GPU scaling and input lag?

[calcifer]

First, sorry for my really poor english, i'm french

I have read some things about scaling and input lag here: http://www.overclock.net/t/1388836/inpu ... ia-drivers
Since my windows 8 (8.1) update, i can't use the "display scaling" option in my nvidia driver (331.82) with my BenQ xl2420t (still available with windows 7 and the same driver) and i'm forced to use "GPU scaling".
So i really want to know if GPU scaling could add some input lag (not logical at all for me) with other than native resolution (and with native resolution ?) or if this is totally bullshit and GPU scaling is fine or better (input lag meaning) than my xl2420t scaling

Thank you!

[SnowHydra]

I have also read that GPU scaling adds input lag and that display scaling is superior, but some displays just do not have the option. It is true that lots of people have been telling nvidia about this (and to ask them to add back in NO SCALING) but nvidia has never responded or done anything.

I do believe that perhaps eventually we will be able to get them to do it if we are loud enough, but I regrettably doubt it. It's disappointing.

I can't say anything about GPU scaling ACTUALLY adding input lag, as I haven't tested it, but it must add some amount; it's an extra process being done, whether it be 1ms or .00001ms. It has been shown that some displays without a scaler have lower input lag than displays with them, so we can say for sure that scaling does add latency. How much, I do not know, if it even matters.

But we're all looking for that holy grail for no latency, right? I know how you feel. Sadly, we cannot do much other than get nvidia to reintroduce "no scaling" option in their drivers... which they probably don't even care about.

[Samhain]

Very interesting! Would love to see someone (with more knowledge than me) test this theory!

I too miss the "no scaling" days as I prefer a 1:1 pixel mapping. For instance, some games have loading screens that are lower resolution than gameplay screens. The loading screen is scaled to fit and ends up looking like arse! I would much rather have the native resolution picture surrounded by vertical and horizontal black bars (like a picture frame).

If "no scaling" helps with low latency, then we really need to push nVIDIA to put this setting back!

[RealNC]

See also: https://forums.geforce.com/default/topi ... e-scaling/

A year ago I've sent NVidia an email about this, but I suspect this kind of email ends up in /dev/null in their offices.

[calcifer]

I don't understand how GPU scaling could add some input lag. I imagine that is some post process that only affects framerate. If not post process AA like FXAA would add some input lag too, no?
No one knows more about this?

[Chief Blur Buster]

I don't understand how GPU scaling could add some input lag.

GPU scaling can essentially copy from one (lower-resolution) frame buffer to another (higher-resolution) frame buffer, while scaling the image -- this would technically add latency.

There are ways to do real-time, zero-buffered scaling by outputting it, line-at-a-time, but that does not appear to be what GPU scaling does.

[calcifer]

I don't understand how GPU scaling could add some input lag.

GPU scaling can essentially copy from one (lower-resolution) frame buffer to another (higher-resolution) frame buffer, while scaling the image -- this would technically add latency.

There are ways to do real-time, zero-buffered scaling by outputting it, line-at-a-time, but that does not appear to be what GPU scaling does.

Tank you for your response! I still have a lot of questions:

This would add "one frame" latency, right? So if i have 300-400 fps (my fps with CS:GO) this would add only ~3ms?
Display scaling use buffering too, i think, so why would it be better than GPU scaling? I have read on some review that the scaling on the BenQ xl2420T adds no latency. How it is possible?
FXAA is post process so could it add some latency too?

[Chief Blur Buster]

This would add "one frame" latency, right? So if i have 300-400 fps (my fps with CS:GO) this would add only ~3ms?

To the best of my knowledge -- one refresh latency, not one frame latency.

Display scaling use buffering too, i think, so why would it be better than GPU scaling?

Not necessarily. You can do scaling in realtime, as you output lines. You would only need about one scanline lookforward (e.g. 1080p 120Hz is 135KHz horizontal scan rate -- so one row of pixels is 1/135,000th of a second -- less than 10 microseconds). This is virtually lagless. However, GPU scaling does not do this.

I have read on some review that the scaling on the BenQ xl2420T adds no latency. How it is possible?

You scale on the fly as you scan out, with probably a one-scanline lookahead buffer.

FXAA is post process so could it add some latency too?

Not sure of the specifics of the FXAA process, so can't answer this question. However, someone else reading this surely knows?

[RealNC]

FXAA is post process so could it add some latency too?

Any computational task adds latency. It doesn't matter if it's post-process or not. If it increases rendering time, it adds latency.

MSAA adds more latency than FXAA, even though FXAA is post-process and MSAA is not.

[srsbsns]

From personal experience running 2x 7970ghz editions I found using GPU scaling reduced input lag on the Asus VG248QE. The setting had no noticible effect when used on a BenQ XL2420TE though.

Side note though the Asus has noticeable input lag where the BenQ did not.