AMD Catalyst 8.5 For Linux

[Vargas]

Did anyone try with AGP cards ? like HD2400/2600 AGP

Yes, finally my HD2400 agp is working, after so many months.
Strange thing is that at login screen the resolution is wrong, after that it's correct. I'm on Ubuntu 8.04.

[Vighy]

Has anyone detected if using the authatieventsd.sh from the 8.5 package locks X on exit?

Maybe this issue only affects Ubuntus, dunno.

Here's diff with the patched file:
--- /etc/ati/authatieventsd.sh 2008-04-11 22:20:43.000000000 +0300
+++ /etc/ati/authatieventsd.sh.dpkg-new 2008-05-22 21:55:19.000000000 +0300
@@ -34,9 +34,7 @@

# Check xdm/kdm

-# XDM_AUTH_MASK=/var/lib/xdm/authdir/authfiles/A$1*
- XDM_AUTH_MASK=/var/run/xauth/A$1*
-
+ XDM_AUTH_MASK=/var/lib/xdm/authdir/authfiles/A$1*
XDM_AUTH_FILE=`ls -t $XDM_AUTH_MASK 2>/dev/null | head -n 1` # Choose the newest file
if [ -n "$XDM_AUTH_FILE" ]; then
SERVER_AUTH_FILE=$XDM_AUTH_FILE

I do get locks at logout! i experienced them with both gdm and slim.

I'm going to try this.



then i'd like bridgman to aswer me why:

- my catalyst control center still displays version: 1.0 since a lot of releases.

- most important: why it detects just 128MB of video ram (while i've got 512MB, checked it also under windows).

- composite performance (not AIGLX) is so poor (my 9200se was a lot faster).

- and if the change to the authatieventsd.sh does have effect, why killing Xorg it hangs! :-)

i'm running a Mobility HD2600 (M76) with 512MB from laptop "Asus Pro31sa" on a 64bit system.

thak's! :-)

[bridgman]

I have mobility radeon x1600 on 32-bit hardy, using gnome.
Using 8.5 with compiz works beautifully, and indeed noticeably faster than 8.3. However, switching to metacity results quickly in an entire garbled screen, full of artifacts. Could this be the fault of my configuration? The relevant section in my xorg.conf is:

Section "Device"
Identifier "Generic Video Card"
Boardname "ati"
Busid "PCI:1:0:0"
Driver "fglrx"
Screen 0
Option "XAANoOffscreenPixmaps" "on"
Option "VideoOverlay" "off"
Option "OpenGLOverlay" "off"
Option "TexturedVideo" "on"
Option "Textured2D" "on"
Option "TexturedXrender" "on"
Option "UseFastTLS" "1"
EndSection

Sounds like you should use Compiz then

Seriously, first thing I would try is turning off TexturedXrender and then maybe Textured2D. I'm not 100% sure, but I believe both of those are work-in-process today and I expect Metacity would lean on TexturedXrender pretty hard.

- my catalyst control center still displays version: 1.0 since a lot of releases.

I believe that refers to the framework version, not the detailed content of tabs & controls. Each of the components also has a (much larger) version number which gets bumped on every change but I don't think we display those externally.

- most important: why it detects just 128MB of video ram (while i've got 512MB, checked it also under windows).

What I've been told is that this is the amount of memory which is mapped into the CPU address space and can be drawn into directly by the CPU; the rest is only accessible to the acceleration hardware. Not 100% sure but this seems to make sense; typically the largest value you see mapped is 256K. Since all of the stuff you see on the screen is either drawn by the acceleration hardware or blitted to the screen by the acceleration hardware it probably doesn't matter what this value is set to, as long as it is big enough to hold the actual front buffer displayed on the screen.

- composite performance (not AIGLX) is so poor (my 9200se was a lot faster).

What exactly do you mean by composite performance here, and why do you say "not AIGLX" ? Under Compiz (are you using Compiz ?) I believe all of the compositing is done via AIGLX. If you are using Metacity and the TexturedXrender paths, those are still work-in-progress -- there seemed to be a huge preference for prioritizing AIGLX.

- and if the change to the authatieventsd.sh does have effect, why killing Xorg it hangs! :-)

Honestly, there seems to be a huge list of things which cause driver-to-driver handoffs to hang and nowhere near enough standards or infrastructure to make sure it succeeds. This seems to be the most system-specific and kernel-specific aspect of the whole X driver space and success seems to be based in part on making the same design assumptions as all of the OTHER driver bits which bang on the same hardware. The issue with atieventsd.sh is only one of maybe a hundred different things that can cause hangs when switching between drivers. One day kernel modesetting will fix most of these problems; in the meantime there's just a lot of testing and tweaking to keep all the different bits of a Linux system roughly in sync.

i'm running a Mobility HD2600 (M76) with 512MB from laptop "Asus Pro31sa" on a 64bit system.

[djdoo]

@Bridgman:

Seriously, first thing I would try is turning off TexturedXrender

Hey John you just made my day man!!
I set Texturedxrender to "off" and now screen is not getting grambled with the 2D Kwin manager at KDE 4!!

So the solution guys is just set "off" TexturedXRender, Textured2D is OK to remain "on"!

I will update my guide at Compiz-Fusion forum now!

Always do the amdpcsdb trick after the change..., aticonfig --input=/etc/X11/xorg.conf --tls=1

[smlbstcbr]

I know that this issue has been reported before but I think that this deserves another comment. I am running a Gentoo amd64 2007.1 system (to say so, but updated the system a couple of weeks ago), I have a Radeon X300, 1 GB RAM, an AMD Athlon 64 3000+, Gnome and Catalyst 8.5. I used to navigate with Google Earth and everything was absolutely fine as well as Open Office 2.4 and GDM 2.20.3. After I installed Catalyst 8.5 GDM appeared corrupted, which I found later to be an issue with Catalyst not detecting the monitor refresh rate properly at 1920*1200 or so, at least to my knowledge (which with 8.4 worked flawlessly) and the context buttons had some annoying artifacts (that look like shadowed text). The same applies to Open Office, which suffers the same problem with the buttons, the drop-down menus, the context messages and text (oddly, it does not affect the text one writes). Google Earth locks the system, seems to change screen resolution and the monitor turns off and the system stops responding, although the same happened with 8.4 (I noted that after reinstalling 8.4 to see what happened). Besides those things, video is running very well, and a general impression shows a solid work.

[djdoo]

Furthermore now with TexturedXrender "off" I take 175fps at RACER whereas before I scored 40fps and all the artifacts at 2D desktop...

I definately believe that it is broken for 8.5 driver...

[djdoo]

Also Compiz-Fusion is even better now! Good work guys!

Any idea when Crossfire will be ready??

[djdoo]

@bridgman:
Hey John

I would have sent this to you in private but I couldn't so here's my message...

I would like to invite you to Compiz-Fusion forum and also I would like you to take a look at our guide there for a good ATI setup and help us a bit with your 'inside' knowledge and experience man...

Your help would be more than welcome not to say needed...

The forum thread locates here http://forum.compiz-fusion.org/showthread.php?t=6794

Truly Yours

Jim

[Maxhysteria]

Maybe Bridgman can answer this question :
What are the default settings used if the section device in xorg.conf is empty ? (except for Identifier and Driver entries )

[Vighy]

I always like your answers because are clear and precise.

What exactly do you mean by composite performance here, and why do you say "not AIGLX" ? Under Compiz (are you using Compiz ?) I believe all of the compositing is done via AIGLX. If you are using Metacity and the TexturedXrender paths, those are still work-in-progress -- there seemed to be a huge preference for prioritizing AIGLX.

I mean without compiz... using a compositor like the one integrated in xfce4. Changing workspace it takes a lot. While without compositing it's instantaneous. I can live without it, i just wanted to know if it was a know issue :-)

Honestly, there seems to be a huge list of things which cause driver-to-driver handoffs to hang and nowhere near enough standards or infrastructure to make sure it succeeds. This seems to be the most system-specific and kernel-specific aspect of the whole X driver space and success seems to be based in part on making the same design assumptions as all of the OTHER driver bits which bang on the same hardware. The issue with atieventsd.sh is only one of maybe a hundred different things that can cause hangs when switching between drivers. One day kernel modesetting will fix most of these problems; in the meantime there's just a lot of testing and tweaking to keep all the different bits of a Linux system roughly in sync.

Ok i solved the hang on logout. It was because I changed to slim lately (from gdm) and in the authatieventsd.sh there's no rule for slim.

I created it, how can i contribute? writing on the unofficial bugzilla could be a way?
If i propose the patch will be added on the next version (once fixed every possible problem.) ?

thank's again Bridgman! :-D