Intel Linux Graphics Performance Q4'08

[josvazg]

Now I benchmarked on Fedora 10 too and there I got 48fps in Tremulous, so the regression in Ubuntu 8.10 seems to be Ubuntu specific and not an upstream one.

Sorry but I tested a Fedora 10 live on my Asus eee 901 and the glxgears performance was about the same as with Ubuntu 8.10, rendering errors INCLUDED.

So, at least in those eee, which were supposed to be the big opportunity for Linux on the desktop, the intel drivers suck in Fedora as well as in Ubuntu.

[chaos386]

Now I benchmarked on Fedora 10 too and there I got 48fps in Tremulous, so the regression in Ubuntu 8.10 seems to be Ubuntu specific and not an upstream one.

It might still be upstream, since Fedora 10 uses newer packages than Ubuntu 8.10. Mesa 7.3 vs 7.2 and Intel 2.5 vs 2.4, for example.

[whizse]

Sorry but I tested a Fedora 10 live on my Asus eee 901 and the glxgears performance was about the same as with Ubuntu 8.10, rendering errors INCLUDED.

Rendering errors? You filed a bug report about those, right?

[josvazg]

Rendering errors? You filed a bug report about those, right?

Yes the bug report is here:
https://bugs.launchpad.net/ubuntu/+s...el/+bug/252094

As you can see there is plenty of people subscribed to this one.

Some people have experienced better performance with Fedora 10, and some others (like me) haven't. Probably it depends on the chipset. [I don't recall anyone using a eee 901 having "good" performance with 2008 distros.]

[kxmas]

Today I noticed, with a clue from a bug report, that somehow the Ubuntu installer failed to add the initial user to the 'video' group. That would be fine if the DRI device defaulted to global read-write permissions.

ls -l /dev/dri/card*
crw-rw---- 1 root video 226, 0 2009-01-01 14:45 /dev/dri/card0

I'm running version 2:2.5.1-1ubuntu5~intrepid of xserver-xorg-video-intel. I had to rollback from the experimental stuff due to problems with dual displays that I didn't care to work through.

Adding my account to the group 'video' resulted in a huge speed up of glxgears.

I hope this helps someone

[whizse]

Yes the bug report is here:
https://bugs.launchpad.net/ubuntu/+s...el/+bug/252094

As you can see there is plenty of people subscribed to this one.

Some people have experienced better performance with Fedora 10, and some others (like me) haven't. Probably it depends on the chipset. [I don't recall anyone using a eee 901 having "good" performance with 2008 distros.]

Yikes! That is one messy bug report, I pity the bug triagers

It's probably a much better idea to grab stuff directly from git, file a bug report directly upstream, and keep things simple: one problem per bug report.

In my experience, things like the aforementioned rendering errors are usually fixed quickly.

[tormod]

Yes the bug report is here:
https://bugs.launchpad.net/ubuntu/+s...el/+bug/252094

Adding your (different) issues to someone else's bug report does not count as reporting. Rendering issues are not even mentioned in the bug description, so please report a new bug.

[tormod]

Today I noticed, with a clue from a bug report, that somehow the Ubuntu installer failed to add the initial user to the 'video' group. That would be fine if the DRI device defaulted to global read-write permissions.

ls -l /dev/dri/card*
crw-rw---- 1 root video 226, 0 2009-01-01 14:45 /dev/dri/card0

This is https://bugs.launchpad.net/bugs/306014
I would think this is only a problem in Jaunty (which is a development version, by the way, and not yet in "beta" like some people seems to believe) and not in Ubuntu 8.10. If you have been installing experimental packages on top that could explain it. If it's a clean install, please report it, and add a link here.

[kxmas]

This is https://bugs.launchpad.net/bugs/306014
I would think this is only a problem in Jaunty (which is a development version, by the way, and not yet in "beta" like some people seems to believe) and not in Ubuntu 8.10. If you have been installing experimental packages on top that could explain it. If it's a clean install, please report it, and add a link here.

Yeah, I don't know where exactly which set of assumptions were made by which packages to cause my permissions problem. I also don't know how much I contributed when I started using experimental packages.

Not adding the initial user to video group could be argued as a feature or as a bug. I do know that in the group/user editor has had an option about accessing video acceleration hardware for a long time. If the group existed, and was referenced in the admin tools, but wasn't being used, that sounds like a bug, albeit a benign bug, to me. I'm glad it's going to be fixed in jaunty.

[tormod]

Not adding the initial user to video group could be argued as a feature or as a bug. I do know that in the group/user editor has had an option about accessing video acceleration hardware for a long time. If the group existed, and was referenced in the admin tools, but wasn't being used, that sounds like a bug, albeit a benign bug, to me. I'm glad it's going to be fixed in jaunty.

Yes, it's unclear whether these groups should be used (or shown), see also https://bugs.launchpad.net/bugs/188759