vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

Cdh · 2011-02-14 15:45:55

Hi,

I have a netbook with intel i945gme.

Since last update with mesa 7.10 running glxgears is not smooth and the fps are ~ 1/2 of the screen refresh rate.

chris@chriseee ~ % glxgears
Running synchronized to the vertical refresh.  The framerate should be
approximately the same as the monitor refresh rate.
192 frames in 5.1 seconds = 37.623 FPS
146 frames in 5.0 seconds = 29.159 FPS
^C
130 chris@chriseee ~ %

But it does work without vsync. It is really smooth then (and proves that the fps are higher then 1/2 of the screen refresh rate when not limited):

chris@chriseee ~ % vblank_mode=0 glxgears
ATTENTION: default value of option vblank_mode overridden by environment.
ATTENTION: default value of option vblank_mode overridden by environment.
1633 frames in 5.0 seconds = 326.594 FPS
1639 frames in 5.0 seconds = 327.623 FPS
^C
130 chris@chriseee ~ %

I have no xorg.conf and no ~/.drirc.

Can anyone reproduce?

Ramses de Norre · 2011-02-14 15:51:09

Working fine here on GM965/GL960 (that's what lspci calls my video adapter, I'm not sure which of those two numbers specifies my GPU).

flan_suse · 2011-02-14 19:32:10

Try glxgears again, without disabling vsync, but as the gears are turning, keep moving your mouse cursor around (never let your cursor idle) and then see what FPS you get. I'm betting you'll see ~60 fps.

Ever since intel drivers >= 2.10, video performance has dropped, and problems have increased. At least for me and others that have noticed the same thing.

It has something to do with a poor power-saving implementation, if I remember correctly.

Cdh · 2011-02-14 21:10:21

Yes, you are right. When moving the mouse pointer it keeps near 60 fps and is smooth. So it's more a redrawing issue instead of actual performance? Intuitively I would have thought that compositing would solve the problem but it doesn't.

I also created an AUR package for mesa-full with gallium for i915:
http://aur.archlinux.org/packages.php?ID=30028

With gallium it happens too...

Gusar · 2011-02-14 21:29:18

How 'bout you guys just stop paying attention to glxgears? Seriously.

Cdh · 2011-02-14 22:35:27

It's not about glxgears. It's about vsync broken which glxgears shows. catalyst once had a similar problem and there it also was not only glxgears but glxgears showing it.

Gusar · 2011-02-14 22:43:50

Can you then report issues in another app?

Cdh · 2011-02-15 00:38:53

Hm, actually I don't have so much 3d Applications on my netbook at the moment. But some basic games seem to run fine.
Maybe you are right that it's only glxgears in this case.

It just feels strange that one of the most basic 3d animations has such issues and everything besides it would work fine.

flan_suse · 2011-02-15 04:20:18

Gusar wrote:

Can you then report issues in another app?

For me, personally? YouTube / Flash and even Neverball and Osmos. If I let the mouse idle, the performance begins to "jerk" a bit. It has gotten better with each new iteration of the intel driver, but it was noticeably worse with 2.10 through 2.12. Now it's not really too big of a deal (but it's still worse than 2.9.1 and earlier.)

I understand this "glxgears is not a benchmark" argument, but it does allow you to observe a change. For instance, running glxgears with intel driver 2.9.1 would not change framerate based on idle/active cursor. Then with 2.10+, it suddenly does, so it at least marks a noticeable difference that you can refer to if you have other issues. "I noticed such and such after I updated from version X to version Y. Maybe that's why this crashes after the recent updates."

Last edited by flan_suse (2011-02-15 04:20:50)

flan_suse · 2011-02-15 04:24:09

Cdh wrote:

Yes, you are right. When moving the mouse pointer it keeps near 60 fps and is smooth. So it's more a redrawing issue instead of actual performance?

My honest answer is that I don't know, but I remember reading a bug report about this where it was mentioned that this has something to do with a power-saving feature.

flan_suse wrote:

It has something to do with a poor power-saving implementation, if I remember correctly.

If I ever find that bug report again, I'll post it in this thread. Maybe others can make more sense of it who have more knowledge about the matter.

I almost forgot to mention, it doesn't happen with full screen applications, so when playing things in full screen (not windowed) mode, I don't notice this effect.

Last edited by flan_suse (2011-02-15 04:59:12)

Gusar · 2011-02-15 10:10:35

flan_suse wrote:

I understand this "glxgears is not a benchmark" argument, but it does allow you to observe a change. For instance, running glxgears with intel driver 2.9.1 would not change framerate based on idle/active cursor. Then with 2.10+, it suddenly does, so it at least marks a noticeable difference that you can refer to if you have other issues. "I noticed such and such after I updated from version X to version Y. Maybe that's why this crashes after the recent updates."

Well, I've seen all sorts of stuff with glxgears (smooth animation, not smooth animation, vsynced, not-vsynced, jerkiness unless I constantly move the mouse...) on my machines as driver versions came and went, but if I wanted to play a game, it always worked the same. I've also never seen differences in Flash behavior. Flash always sucks

kokoko3k · 2011-02-15 10:38:34

Gusar wrote:

How 'bout you guys just stop paying attention to glxgears? Seriously.

How about to stop to be paranoid as soon as someone type "glxgears" too?

Somebody screaming "glxgears is not a benchmark!"
...Then someone uses teapot
https://bugzilla.redhat.com/show_bug.cgi?id=582861

Bug reference:
https://bugs.freedesktop.org/show_bug.cgi?id=30364

Last edited by kokoko3k (2011-02-15 10:42:33)

Gusar · 2011-02-15 12:09:38

kokoko3k wrote:

How about to stop to be paranoid as soon as someone type "glxgears" too?

I've seen more than once "glxgears shows 'blah', what do I do?", and when you ask if they have problems elsewhere, they say "no". It's not paranoia, it's that glxgears is simply bad at everything. That redhat link you posted, among others points to this: http://qa-rockstar.livejournal.com/7869.html

Last edited by Gusar (2011-02-15 12:11:47)

Arch Linux

#1 2011-02-14 15:45:55

vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#2 2011-02-14 15:51:09

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#3 2011-02-14 19:32:10

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#4 2011-02-14 21:10:21

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#5 2011-02-14 21:29:18

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#6 2011-02-14 22:35:27

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#7 2011-02-14 22:43:50

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#8 2011-02-15 00:38:53

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#9 2011-02-15 04:20:18

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#10 2011-02-15 04:24:09

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#11 2011-02-15 10:10:35

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#12 2011-02-15 10:38:34

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

#13 2011-02-15 12:09:38

Re: vsync broken with i915 since [mesa/intel-dri/libgl]-7.10

Board footer