Testing Input Latency

General Discussion
1

I was interested in seeing what settings actually influenced input latency on my current machine, it was always a problem back in the laptop days but I’ve never really looked into it extensively since.

Current PC;
CPU - 5900x
GPU - 3090
RAM - 2x 8GB 3600MHz CL16
SSD - Samsung 970 Evo Pro
Display - 1080p 1ms @ 240Hz

Average input latency - 2.5ms
Lowest obtained input latency - 2.5ms
Highest obtained input latency - 16.7ms

Now I’ll put in the settings and what the outcomes were for me, this was all observed through the in game netgraph / CTRL + Shift + N. Assume everything was tested with optimal settings and only it being the factor of influence at 1080p with a consistent 400 FPS in a simulated Quickplay (against bots).

In game settings
Fullscreen - 2.5ms
Windowed - 2.5ms
Borderless Windowed - 2.5ms
FoV 55-103 - 2.5ms / no difference.
Vsync 240 - 4.5ms
Triple Buffering On / Off - 2.7ms / 2.5ms
Reduce Buffering On / Off - 2.5ms / 2.5ms
Render Scale 100% / 200% - 2.5ms / 2.5ms
Texture Filter Quality - 2.5ms all settings.
Dynamic Reflections - 2.5ms all settings.
Texture Quality - 2.5ms all settings.
Local Fog Detail - 2.5ms all settings.
Shadow Detail - 2.5ms all settings.
Model Detail - 2.5ms all settings.
Effects Detail - 2.5ms all settings.
Lighting Quality - 2.5ms all settings.
Antialiasing - 2.5ms all settings.
Refraction Quality - 2.5 ms all settings.
Local Reflections On / Off - 2.5ms / 2.5ms
Ambient Occlusion On / Off - 3.1ms / 2.5ms

What I found to be the biggest contributor, which is an axiom anyways, is the framerate -
FPS @ 60 - 16.7ms
FPS @ 120 - 8.3ms
FPS @ 144 - 6.9ms
FPS @ 240 - 4.5ms
FPS @ 300 - 3.3ms
FPS @ 400 - 2.5ms

Nvidia Settings
Vsync 240 - 4.5ms
Threaded Optimization On / Off - 2.5ms / 2.5ms
Trilinear Optimization - 2.5ms.
Texture Filtering (all settings ) - 2.5ms / no deviation.
Shader Cache On / Off - 2.5ms / 2.5ms.
MFAA On / Off - 2.5ms / 2.5ms
Low Latency Mode Off / Ultra - 2.5ms / 2.5ms.
Antialiasing settings - 2.5ms / no deviation.
Anisotropic Filtering - 2.5ms / no deviation.
Image Sharpening - 2.5ms / no deviation.

Some windows settings
Disable full screen optimizations On / Off - 2.5ms / 2.5ms
Power plan balanced - 2.5ms
Power plan performance - 2.5ms

TLDR
On a high end system input latency seems to have little relevant settings outside of strictly FPS. The lowest obtainable input latency at a consistent 400 FPS is 2.5ms, which also seems to remain the stable average when testing singular settings outside of ambient occlusion, which has a very nominal <1ms impact.

Deduction? Best way to remove input latency is upgrade your computer, outside of that it seems other settings will have different impacts on different systems and frame rates could be a corroborative factor.

5 Likes
2

This shows how much overwatch is optimized, in a lot of games your graphic settings have a big impact on your latency, IIRC someone did almost the same testing in Fortnite and his results showed that your input latency can increase by about 10ms just from the graphic settings alone, and to see that overwatch have minimal effect on your input delay is great, as it means that if a player can afford having his game look nicer without losing fps (as it is capped), it wouldn’t hurt his in-game performance.

Great testing BTW.

1 Like
3

Correct me if I am wrong, but I have also heard of client-server frame-discrepancy or discrep. Where extra filtering/buffering/quality slightly dissociates what you see with what server sees. And this can be a static/permanent offset. E.g. where you see their head is something is like 30 frames behind where the server believes it is, not accounting for lag and other effects.

If that isn’t corrected for (client or server side), you’re getting an additional ‘effective’ input lag in the 2-3 ms ballpark. I’m not sure how you would track/measure this.

1 Like
4

No idea, though it could certainly be a factor. There’s yet to be a game that seems to have perfected it’s server / client sided shambles.

5

Well this engine certainly feels optimized. Maybe it’s no longer state-of-the-art, but it handles things reasonably well.

Where I expect a performance breakdown is in their migration to Google servers. IIRC they used to be with team amazon(ttv), but have been going full team google/yt. I don’t know if I trust the google servers to be as sharp. They seem tuned for throughput customers, and not esports-tier latency.