Tuesday, September 21, 2010

Are Your Desktop Effects Slowing You Down?

Whenever I perform a 3D benchmark in Linux one of the first questions I get asked about the results is:

Where your desktop effects turned off?

For those who are not aware, desktop effects are the "flash" that is enabled by default in many popular Linux distros (namely Ubuntu and it's derivatives) such as the wobbly windows, desktop cube, and sleek sliding effects. Something that has always been questioned is whether or not desktop effects slow down your 3D performance in other applications and if so, how much do they slow it down by?

The Test:
I am going to use Unigine Benchmarks on Pinguy OS (for Gnome) and Chakra (for KDE) with and without desktop effects enabled. Both systems are clean, fully up to date installs with the nVidia 260 beta driver installed. Gnome will be using Compiz for it's desktop effects and KDE will be using Kwin.

The Hardware:
I'm using my same gaming rig that I've used for all my other benchmarks: Processor - Intel p9700 2.8ghz Dual Core, RAM - 4gigs of DDR3 1066, Video Card: nVidia 260m with 1gig DDR3 dedicated memory.

The Results:
Lets dive into some graphs shall we?

Graphs can be hard to read exact numbers on, so here is the numerical breakdown of the results:

As you can see, desktop effect do indeed decrease your overall 3D performance. Compiz is far more detrimental to 3D scores than KDE's Kwin is. Across all three benchmarks having Compiz enabled on the system caused a 10.7% performance decrease, while Kwin only caused a 1% decrease.

So it appears that if you are going to be gaming on Ubuntu/Gnome it is worth taking the time to toggle Compiz off before loading up that game.

~Jeff Hoogland
Please note while these benchmark scores presented are accurate to the best of my abilities, they only represent my personal hardware and software configurations. Your results on your own system(s) may vary (and if they do, please share them!).


  1. These tests confirm what is immediately obvious on many good PCs. 3-D desktop effects are shameless resource hogs that cause sluggish desktop user experiences.

    I find it unacceptable for a typical 20% CPU usage before you even start to use applications.

    A much better compromise is to use Xrender based effects (mostly fade-in/out and translucency) instead of OpenGL. I use Cairo CM. It's not perfect, but CPU usage is much lower.

  2. These are interesting test results. It makes me want to give a KDE-based distro another spin. One thing I wonder, though, is whether the KDE-Kwin and Gnome-Compiz differences could be attributed to the distros you ran. For example, in your earlier test on best distro for 3d performance, Chakra scored highest, while Ubuntu (which Pinguy is based on) came in near the bottom. I wonder if the Gnome-Compiz results would have been different had you ran Fedora, Sabayon or Lint Minux Debian?

  3. This seems to lend support to KDE's decision that Compiz was too bloated and messy to use by default in the $.x rewrite of KDE years ago, which led to KWin's rewrite to more or less duplicate these effects, but much more cleanly.



    Would like to see the same benchmark also compared against 7 if possible.

  5. And presumably it makes a difference how many compiz effects are enabled? On my laptop I have wobbly windows and window previews enabled, I can't do without these :D I have the compiz cube on my desktop machine. It would be good if compiz could automatically detect a game running full screen and automatically disable effects an re-enable them afterwards.

  6. Yes, exactly; Aero vs Non-Aero. By the way, this is really interesting stuff. It's the kind of thing that a lot of us have known or pondered for a long time, but you are really presenting it in a great way.

  7. Thanks FEWT. I actually just reinstalled Windows 7 to get some updated OpenGL and DirectX scores. I'll do some Areo vs non-Aero as well. Should be able to get a posting done within the next week.

  8. Thanks for the great article. I think it's very nice to be able to turn on the effects if you want to. There are some very flashy things in compiz/beryl that are really cool for show. But I think think everything should be set to basic for default. The first thing I always do is turn off desktop visual effects when using Ubuntu, and I would do the same with windows if I used it. Performance is king. In the words of Marissa, never underestimate "fast".

  9. Hey Jeff, If you are doing a benchmark with windows7 can you compare it with Ubuntu as well? So we can really put the 10% in to perspective.

    @Shannon - I like fast as well but since I don't use Ubuntu for gaming and i have a decent machine I keep compiz on. I find it easier to get by with 'expo' and 'scale' to manage my applications.

  10. The simple solution on a Linux box is to create a separate gaming account...

  11. I keep getting impressed by the really interesting articles on your blog.
    Keep them comming!

    One benchmark I'd find interesting is ArchLinux with KDE vs Chakra.
    Since Chakra is the distro that seemingly performs the best out of all benchmarks it would be interesting to see how it performs vs ArchLinux, from which it is derived.

    I'm using ArchLinux myself. If Chakra happens to get better scores, I might make the switch.

  12. This is quite contrary to what I encountered with Kwin. I sometimes play World of Warcraft with wine (opengl based). Turning on Kwin gives me only roughly half of the performance (nvidia).

  13. My rig isn't quite as powerful as the Jeff's test machine, but frankly, I don't think I miss that 10% I'm losing by running with desktop effects.

    Back in the days when I was overclocking my rig and doing all sorts of foolishness to get better clock-speeds (custom cooling apparatus, etc) a 10% performance change would have been earthshattering. Now, it seems fairly meaningless, and if it really bothered my I could go upgrade my processor.

    I would be interested, though, in seeing how a quad core processor performs vrs. a dual in this test. While the graphics card seems like a likely bottleneck, it probably isn't pumping many instructions from the window manager during a gaming session. Certainly not 1 in 10?

  14. @Fred

    "I sometimes play World of Warcraft " does NOT mean much. You need to be more specific and present real numbers to be credible.

    About the test, it is pretty good but would have been more accurate if the same distro was used. i.e. Ubuntu & Kubuntu since they are basically the same except for the DE and effects generator.

  15. Jeff, as you say, "Graphs can be hard to read exact numbers on". It's easier to read them if you turn off spurious 3-D effects. Edward Tufte's books are a good read and an excellent reference for how to present information. He will educate you about "chartjunk".

  16. I run a Sony VAIO Notebook which has a medium range GM45 Intel graphic chip. I run Kubuntu 10.4 with KDE 4.5.1. Without Desktop Effects Stellarium gives me an idle setting of 15 fps. With desktop effects it is 5 fps.

    I imagine that if I was running a high end chip the difference wouldn't be so noticeable, and with a low end chip the difference would be much worse.

    Playing combat games on Second Life I found that those who used fast graphics could easily outfight those who did not. You are often killed before you see their weapon fire.

  17. I look forward to the obvious Aero vs. Non-Aero results. :)