I am not much of a systems programmer or developer of any sort. I don’t have the experience to write this app myself. To be clear, this thread is not a request from me asking Vaxry (or anyone) to implement a feature. I aim to distance myself from any such sense of entitlement. I realize Hyprland is an open source project and users are welcome to contribute on their own terms. The intent of this thread rather is to share a quirky observation and offer an interesting possibility.
Summary: rasterman’s E17 window manager led the X11 ecosystem in performance in 2005 in benchmarks. A similar reproducible benchmarking technique for modern Wayland compositors would serve as an interesting comparison of frame rates and responsiveness across different visual effects, plugins, bars, and workloads which would establish common standards, detect regressions, and measure improvements.
All the way back in May 2005 when E17 was a Development Release and the GeForce 6600 GT graphics card and Intel Pentium 4 3.4Ghz were mid-range hardware, E17 lead developer rasterman made a personal blog post showcasing a way to measure and compare E17’s performance with other window managers. That’s a link to raster’s blog post on the Wayback Machine. Here is the screenshot embed:
To describe rasterman’s results in the image copied above: E17 configured with certain modules and smart placement plugins disabled came out as the fastest, drawing 267.572 windows per second. However with modules and smart placement enabled, E17 reached 103.462 windows per second. By comparison, Metacity drew 73.858 windows per second while KDE 3.3 was drawing 34.185 and Kwin 3.2 was drawing 30.620. E17 was clearly the most performant. As a reminder, this was in 2005 so as you can tell by the software versioning, this was all a very long time ago. This was a time when Sawfish, Metacity, Xfwm4, Icewm were common. Torvalds had just released git and E17 code was hosted using CVS. How the landscape has changed!
It’s a given that Hyprland is a tiling WC whereas Wayfire is a stacking WC so they are completely different UI paradigms …but Hyprland is still able to float a tile like a window and now wobble. It would be interesting to compare frames per second as well as response time on Hyprland:
- With blurry terminals and wobbly windows turned on or off
- With all plugins disabled in a vanilla scenario vs. having multiple plugins enabled
- With different bars (Quickshell, Waybar) vs. no bar
- Using different high end vs. low end GPUs
- Using different GPU vendors in the same class like performance on an AMD Radeon RX 9070 XT vs. an Nvidia Geforce RTX 5070
You could stress test performance with +1,000 floating windows open at the same time compared to having just 3 open (or any arbritrary number range of windows in between).
My conjecture is that other WC’s which offload rendering to the GPU like Wayfire and Niri would be close to Hyprland in terms of performance whereas the single threaded JS-based Mutter would come dead last, like Xfce did in X11 in 2005.
With his talent for systems programming and expert C++ algorithmic design, I’d wager that @vaxry would make sure Hyprland is #1. Bragging rights aside, a WC benchmarking tool would be useful for regression testing from one release to the next and measuring incremental progress in a reproducible way. The tool could even be integrated into Michael Larabel’s Phoronix Test Suite to establish a WC ecosystem benchmarking standard and to demonstrate true Hyprland superiority.
Thoughts?