Ubuntu 13.04 Desktop Comparison: 6 Desktops, 5 Driver/GPUs

[StephanG]

I for one am disappointed by the lack of quality of the article. Don't get me wrong, the data is great. But, with a massive comparison like this, it would have been great to glean a little more insight from the article itself.

For example, the World of Padman V1.2 test, shows that for the Radeon test, KDE got 15 FPS, on hardware that is clearly capable of delivering almost 200 FPS on EVERY other DE, using the same drivers. No matter how bad KDE is at gaming, it shouldn't lag that far behind. Clearly there is some sort of anomaly. If you don't want to take the time to figure out why, at least mention it as an anomaly. Perhaps the test machine had a problem?

Secondly, and yes it's been mentioned before, but I feel it should be mentioned again. The colours are REALLY annoying. You can clearly see KDE, and the rest all look the same.

Thirdly, I don't think you should allow data to be compared for any values that are unplayable, such as 10 FPS. The bars seem to indicate that DE X is slightly better than Y, when in reality, all of them are unplayable.

EDIT: In short, at least give us the impression that you've given a few minutes thought about the results. Currently, it feels like the entire article was simply run by a script and then copied as an 'article'.

[GreatEmerald]

I?m glad someone did bite. Thanks .

I interpreted your post as a joke. It was rather amusing, too. After all, it does say that glxgears is not a benchmark the second you launch it

[duby229]

I am not familiar with the current state of xfce, but I do believe that its compositor is 2d only, you wouldnt notice a difference with 3d apps... and again the reason that glxgears gave differing results is because it is not a benchmark.

[stqn]

I?m glad someone did bite. Thanks .

I just did a few benchmarks with and without Xfce compositing.

glxgears:
without compositing: 1230 fps
with compositing: 1450 fps

Trackmania Nations Forever (lowest possible settings):

1280?960 windowed:
without compositing: 37.6 fps.
with compositing: 37.6 fps.

1920?1200 fullscreen:
without compositing: 29.4 fps.
with compositing: 30.1 fps.

(Intel Core i3 550, linux 3.6.9 64 bit, mesa 9.0.1, xf86-video-intel 2.20.15, wine 1.5.19, vblank_mode=0)

So it appears that under Xfce compositing has no impact on a ?real? game, but improves things on a program that renders very little. (Unfortunately I don?t think I have another game to benchmark.)

[duby229]

Maybe Xfce had compositing enabled and it slows things down?
However, last time I tried the excellent benchmarking tool glxgears under Xfce, enabling compositing actually improved the frame rate significantly!

glxgears is definitely not a benchmark. framerates change by doing nothing more than moving the window. it is absolutely impossible to get comparable data using it. It just isnt possible to get repeatable data.

Again glxgears is not a benchmark. it is useful in showing that direct rendering is working, thats all, and even then its just for a show.

[stqn]

but how can there be significant differences sometimes between xfce and lxde?

Maybe Xfce had compositing enabled and it slows things down?
However, last time I tried the excellent benchmarking tool glxgears under Xfce, enabling compositing actually improved the frame rate significantly!

I am a bit shocked about the differences from the Desktops in gaming... I mean ok I can understand why the "3d"-Desktops perform sometimes bad... but how can there be significant differences sometimes between xfce and lxde? And is there no general rule, which one is faster? I just dont get it... the window manager should open a window in full-screen and then the driver should directly write in this window, does it say to the window manager please draw me a poligon in this window or something like that?

So is there a general rule, if as example I would use something like awesome-wm which is pretty low-end would it perform best in all test? whats the bottle-neck, does the dms eat some ram behind or does they calculate some tetris games its playing with it self? Strange...

but very interesting that at least if you use the right dm in some game radeon driver kicks ass... at least in speed terms the radeon driver improves very well, even I think thats not the most important field...


UPDATE:

another thought... does that differences go away at least in the cases where the desktops switch properly to 2d in background if a 3d window in fullscreen gets startet... with wayland... so is wayland be more direct in rendering games?

With same settings there's no difference between KDE and Unity. I bet there's no difference with others as well. To have full performance just suspend compositions or use unredirect fullscreen windows. This works with Nvidia and I don't know about others.

[blackiwid]

I am a bit shocked about the differences from the Desktops in gaming... I mean ok I can understand why the "3d"-Desktops perform sometimes bad... but how can there be significant differences sometimes between xfce and lxde? And is there no general rule, which one is faster? I just dont get it... the window manager should open a window in full-screen and then the driver should directly write in this window, does it say to the window manager please draw me a poligon in this window or something like that?

So is there a general rule, if as example I would use something like awesome-wm which is pretty low-end would it perform best in all test? whats the bottle-neck, does the dms eat some ram behind or does they calculate some tetris games its playing with it self? Strange...

but very interesting that at least if you use the right dm in some game radeon driver kicks ass... at least in speed terms the radeon driver improves very well, even I think thats not the most important field...


UPDATE:

another thought... does that differences go away at least in the cases where the desktops switch properly to 2d in background if a 3d window in fullscreen gets startet... with wayland... so is wayland be more direct in rendering games?

[bug77]

I wonder how much NVIDIA has paid Michael for those precise and fair tests in his article

Feel free to publish some tests proving him wrong.

[gutigen]

yeah.. really.. especially considering AMD card costs ~160$, Intel graphics are integrated in CPU, and GTX is just a high-end card for ~500$

EDIT:


ha ha, you were few minutes faster on posting exactly same thing I was thinking

I wonder how much NVIDIA has paid Michael for those precise and fair tests in his article

Also why test modern GPU with old, OpenGL 2.0 games when you got Serious Sam, TF2 and Unigine Heaven out there... WTF