News for hackers. Stuff that matters.
Re:What Metric to Use When Benchmarking? by ajuc on Friday July 01, 2022 @03:53AM (#31943611)
In interactive programs (for example games) it's often more important to be consistent than to be fast.

For example a game which runs at 120 fps but every 10 seconds has 1 frame that takes 1/30th of a seconds feels awful.

A game that runs at constant 60 fps feels much better.

In this case it's better to just count the frames that took too long and by how much.

Re:What Metric to Use When Benchmarking? by zigzag312 on Friday July 01, 2022 @06:10AM (#31944416)
This. When optimizing realtime apps, focus on worst cases.

Also, FPS is inconvenient metric because framerate is not a linear scale. You can't say a task takes x FPS to compute, but you can say it take x milliseconds, since time is a linear scale.

For example: if a task takes 10 ms and other stuff takes 1 ms. FPS delta for the task is ~909 FPS (1000 FPS - 91 FPS). But if other stuff takes 20 ms, we get only a ~17 FPS difference (50 FPS - 33 FPS).