Blur Busters Forums

Hey Everyone,

I am currently looking to optimize the input latency in Gears 5 with gsync, and eliminate shot errors in-game. For the purpose of context in terms of nominal settings, I've capped the framerate at 237 fps via the game config file and currently have vsync enabled through the NVCP only. After experimenting with the graphics options in game, and using the in-game benchmark, I was able to minimize the GPU Bound percentage to as low as 90%. Multiplayer is of course a much less demanding experience for the game.

I've been using the "Reduced Buffering" option in-game and have the low latency mode disabled since the game does a decent a job at minimizing the frame queue. This combo has optimized the benchmark at a total frames rendered amount of 11,382 at an average of 189fps. However, whenever I do this, I'll end up missing a lot of shots that should've hit other players in multiplayer, and experience a slight stutter here and there. Even worse, the game will feel like I'm moving my mouse through sand at random intervals. I can usually fix this by deleting the temporary shader cache files and restarting the game. Unfortunately, the missed shot error still occurs, despite a point blank shotgun to the other player's face. After much frustration, I currently tested both NULL and the reduced buffering option together while also capping the frame rate to 237 fps. This provided a slightly lower amount of frames rendered for a total of ~11,200 at an average fps of ~185. I've tested both "Fullscreen Mode" and "Fullscreen and Windowed Mode" with this setting NULL and RB combo, and found that the "Fullscreen Mode" had the best effect in terms of hitting shots. With NULL enabled, I found that the stutters were almost non-existent and the game felt incredibly responsive, and I was able to hit more shots. However I cannot confirm this feeling since I currently do not have any method of measuring input latency.

In conclusion, my main questions are:

Would running both reduced buffering and NULL create errors in terms of managing frames queued?
The game does not run at an exclusive fullscreen mode since alt-tabbing is not necessary for when I want to click on something on my secondary monitor. Should I keep the "Fullscreen and Windowed Mode" on?
Are NULL and Reduced Buffering the same operation in terms of minimizing the frames queued?
Is there anybody here that feels like they have optimized the input latency in Gears 5 specifically?

joselaba97 wrote: ↑

11 Nov 2020, 19:03

Would running both reduced buffering and NULL create errors in terms of managing frames queued?

Gears 5 runs in DX12, DX12 and Vulkan don't support Nvidia Low Latency Mode settings.

jorimt wrote: ↑

11 Nov 2020, 21:33

joselaba97 wrote: ↑

11 Nov 2020, 19:03

Would running both reduced buffering and NULL create errors in terms of managing frames queued?

Gears 5 runs in DX12, DX12 and Vulkan don't support Nvidia Low Latency Mode settings.

Interesting, didn't know that about Dx12 or Vulkan. Therefore, reduced buffering, LLM:OFF, and the FPS cap would be the most optimal settings in this game. Would you suggest lowering the fps cap even lower? Say like 232 like Battle Nonsense said in one of his most recent videos? Could it be that the game is occasionally reaching 240fps and enabling full-on vsync?

joselaba97 wrote: ↑

12 Nov 2020, 16:35

Would you suggest lowering the fps cap even lower? Say like 232 like Battle Nonsense said in one of his most recent videos? Could it be that the game is occasionally reaching 240fps and enabling full-on vsync?

As you noted, if you're using G-SYNC, you must keep the framerate within the refresh rate, or it will revert to V-SYNC or V-SYNC behavior (depending if you have the V-SYNC option on or off).

Regarding setting an even lower limit than the recommended minimum -3 FPS, you certainly can. If you limit the FPS to a level where your GPU is never maxed, for instance, you'll reduce input lag even more than you would if you used an LLM setting, as the pre-rendered queue will effectively remain empty instead of just reduced.