Intel 945GM, Xorg, Kernel - performance

eldragon · 2009-04-17 14:31:49

gborzi wrote:

I don't think DRI is really needed, I tried using the "bare" ubuntu xorg.conf and got similar results.
Section "Device"
      Identifier      "Configured Video Device"
      Driver      "intel"
EndSection
Take a look at /var/log/Xorg.0.log to see which option is really used.

ive been reading, and only options being used are the ones i proposed, it tries to use DRI2 and fails (normal)

gborzi · 2009-04-17 17:36:58

I tried separately "MigrationHeuristic" "greedy" and "AccelMethod" "XAA". The former doesn't affect performance at all, the latter gives a worser performance for glxgears, 3D games are unaffected. Obviously using intel-legacy.

eldragon · 2009-04-17 18:04:47

gborzi wrote:

I tried separately "MigrationHeuristic" "greedy" and "AccelMethod" "XAA". The former doesn't affect performance at all, the latter gives a worser performance for glxgears, 3D games are unaffected. Obviously using intel-legacy.

one last thing i had to do in order go get descent 3d performance is include "intel_agp drm i915" to the list of MODULES in /etc/mkinitcpio.conf, i still dont know why i have to.

im using 2.7.0 from aur's xf86-video-intel-newest, give it a go, and reinstall kernel26 for it to take effect. oh, you will have to reboot

EDIT: and for accelmethod, im using EXA, not XAA, i havent tried with XAA.

Last edited by eldragon (2009-04-17 18:05:20)

gborzi · 2009-04-17 19:07:21

eldragon wrote:

one last thing i had to do in order go get descent 3d performance is include "intel_agp drm i915" to the list of MODULES in /etc/mkinitcpio.conf, i still dont know why i have to.

I think it's needed for Kernel Mode Setting, see here. I've used the "late" activation. Did you put "options i915 modeset=1" in /etc/modprobe.conf ?

im using 2.7.0 from aur's xf86-video-intel-newest, give it a go, and reinstall kernel26 for it to take effect. oh, you will have to reboot

I've tried xf86-video-intel-newest when it used the release candidates, the latest is the same as the xf86-video-intel in testing, same source and no patches or different configure options.

EDIT: and for accelmethod, im using EXA, not XAA, i havent tried with XAA.

I'm not shure, but I think the old XAA acceleration isn't available with the new driver.

eldragon · 2009-04-17 19:12:22

gborzi wrote:

eldragon wrote:
one last thing i had to do in order go get descent 3d performance is include "intel_agp drm i915" to the list of MODULES in /etc/mkinitcpio.conf, i still dont know why i have to.
I think it's needed for Kernel Mode Setting, see here. I've used the "late" activation. Did you put "options i915 modeset=1" in /etc/modprobe.conf ?
im using 2.7.0 from aur's xf86-video-intel-newest, give it a go, and reinstall kernel26 for it to take effect. oh, you will have to reboot
I've tried xf86-video-intel-newest when it used the release candidates, the latest is the same as the xf86-video-intel in testing, same source and no patches or different configure options.
EDIT: and for accelmethod, im using EXA, not XAA, i havent tried with XAA.
I'm not shure, but I think the old XAA acceleration isn't available with the new driver.

yes, i know its used for KMS, but after an update, i had huge performance issues with EXA.. compiz would give a white cube, and lots of problems....adding those modules to the initial load fixed those issues for me. even if im not using KMS, i had to do it. no harm there though.

i suggest you do that and report back. maybe you are having the same issue.

about the aur package, i cannot find the i686 version in testing, so im using AUR's, its true its the same. when the i686 version reaches testing, i'll switch

gborzi · 2009-04-17 20:31:04

eldragon wrote:

yes, i know its used for KMS, but after an update, i had huge performance issues with EXA.. compiz would give a white cube, and lots of problems....adding those modules to the initial load fixed those issues for me. even if im not using KMS, i had to do it. no harm there though.
i suggest you do that and report back. maybe you are having the same issue.
about the aur package, i cannot find the i686 version in testing, so im using AUR's, its true its the same. when the i686 version reaches testing, i'll switch

I tried your suggestion, i.e. to add intel_agp and i915 in mkinitcpio.conf and rebuild the images (I actually used "mkinitcpio -p kernel26" to rebuild the images, instead of reinstalling). The performance is still bad but I didn't see the slightly garbled menus.
glxgears gives 150-200 FPS, gtkperf -a takes 39s, assaultcube unplayable at ~8 FPS. I also retried KMS with the early activation, getting better performances (but still well below the intel-legacy driver) but garbled menus and strange behaviour for gtkperf (no visible ComboBox). So I have reverted to intel-legacy and disabled KMS, but left intel_agp and i915 in kernel images. Perhaps the difference is due to the different architecture we two are using, i.e. x86_64 and i686.

karabaja4 · 2009-04-17 20:49:48

I wonder if intel cards will ever work properly again in 3D (compared to performance in Windows XP with the same card)?

Llama · 2009-04-18 13:31:03

I did everythins as in Allan's howto; it works, but I have a couple of questions:

(1) The boot process looks quite ugly. First, it runs in a default vga resolution; then it changes somewhere about the loading of modules to native (1280x800), but runs on the left half of the screen; after entering runlevel 3 it runs normally, at last. BTW, I had to put intel_agp and i915 in MODULES=(...) (/etc/rc.conf) which isn't a part of Allan's howto. Does it ring a bell?

(2) After that, X doesn't start any faster. Yes, Ctrl-Alt-F1 becomes really snappy . Is it the much sought after benefit ?

eldragon · 2009-04-18 14:16:21

karabaja4 wrote:

I wonder if intel cards will ever work properly again in 3D (compared to performance in Windows XP with the same card)?

im sure they will, we just have to wait a bit.... after all, there have been important changes to the driver these last monthes...

Damnshock · 2009-04-18 14:33:21

karabaja4 wrote:

I wonder if intel cards will ever work properly again in 3D (compared to performance in Windows XP with the same card)?

Wanna know what? I do not know how my intel card worked on Windows 'cause never actually used it. First thing I did when I got this laptop was install archlinux on it!

Anyway, I just set up KMS and the performance has imprived quite a bit! KDE 4.2 with composite effects works almost perfectly.

Drawbacks? Screen is "cut" on my external monitor while I'm at the terminal, I mean it does not get to the bottom although the resolution seems to be correct (1920x1080).

Nothing else to say after 20minutes of using it, I'll be back if I find something worth to mention!

breakdown · 2009-04-18 14:58:30

Llama wrote:

(1) The boot process looks quite ugly. First, it runs in a default vga resolution; then it changes somewhere about the loading of modules to native (1280x800), but runs on the left half of the screen; after entering runlevel 3 it runs normally, at last. BTW, I had to put intel_agp and i915 in MODULES=(...) (/etc/rc.conf) which isn't a part of Allan's howto. Does it ring a bell?

Just compile intel_agp and i915 as built-in kernel options instead of modules, and should work.

gborzi · 2009-04-19 02:05:16

Finally I got xf86-video-intel 2.6.3 working with good performance, better than xf86-video-intel-legacy. Without enabling KMS, I only added

Option      "Tiling" "false"

in xorg.conf. I decided to try this option after reading some error messages in /var/log/Xorg.0.log about the failure to allocate buffer for tiling. Googling I found some suggestions about disabling tiling.
My current benchmarks
glxgears ~700 FPS, gtkperf -a ~18 seconds, assaultcube ~50 FPS, torcs 15-20 FPS, tremulous ~60 FPS.
And compiz at 140+ FPS.

Hope this helps.

moljac024 · 2009-04-19 06:57:22

gborzi wrote:

Finally I got xf86-video-intel 2.6.3 working with good performance, better than xf86-video-intel-legacy. Without enabling KMS, I only added
Option      "Tiling" "false"
in xorg.conf. I decided to try this option after reading some error messages in /var/log/Xorg.0.log about the failure to allocate buffer for tiling. Googling I found some suggestions about disabling tiling.
My current benchmarks
glxgears ~700 FPS, gtkperf -a ~18 seconds, assaultcube ~50 FPS, torcs 15-20 FPS, tremulous ~60 FPS.
And compiz at 140+ FPS.
Hope this helps.

And what chipset ? 945G, GM965... something else ?

EDIT: Ok, that was a stupid question, considering the thread title....what happened to the option of deleting your post ?

Last edited by moljac024 (2009-04-19 07:34:07)

Dr4go · 2009-04-19 09:46:20

@ gborzi: Yes, there are a few bugs in the current drm-intel code inside the intel driver and the linux kernel. Just wanted to tell, that - as soon as v2.6.30-rc3 comes out - you should re-enable tiling on 945 and/or 965 again. Well... at least when you are using the Intel driver v2.7.

The combination of v2.6.30-rc3 and v2.7 should _finally_ work awesome.

gborzi · 2009-04-19 12:31:33

@moljac024
It's not a stupid question, I've read some post in this thread by people with a different GPU. Mine is a 945GM.
@Dr4go
From what I've read on the net (mainly an ubuntu bug report and the following discussion) it's a bug in the i915 kernel module code. There is a kernel26 (2.6.29-4) in testing now and I've read that with kernel26-2.6.29-2 (it was in testing, never hit the core repo) things worked fine and broke with the current kernel26. I'll try this new kernel and the intel driver 2.7.0 and report the results.

Shapeshifter · 2009-04-19 13:50:09

Turning off tiling fixed the performance issues for me. Awesome!

Two things are still broken:
- Weird transparency issues when setting opacity <100% on a window
- Sub-pixel font smoothing on these transparency-broken windows.

This screenshot illustrates this quite well. The windows on the left have 95% transparency, the ones on the right have 100%. The left ones are broken, and font smoothing ain't working. I also miss font smoothing in urxvt, although there's no transparency-brokenness:

Anyone else seeing this? Solutions?

Last edited by Shapeshifter (2009-04-19 14:01:18)

gborzi · 2009-04-19 15:20:47

@Shapeshifter
I don't have this problem, but it may appear only for some theme/gtk engine/application. So, which theme, gtk engine and application did you use for the above screenshots? BTW, in the meantime I tried xf86-video-intel from testing with the current core kernel26 and disabled tiling. It works like 2.6.3 for gtkperf, compiz and glxgears, but the 3D games I mentioned previously are slower, expecially assaultcube.
Also, with intel-legacy I experienced some Xorg crash while playing with glest, now with intel 2.6.3 it doesn't crash anymore.

eldragon · 2009-04-19 17:17:19

Shapeshifter wrote:

Turning off tiling fixed the performance issues for me. Awesome!
Two things are still broken:
- Weird transparency issues when setting opacity <100% on a window
- Sub-pixel font smoothing on these transparency-broken windows.
This screenshot illustrates this quite well. The windows on the left have 95% transparency, the ones on the right have 100%. The left ones are broken, and font smoothing ain't working. I also miss font smoothing in urxvt, although there's no transparency-brokenness:
http://stuff.moritzg.ch/kmsfail3.png
Anyone else seeing this? Solutions?

thats a known bug when using UXA....the infamouse alpha channel gets lost somewhere, and then treated FF, which in turn when added to background saturates the image....i dont know if there is a fix for it yet.....live with it, or use EXA

vinvin · 2009-04-20 10:24:45

It seems that putting the Tiling option to false is the answer to everything to me.
I have the better 3D performance ever with my 915GM.
Thanks a lot gborzi for this tip!

I also installed the intel driver 2.7 from testing, and i have no problem with it.

edit: Oh yeah, KMS + UXA is fine, excepts that I forgot that waking up from sleep didn't work with my chip.

Last edited by vinvin (2009-04-20 12:41:50)

gborzi · 2009-04-20 15:31:09

Today I updated the kernel to 2.6.29-4, nothing seems to have changed for the intel driver on my system.

eldragon · 2009-04-20 15:54:36

gborzi wrote:

Today I updated the kernel to 2.6.29-4, nothing seems to have changed for the intel driver on my system.

i think that update has nothing to do with graphics....

Damnshock · 2009-04-20 16:19:22

eldragon wrote:

gborzi wrote:
Today I updated the kernel to 2.6.29-4, nothing seems to have changed for the intel driver on my system.
i think that update has nothing to do with graphics....

Sometimes it might have something to do. Keep in mind that there have been some graphics related staff merged into ther kernel (GEM, if I'm not wrong is an example as well as KMS).

eldragon · 2009-04-20 16:46:15

Damnshock wrote:

eldragon wrote:
gborzi wrote:
Today I updated the kernel to 2.6.29-4, nothing seems to have changed for the intel driver on my system.
i think that update has nothing to do with graphics....
Sometimes it might have something to do. Keep in mind that there have been some graphics related staff merged into ther kernel (GEM, if I'm not wrong is an example as well as KMS).

from 2.6.29.1-3 to 2.6.29.1-4: patches applied had nothing to do with graphics, thats what i meant.

in fact, straight from the horse's mouth: thats what that update was about.

> > http://bugs.archlinux.org/task/14263
> > http://bugs.archlinux.org/task/12992

Dr4go · 2009-04-22 08:48:08

Dr4go wrote:

[...]
The combination of v2.6.30-rc3 and v2.7 should _finally_ work awesome.

Just to correct myself... I thought the drm kernel fix would make it into the rc3 of the kernel version 2.6.30, which indeed was a mistake, because it didn't happen. I guess we have to wait until rc4 then (or merging it on our own).

bangkok_manouel · 2009-04-22 12:38:24

Dr4go wrote:

Dr4go wrote:
[...]
The combination of v2.6.30-rc3 and v2.7 should _finally_ work awesome.
Just to correct myself... I thought the drm kernel fix would make it into the rc3 of the kernel version 2.6.30, which indeed was a mistake, because it didn't happen. I guess we have to wait until rc4 then (or merging it on our own).

FTR, the patch is available here

Arch Linux

#101 2009-04-17 14:31:49

Re: Intel 945GM, Xorg, Kernel - performance

#102 2009-04-17 17:36:58

Re: Intel 945GM, Xorg, Kernel - performance

#103 2009-04-17 18:04:47

Re: Intel 945GM, Xorg, Kernel - performance

#104 2009-04-17 19:07:21

Re: Intel 945GM, Xorg, Kernel - performance

#105 2009-04-17 19:12:22

Re: Intel 945GM, Xorg, Kernel - performance

#106 2009-04-17 20:31:04

Re: Intel 945GM, Xorg, Kernel - performance

#107 2009-04-17 20:49:48

Re: Intel 945GM, Xorg, Kernel - performance

#108 2009-04-18 13:31:03

Re: Intel 945GM, Xorg, Kernel - performance

#109 2009-04-18 14:16:21

Re: Intel 945GM, Xorg, Kernel - performance

#110 2009-04-18 14:33:21

Re: Intel 945GM, Xorg, Kernel - performance

#111 2009-04-18 14:58:30

Re: Intel 945GM, Xorg, Kernel - performance

#112 2009-04-19 02:05:16

Re: Intel 945GM, Xorg, Kernel - performance

#113 2009-04-19 06:57:22

Re: Intel 945GM, Xorg, Kernel - performance

#114 2009-04-19 09:46:20

Re: Intel 945GM, Xorg, Kernel - performance

#115 2009-04-19 12:31:33

Re: Intel 945GM, Xorg, Kernel - performance

#116 2009-04-19 13:50:09

Re: Intel 945GM, Xorg, Kernel - performance

#117 2009-04-19 15:20:47

Re: Intel 945GM, Xorg, Kernel - performance

#118 2009-04-19 17:17:19

Re: Intel 945GM, Xorg, Kernel - performance

#119 2009-04-20 10:24:45

Re: Intel 945GM, Xorg, Kernel - performance

#120 2009-04-20 15:31:09

Re: Intel 945GM, Xorg, Kernel - performance

#121 2009-04-20 15:54:36

Re: Intel 945GM, Xorg, Kernel - performance

#122 2009-04-20 16:19:22

Re: Intel 945GM, Xorg, Kernel - performance

#123 2009-04-20 16:46:15

Re: Intel 945GM, Xorg, Kernel - performance

#124 2009-04-22 08:48:08

Re: Intel 945GM, Xorg, Kernel - performance

#125 2009-04-22 12:38:24

Re: Intel 945GM, Xorg, Kernel - performance

Board footer