Hacker News Reader

For those of you who are application developers on Linux/Wayland, do you use any other methods for benchmarking true latencies between input events and screen? Less heroic than the high-speed camera method. I'm embarrassed I've still never figured this out properly.

View on HN

A photocell is I think the standard (or at least "the standard" in psychophysics applications I have also used). You need, though, to make sure your event causes a change that a simple photocell would detect (eg you have a corner of your screen or application window having a small square that turns from black to white and vice versa). You can connect that to an arduino or whatever you can use to get both signals in some low latency way (one from user interaction or whatever causes the event, and one from the event itself with that light intensity change). There are also photocells with a higher granularity for luminosity changes than just light/dark, so maybe you do not need to have the black/white square thing, just something that causes enough luminosity change to be detected. This should, in principle, give you very accurate timings if you can also get the event initiator somehow with low latency (or you can do stuff like simulate an event periodically and look for skipped frames/variance in the screen presentation rather than latency).

You might like Typometer. I believe it was created by a JetBrains developer to improve input latency in IDEA.

> Typometer works by generating OS input events and using screen capture to measure the delay between a keystroke and a corresponding screen update. Hence, the measurement encompasses all the constituents of processing latency (i. e. OS queue, VM, editor, GPU pipeline, buffering, window manager and possible V-Sync).

https://pavelfatin.com/typing-with-pleasure/

Reply on news.ycombinator.com