyanax could give some insight? He seems quite knowledgeable regarding these things, although his manner of presentation of such knowledge is frowned upon by some. He's a 'No Nonsense' kind of personality if you get my drift ?

 

As far as I recall it specifically suggest using Ultra when using G-sync in the tips at the bottom in the Nvidia Control Panel? Not on my PC to check right now though.

 

Yeah, I see that bro but it refers only to G/F-Sync + VSync. When Using G/F-Sync + VSync + FRTC it should be set to 'On' for best performance. I set my F-Sync monitor + NVCP according to BattleNonSense advice as G & F Sync may be different technologies, they ultimately perform their function in the same way making settings applicable across both platforms.

I may be wrong though as you pointed out which is why I called for some more brainpower/input on this conundrum

*** Edit addition ***

As far as I can tell, NVCP requires pre rendered frames to be set at '1' for LLM to function optimally within the driver. Setting LLM to 'On' locks pre rendered frames to '1' automatically ensuring optimal performance from the driver regarding LLM.

In specific cases/workloads a pre rendered frames setting of less than or greater than '1' 'May' be beneficial. In such cases 'ULLM' would be the corrrect setting as this does not lock the pre rendered frames to '1' but instead allows the pre rendered frames be set lower or higher whilst still applying the benefits of LLM + pre rendered frames '1' to its best ability.

For gaming purposes (GRD), this driver, as are all GRD's is tuned towards pre rendered frames '1' as a standard for some of the feature implementations.

 

I don't believe that's true. ULLM basically tries to make the rendering pipeline act as if it was close to a pre-rendered queue size of 0. Which is what a frame limiter will do as well, btw, unless the FPS cap target is not reached.

 

I think that in the absence of official word on this dilemma, we each need to find what's good for our own particular problem. I'm running f-Sync + VSync + LLM 'On' + FRTC (143fps) without any issues or discernible input lag. 1080ti running dual 32" 1440p F-Sync monitors via DP's. Driver is 460.79.

In light of no difinitive answer I think if nothing else, debating this question has been interesting

 

Thanks! That's really helpful to know -- you da man ReacNC