Avean wrote: ↑14 May 2020, 15:17
BattlenonSense
https://www.youtube.com/watch?v=7CKnJ5ujL_Q
If youre enabling it when youre not gpu bound you get the opposite effect. You get worse input lag by having it on as shown in his tests here. Ive also read some comments about it from the staff here i think.
All of his tests were done with low latency "Ultra" which we already know adds lag, and also caps fps on a driver level (slower but still faster than uncapped when paired with gsync) which I even already covered that in this very same topic;
Dmoney405 wrote:
Use low latency mode "On" in nvidia control panel. Do not use "Ultra" unless you are GPU bound BELOW your 141 cap. (low latency mode only works for DX11 games...no vulkan or DX12)
Then I further explained it in in my next post on here;
Dmoney405 wrote:
Turning it to 'ultra' is okay on grandmas computer because it take the thought process out of it. Just set it and go. The issue with this is that in games where you can easily cap your refresh rate will be capped at 141hz instead of 144 by nvidia software instead of an ingame limiter, potentially giving up one to two frames of lag resulting in 6-12ms of added input lag at 144hz and like 8-16ms at 120hz. Leaving it set to 'On" instead of "Ultra" allows you to still choose how you want to cap the framerate for more flexibility and potentially lower lag. It is called low latency mode 'ultra' because it is still a net reduction in lag over the standard of not capping fps and allowing vsync to take over gsync at the monitors selected refresh rate, which gives 3 frames of input lag. Since 6-12ms < 18-21ms of input lag they are still correct by labeling it a low latency solution. But power users like most people here can squeeze more performance out of "on" and a manual -3fps cap ingame or with RTSS, if needed.
I will now take an important part of out of the above quote you need to read again;
Dmoney405 wrote:
It is called low latency mode 'ultra' because it is still a net reduction in lag over the standard of not capping fps and allowing vsync to take over gsync at the monitors selected refresh rate, which gives 3 frames of input lag. Since 6-12ms < 18-21ms of input lag they are still correct by labeling it a low latency solution.
Since this is a discussion about about input lag and gsync I was offering ultra mode as a lazy setting that technically gives you lower latency but isn't the fastest option. The lag you get from that mode is solely from the driver level frame cap and has nothing to do with lowering the queue size. Even with the added delay this is still lower input lag than using gsync at capped >>refresh rate<< *not fps* because gsync turns off and vsync takes over giving you the wonderful 3 giant frames of lag we all know and hate from it.
The queue in question is referencing how many frames the cpu can render before giving it to the gpu. It has no bearing on techniques the gpu uses to display a picture. When the gpu finishes a frame and asks the CPU for the next one the CPU will send the one at the end of the queue. With a frame buffer of 3 that frame is 2-3 frames behind the one the cpu is drawing currently. Allowing a large buffer with a cpu bottleneck or even a moderately stressed cpu allows it to pool these so it is more likely to have one ready exactly when the GPU calls for it, reducing stutters but adding lag. The stutter is the gpu sitting there with nothing to do waiting for the cpu to send the next job. If the gpu is at a high utilization and is severely bottlenecked you can use
>Low latency: Ultra< because the gpu at this point has the highest delay in the system and the cpu can easily render one frame at a time for it in real time when called; or if the cpu and GPU are both equally stressed (or not stressed) you can use
>Low Latency: On< to reduce this queue size since the cpu can easily keep up with a single frame buffer (not to be confused with zero buffers from low latency ultra) allowing a lower input lag. If at any point you start to stutter, that is a good sign that your CPU is having issues keeping up and you should increase the buffer by 1 until it smooths out. To think having a smaller buffer would INCREASE input lag is ridiculous. It is designed to do the exact opposite.
The best part is this works at all framerate but you get the highest benefit from lower framerates because......SCALING! Yes just like how going from 60hz to 120hz feels like a MASSIVE improvement but going from 120hz to 180hz feels better but definitely not on the same level. These frame buffer settings do not require some arbitrary number of frames to work. 1 frame at 60 fps saved is 16.67ms 1 frame at 144hz is 6-7ms. So while you do get more for your money at lower frames it is still 100% something you should be doing at all times, if you can manage this without the dreaded cpu bottleneck stutter. (cpu bottleneck is read as a PER USED CORE and not a TOTAL CPU usage, if the game uses a single core in your 10 core processor then you are considered bottlenecked when that single core reaches 100% usage and must increase frame buffers to avoid stutter)
So to recap; your youtube friend used "ultra", which adds some driver level frame cap hook that still seems to be in effect even if you use an ingame fps cap for whatever reason and got misleading results. I am quite certain that if he does the same tests with 'On' instead of 'Ultra' in a dX11 game he would see an inprovement.
..............can we please stop treating youtube as gospel and instead try to dissect their videos to validate their information before spreading it? All of this information is already on this website, you wouldn't even have to search very hard.
https://blurbusters.com/gsync/gsync101- ... ttings/14/ Hit ctrl + F and type in low latency. Read the article. Realize that the people who are in charge of this site do a crap load of research before posting a potentially bogus guide and probably know better than youtuber #232,389,178,194; especially when he admits he knows little about the subject. He didn't even mention the auto fps cap with Ultra a single time.