Determine how best to measure perceived input latency (character input to screen) #7727

@vadimkantorov commented on GitHub (Apr 27, 2020):

When I step on a very long line in vim and just press key "Right", WT glitches noticeably, sometimes even hanging for what looks like ~100ms. I don't know if it's cursor/character rendering that's slow or underlying controls being slow or sth else...

This may sometimes be the consequence of slow SSH / internet, but in this time it's not the case I think. wsltty renders cursor / characters a bit faster

@vadimkantorov commented on GitHub (Apr 27, 2020): When I step on a very long line in vim and just press key "Right", WT glitches noticeably, sometimes even hanging for what looks like ~100ms. I don't know if it's cursor/character rendering that's slow or underlying controls being slow or sth else... This may sometimes be the consequence of slow SSH / internet, but in this time it's not the case I think. wsltty renders cursor / characters a bit faster

claunia commented

@DHowett-MSFT commented on GitHub (Apr 27, 2020):

So, the software I usually use to measure things like this is called "typometer". It's an old piece of Java software that injects input and waits for the output to be rendered on the screen.

We've gotten a number of improvements in rendering speed since 0.11 landed, so I'm inclined to say that your issues are going to be largely addressed.

If you have VS 2019, I welcome you to try out a Release build from master 😄

@DHowett-MSFT commented on GitHub (Apr 27, 2020): So, the software I usually use to measure things like this is called "typometer". It's an old piece of Java software that injects input and waits for the output to be rendered on the screen. We've gotten a number of improvements in rendering speed since 0.11 landed, so I'm inclined to say that your issues are going to be largely addressed. If you have VS 2019, I welcome you to try out a Release build from `master` :smile:

claunia commented

@vadimkantorov commented on GitHub (Apr 27, 2020):

So, the software I usually use to measure things like this is called "typometer". It's an old piece of Java software that injects input and waits for the output to be rendered on the screen.

Spooky! :) Looks like it's a valid "testing infra" feature request? (installing Java for end-user input on this can be quite a heavy dependency) :) Cursor rendering speed should also be supported for measurement

@vadimkantorov commented on GitHub (Apr 27, 2020): > So, the software I usually use to measure things like this is called "typometer". It's an old piece of Java software that injects input and waits for the output to be rendered on the screen. Spooky! :) Looks like it's a valid "testing infra" feature request? (installing Java for end-user input on this can be quite a heavy dependency) :) Cursor rendering speed should also be supported for measurement

claunia commented

@vadimkantorov commented on GitHub (Apr 27, 2020):

If you have VS 2019, I welcome you to try out a Release build from master 😄

Good news it's improved, but it's too heavy-lifting for me for now. I wish there were *.msixbundle artefacts in some nightly builds?

@vadimkantorov commented on GitHub (Apr 27, 2020): > If you have VS 2019, I welcome you to try out a Release build from `master` 😄 Good news it's improved, but it's too heavy-lifting for me for now. I wish there were `*.msixbundle` artefacts in some nightly builds?

claunia commented

@DHowett-MSFT commented on GitHub (May 1, 2020):

I'll repurpose this one into the Task for "determine a good way to measure performance". 😄

@DHowett-MSFT commented on GitHub (May 1, 2020): I'll repurpose this one into the Task for "determine a good way to measure performance". :smile:

claunia commented

@DHowett-MSFT commented on GitHub (May 1, 2020):

This will require #4001.

@DHowett-MSFT commented on GitHub (May 1, 2020): This will require #4001.

claunia commented

@vadimkantorov commented on GitHub (Jun 8, 2020):

a related request (also by me) on time-to-rendering measurements and loading: https://github.com/microsoft/terminal/issues/6409

@vadimkantorov commented on GitHub (Jun 8, 2020): a related request (also by me) on time-to-rendering measurements and loading: https://github.com/microsoft/terminal/issues/6409

claunia commented

@lhecker commented on GitHub (Oct 25, 2020):

As a stopgap you can use an app called "Is It Snappy?" with a modern iPhone (those supporting Slow-Mo videos).
At 240 FPS recording speed you can get decently accurate end-to-end latency measurements that way.

@lhecker commented on GitHub (Oct 25, 2020): As a stopgap you can use an app called ["Is It Snappy?"](https://itunes.apple.com/us/app/is-it-snappy/id1219667593?ls=1&mt=8) with a modern iPhone (those supporting Slow-Mo videos). At 240 FPS recording speed you can get decently accurate end-to-end latency measurements that way.

claunia commented

@NelsonMinar commented on GitHub (Apr 8, 2024):

Some 2024 measurements with method explained: https://chadaustin.me/2024/02/windows-terminal-latency/

@NelsonMinar commented on GitHub (Apr 8, 2024): Some 2024 measurements with method explained: https://chadaustin.me/2024/02/windows-terminal-latency/

claunia commented

@lhecker commented on GitHub (Apr 8, 2024):

I have the same monitor as you do, running at 144 Hz, but a better CPU (AMD 5950X) and GPU (RTX 4090), so my experience may be better because of faster hardware. Additionally, I don't have WSL 1 nor Windows 10 available at the moment, which may also result in quite a different experience. Lastly, I've customized my keyboard to have a poll rate of 1000 Hz.

I noticed that your best and worst results in the first table are pretty much exactly 30 and 60 FPS respectively, which is, if I'm entirely frank, a little bit suspicious. I would expect there to be a lot more spread, the results to be generally lower, and for them to not coincide with such common dividers.

For instance, on Windows 11 build 26060, conhost, WSL2 and bash, I consistently get an input latency of 16ms which is 2 frames. For Windows Terminal I get a latency of 25-32ms (3-4 frames).

~~Just to be sure, could you check whether your input latency is better when you disable G-SYNC for your monitor?~~

@lhecker commented on GitHub (Apr 8, 2024): I have the same monitor as you do, running at 144 Hz, but a better CPU (AMD 5950X) and GPU (RTX 4090), so my experience may be better because of faster hardware. Additionally, I don't have WSL 1 nor Windows 10 available at the moment, which may also result in quite a different experience. Lastly, I've customized my keyboard to have a poll rate of 1000 Hz. I noticed that your best and worst results in the first table are pretty much exactly 30 and 60 FPS respectively, which is, if I'm entirely frank, a little bit suspicious. I would expect there to be a lot more spread, the results to be generally lower, and for them to not coincide with such common dividers. For instance, on Windows 11 build 26060, conhost, WSL2 and bash, I consistently get an input latency of 16ms which is 2 frames. For Windows Terminal I get a latency of 25-32ms (3-4 frames). ~~Just to be sure, could you check whether your input latency is better when you disable G-SYNC for your monitor?~~

claunia commented

@NelsonMinar commented on GitHub (Apr 8, 2024):

could you

I am not the person who wrote the blog post I linked so I can't answer your questions, sorry.

@NelsonMinar commented on GitHub (Apr 8, 2024): > could you I am not the person who wrote the blog post I linked so I can't answer your questions, sorry.

claunia commented

5f3a857192/src/renderer/base/RenderEngineBase.cpp (L72-L78)

@zadjii-msft commented on GitHub (Apr 15, 2024):

@chadaustin might be able to answer those questions though - or provide any other thoughts he may have here.

I'd reckon the Terminal's input latency is from the conpty buffering:

@zadjii-msft commented on GitHub (Apr 15, 2024): @chadaustin might be able to answer those questions though - or provide any other thoughts he may have here. I'd reckon the Terminal's input latency is from the conpty buffering: https://github.com/microsoft/terminal/blob/5f3a857192b2fa6f4e5ae87eb3ee59a1571515a2/src/renderer/base/RenderEngineBase.cpp#L72-L78

claunia commented