Does Compiz Still Slow Down Your System?

[deanjo]

Hi Michael,

I have noticed too that Compiz mostly doesn't decrease performance. However, KWin does and the difference can be huge. It would be cool if you made an article that compares:
- Gnome without Compiz
- Gnome with Compiz
- KDE without KWin
- KDE with KWin

I bet the numbers will be very interesting.

With full screen apps you shouldn't see any difference in performance if Kwin is configured with:

UnredirectFullscreen=true

This setting is on by default in openSUSE.

[mtippett]

What happened to the Catalyst team starting around late 2009 and through 2010 to the present? Their driver stopped sucking. Maybe they got lucky and hired one or more brilliant minds.

Now if only we can make it not suck and make it open source. Damn. The numbers for r?00g are encouraging though.

Late 2009 is when the driver hit a sweet spot. The team (my team) had basically got the tempo and vibe to match the development speed of Windows for the Radeon HD 5000 series launch.

Driver development with 10's of millions of lines of code is slow and steady (particularly with the ecosystem continually changing). ATI is great at maintaining quality as it gets there.

I made the move from AMD when I felt we had hit a good spot (late 2009), from there they've continued to hold the ship steady and continue pushing out good features.

[pingufunkybeat]

I've made a quick check with glxgears:

PCI 4350:
KWin (+compositing): 240
KWin (-compositing): 440

PCI-E 4550:
KWin (+compositing): 2400
KWin (-compositing): 3050

Glxgears is not a benchmark, but it does stress the copying. The terrible performance of the PCI card and the huge hit it gets from compositing make me believe that the problem is in the extra blit, caused by the compositing environment.

If I recall correctly (it's all muddy and I'm not a developer), the extra blitting causes a large hit with Mesa drivers in general. This is also why the KMS path was so much slower when it was first introduced, right?

The binary blobs have it really well-optimised.

That's why a check with a non-compositing window manager would be useful.

[marek]

Hi Michael,

I have noticed too that Compiz mostly doesn't decrease performance. However, KWin does and the difference can be huge. It would be cool if you made an article that compares:
- Gnome without Compiz
- Gnome with Compiz
- KDE without KWin
- KDE with KWin

I bet the numbers will be very interesting.

[pingufunkybeat]

FWIW, I have a PCI card (not express) in my computer, and there is a very apparent difference between fullscreen video (no compositing) and windowed video (composited environment) in KWin.

[allquixotic]

What happened to the Catalyst team starting around late 2009 and through 2010 to the present? Their driver stopped sucking. Maybe they got lucky and hired one or more brilliant minds.

Now if only we can make it not suck and make it open source. Damn. The numbers for r?00g are encouraging though.

[pingufunkybeat]

Now we know how to make Michael do new benchmarks

Does Metacity in Ubuntu use compositing? Because that's what made 3d apps slower under Compiz. If Metacity uses compositing as well now, then it's no wonder that the results are closer.

[blacknova]

No it doesn't. But it is annoying that desktop behavior (shortcuts, virtual screens and so on) changes with compiz and without it.

[curaga]

Nice of you to cast doubt of FUD into conclusions based on your own article

[Raine]

Gallium3D

Looking at those results, mostly on Nvidia side, makes me wonder how much efforts are being dedicated by Gallium3D team to improve performance on Gallium3D infraestructure.

Does anyone know if there is any work beeing done to improve Gallium3D infraestructure performance in the wild?

I didn't notice any special/cool patch for Gallium3D on 2.6.38. I could be wrong, and wish to Please correct me if i'm wrong....