NVIDIA Low Latency Mode question

[speancer]

Does enabling Low Latency Mode in NVIDIA Control Panel (either "on" or "ultra") do anything at all in terms of input lag reduction if you play a game at very high frame rate (200+ at all times) that is not GPU-bound at all (e.g. CS:GO) ? Is there any negative performance impact coming from enabling this option in high frame rate non GPU-bound scenario?

[jorimt]

If you're system isn't GPU-bound, the only thing LLM might theoretically do is affect frame pacing, depending on how the given engine handles the the pre-rendered frames queue, and if it allows LLM override to take effect (a LOT of games don't, and DX12/Vulkan games don't at all).

LLM "On" is typically safe to leave enabled on a gaming-capable system, GPU-bound or no.

Battle(non)sense found LLM "Ultra" slightly increased input lag in non-GPU-bound scenarios. I haven't tested that myself, so I can only guess why he found that to be the case, said guess being it caused repeat frames here and there due to its "just-in-time" delivery component acting up when the system wasn't GPU-bound, but that's up in the air.

Anyway, as I've said elsewhere, LLM is a highly overvalued setting. Even at its best, it doesn't do much to reduce input lag; typically around 1 to a little over 1 frame reduction in GPU-bound scenarios.

[MatrixQW]

If you're system isn't GPU-bound, the only thing LLM might theoretically do is affect frame pacing

What does/do you mean by frame pacing?
Detrimental or beneficial?

[jorimt]

If you're system isn't GPU-bound, the only thing LLM might theoretically do is affect frame pacing

What does/do you mean by frame pacing?
Detrimental or beneficial?

In non-GPU-bound scenarios LLM "On" potentially neutral/beneficial, LLM "Ultra" potentially neutral/detrimental.

[disq]

btw and sorry to (kinda) hijack the thread

when a game reaches 95%+ of GPU usage for a couple of frames, like this:



would this be considered "GPU bound"? and so the need to use LLM?

[jorimt]

would this be considered "GPU bound"? and so the need to use LLM?

Unless you've closely monitored each game, determined your minimum achievable framerate, and set the FPS limiter to that, the likelihood is, in many games, unless said games are heavily CPU-dependent (e.g. CS:GO), your system is going to drop in and out of GPU-bound scenarios at any given time depending on the demands of the given scene.

This is why I typically recommend keeping LLM "On" (with or without G-SYNC), so IF your system becomes GPU-bound at any point, and IF the game in question supports LLM override, you'll potentially keep the pre-rendered frames queue lower.

I personally just keep LLM globally set to "On," since at worst, it does nothing, and at best, you could reduce input lag by about 1 frame in GPU-bound scenarios.

[disq]

Forgot to mention that i already capped my frames, and i'm only talking about one game, which i locked at 120fps (with GSYNC) but as you can see above, it still goes to 95%+ GPU Usage in some frames

My confusion comes from Battle(non)sense video (https://youtu.be/Gub1bI12ODY?t=277) where in the smooth case he keeps LLM off with a cap

I guess that in my case 120fps is still too much and i need to lower it more to avoid those tiny spikes to 95% or above in few frames?!

[jorimt]

I guess that in my case 120fps is still too much and i need to lower it more to avoid those tiny spikes to 95% or above in few frames?!

< 95% can still be in the "safe" range, but yes, if you want to ensure your system is never GPU-bound in that game, you'll need to lower your cap to more closely match your 1% lows, for instance.

[disq]

that's what i was thinking, thanks for your input @jorimt

and sorry @speancer for the slightly thread hijack

[jorimt]

that's what i was thinking, thanks for your input @jorimt

and sorry @speancer for the slightly thread hijack

No worries @disq, and it was still within subject