ATI radeon: poor 3D performance

[Mika Fischer]

I have an IBM Thinkpad R40 2722-CDG with a Radeon Mobility 7500 graphics chip.
See https://wiki.ubuntu.com/LaptopTestingTeam/ThinkpadR40-2722 for details.

2D performance is very sluggish overall (moving windows, resizing windows, etc.). Under Windows XP I have no such problems.
The worst applications which show this effect extremely well are firefox and thunderbird where I can sometimes even watch it redraw the widgets one after the other...

I'd be interested to know if there is a kind of benchmark for measuring 2D drawing performance of X...

But the 3D perfomance is even worse than that. With flightgear I get only 3-4 frames per second, whereas on Windows i have about 50.
And those 3-4 I get only if I add the following to my xorg.conf:
Option "AGPMode" "4"
Option "AGPFastWrite" "on"
Option "EnablePageFlip" "true"
Without those options I get only about 1 frame per second...

Again I'd be interested to know if there's a real benchmark for 3D performance.

I'll try to test this on Dapper, and Breezy Live CDs to see if it's a regression...

[Jeremy Vies (jeremy.vies)]

Hi Mika,
do you know what driver you are using ? vesa or radeon ?

I do have a radeon mobility U1 (~ radeon 7000), and 2D performances are acceptable. I don't have any trouble when I move windows...

[Mika Fischer (zoop)]

Hi Jeremy,

I'm using the radeon driver of course. :)

I've attached my xorg.conf for reference. It's the default with the addition of the Options mentioned above.

Perhaps one could call the 2D performance "acceptable", it certainly is usable. But when I see what this card can do under Windows then the results on Linux are certainly disappointing, to say the least.

On my girlfriend's computer which has a Geforce2 MX in it (using the closed-source nVidia drivers), half the RAM and a slower processor, working with Ubuntu is a lot more fun than on my laptop...

[Mika Fischer (zoop)]

Ah, and I removed the two or three input devices relating to graphics tablets, which I don't have and which generate spurious warnings which bugged me...

[Jeremy Vies (jeremy.vies)]

Have you looked at logs from Xorg server ? maybe some troubles are reported in it.

(logs are in /var/log/Xorg.0.log)

[Mika Fischer (zoop)]

Yeah, I did that too. Sorry for not attaching them in the first place...

There are some warnings. I don't quite understand them but they seem to relate to DRI, so may be connected to the bad 3D performance...

[Jeremy Vies (jeremy.vies)]

I see a [resume] line by the end of the file. Does the problem happen after a suspend / resume ?

Else, the log seems clean to me.

[Mika Fischer (zoop)]

No. The performance is the same befor suspending.

[Mika Fischer (zoop)]

Some recent update fixed the 2D performance. It's still not as good as under Windows but now it's acceptable.

The 3D problem remains, though...

It might have to do with the following error message I get when starting GL apps:
libGL warning: 3D driver claims to not support visual 0x4b

Searching for this leads to an X.org bug report:
https://bugs.freedesktop.org/show_bug.cgi?id=6624

There it is hinted that this might be a general DRI/Mesa bug since Intel chips are also affected.

I'm setting this bug to "Confirmed" because several people on that bug have the same problem.

[Mika Fischer (zoop)]

I've reported my problem upstream.

[sam tygier (samtygier)]

this and bug 58541 both link to the same upstream bug.

[Mika Fischer (zoop)]

I don't have this laptop anymore. So I can't help in any way. Feel free to close this bug.

[Nicolò Chieffo (yelo3)]

Hello. I'm having the same problems with the radeon driver. I've discovered that if you set the ColorDepth to 16 the performance is smoother, at least using compiz. I don't know about 3d games.

[Nicolò Chieffo (yelo3)]

One other strange thing is that if viewing a firefox page with animated gifs or flash animations, (compiz enabled) the cpu is working at 10%. Of course this does not happen when compiz is disabled. I see this debug message when starting compiz:

compiz.real: GLX_EXT_texture_from_pixmap is not supported by direct rendering context, trying indirect rendering context instead

[Tormod Volden (tormodvolden)]

(Intel was only mentioned in an earlier upstream bug about the harmless libGL warning).

Nicolo, what kind of card do you have? Please attach Xorg.0.log and xorg.conf

[Nicolò Chieffo (yelo3)]

I have a mobility radeon 9700.
I can't give what you asked now, because I'm not at home. But I can
say that I've tried these options and the performance is the same:
EXA, PageFlip, AGPMode, RenderAccel (which is not supported by my
model, as I can see in Xorg.0.log).

In each case the CPU is highly used (I think instead of the GPU) and
causes the system to become slow (when using a composite manager) or
the game to be choppy (even planet penguin racer is not fluid)

[Nicolò Chieffo (yelo3)]

even glxgears uses lots of CPU. I've discovered that with driconf I
can set an option named "sleep for brief intervals" instead of using
the interrupts. This option causes no cpu usage in glxgears, but
others applications are still slow

[Tormod Volden (tormodvolden)]

Can you run:
LIBGL_DEBUG=1 glxinfo >gldebug.txt 2>&1
and attach the resulting gldebug.txt file please?

[Nicolò Chieffo (yelo3)]

xorg 7.2 has much better performance! I've tested it from a repository found in the forums

[Erik Andrén (erik-andren)]

Nicolo, xorg 7.2 has been merged to feisty current. Is it still working fine without the extra repository?

[Timo Aaltonen (tjaalton)]

Mika: can you try Feisty Herd5 livecd to test if the performance is better for you?

[Nicolò Chieffo (yelo3)]

compiz performance is higher than xorg 7.1 but still not metacity comparable. some options are still quite slow... such as scale, maximize, restore a minimized window, scroll.

Anything I can do? some other debug... don't know...

[Johannes H. Jensen (joh)]

I'm experiencing similar issues on my powerbook5,6 with an ATI Technologies Inc RV350 [Mobility Radeon 9600 M10] card which is supported by the r300 driver.

Even though glxinfo reports that direct rendering is working, I only get around 35fps from glxgears and ~0.5fps in games like ppracer and neverball.

I've heard numerous reports that the R300 driver now is stable and provides good 3d acceleration for these cards. Johannes Berg even reports to have 3d acceleration working on his identical PowerBook (http://johannes.sipsolutions.com/PowerBook). So my question is this: is the (latest?) r300 driver even included in Edgy? My Xorg.0.log seems to suggest otherwise:

(WW) RADEON(0): Enabling DRM support

 *** Direct rendering support is highly experimental for Radeon 9500
 *** and newer cards. The 3d mesa driver is not provided in this tree.
 *** A very experimental (and incomplete) version is available from Mesa CVS.
 *** Additional information can be found on http://r300.sourceforge.net
 *** This message has been last modified on 2005-08-07.

I would guess that this message is from some old version of the r300 driver, but I'm not sure...

[Johannes H. Jensen (joh)]

Tested this on Feisty now and it seems the problem is gone. I have great 3d acceleration now! :)

[Timo Aaltonen (tjaalton)]

Johannes: yes, that's a different card, so this bug still remains open until we get feedback from Mika.

[Johannes H. Jensen (joh)]

But isn't it the same driver? The symptoms were identical to mine, though as you say the problem might be card-specific.

Note that my comment was by no means a suggestion to close this bug, merely an update from my side.

[Nicolò Chieffo (yelo3)]

I have a sort of regression in glxgears. the fps count is aroud 1000
and uset do be a bit less than 2000... I'm using the "ati" driver
without any additional option, and I have a mobility 9700, in feisty.
And again glxgears is using more than 60% of my cpu. What about your
cpu usage Johannes?

[Johannes H. Jensen (joh)]

As far as I could see from the short time I got to test the 3D performance on Feisty (gnome packages broke yesterday so now I'm back on os x) the cpu usage seemed normal while running glxgears and ppracer though I'm not 100% sure. Maybe you could try the Feisty LiveCD on your computer? I'm unable to on mine as there is no LiveCD available for PowerPC yet...

[Nicolò Chieffo (yelo3)]

I've got feisty installed and those are my results: lots of cpu used in 3d apps.

[sam tygier (samtygier)]

there are feisty live cds (herd and daily) they have move to the ports section of the server http://cdimage.ubuntu.com/ports/

[Nicolò Chieffo (yelo3)]

I can confirm again poor performance. I've installed google earth and
also without doing anything the cpu is at 70%. with fglrx is no used
so much instead

[Roy Verrips (rverrips)]

I have an IBM Thinkpad Z60m with ATI Mobility Radeon X600 128MB and with Feisty Herd 5 the "Desktop Enhancements" performance using the default drivers is very poor ... windows drag slowly, take forever to open etc.

When changing to the Ubuntu standard "Restricted Drivers" Composite is disabled and "Desktop Enhancements" thus not available.

Have also tried the "official" ATI drivers from, and although 3D performance is create, failure on their side to have Composite Enabled means no "Desktop Enhancements"

[Nicolò Chieffo (yelo3)]

I'm running feisty xorg 7.2-0ubuntu11 xserver-xorg-video-ati 6.6.3-2ubuntu4
xorg.conf default options (reconfigured using -phigh)
what does this output mean (the output of ppracer)?
what is GART size? and what value should it have? now I have 8MB (aperture) and 5MB (textures) by default

*********************************WARN_ONCE*********************************
File r300_vertexprog.c function valid_dst line 333
Output 3 not used by fragment program
***************************************************************************
*********************************WARN_ONCE*********************************
File radeon_mm.c function radeon_mm_alloc line 216
Ran out of GART memory (for 1048576)!
Please consider adjusting GARTSize option.
***************************************************************************

[Tormod Volden (tormodvolden)]

What does the performance look like in Tribe 1? I understand the original reporter can't check any longer, but the rest of you?

[Nicolò Chieffo (yelo3)]

In my case the performance is very bad.
I have a Mobility Radeon 9700
I cannot run any 3d games without an excessive use of the cpu. And the
performance is about 5fps
even with the new gnash (plugin for flash) when rendering a simple
animation all the cpu is used.

[Tormod Volden (tormodvolden)]

Nicolo, can you quantify the regression with for instance our favourite benchmark glxgears?

[Nicolò Chieffo (yelo3)]

this is the result of glxgears running in a small window
1297.375
this is instead the result with the window maximize
140.581

again glxgears puts the cpu to 100% usage
I cannot play games such as lincity, smc because the fps even in the menu is ~1

[Tormod Volden (tormodvolden)]

Nicolo, have you got any better numbers before, with another release for instance, or has it always been that bad?

[Nicolò Chieffo (yelo3)]

I've always had a very very bad performance

[Tormod Volden (tormodvolden)]

There are some debugging tips on http://dri.freedesktop.org/wiki/DriTroubleshooting but most of it should not be necessary on a clean Ubuntu install.

[Tormod Volden (tormodvolden)]

Some people having issues with googleearth have used the program driconf and selected "Disable Low-Impact Fallback". Maybe worth a try.

[Nicolò Chieffo (yelo3)]

this is my result following the guide:

some x.0.log lines

drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is -1, (No such device or address)
drmOpenDevice: open result is -1, (No such device or address)
drmOpenDevice: Open failed
drmOpenByBusid: Searching for BusID pci:0000:01:00.0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmOpenByBusid: drmOpenMinor returns 7
drmOpenByBusid: drmGetBusid reports pci:0000:01:00.0
---------
(WW) RADEON(0): DRI init changed memory map, adjusting ...
(WW) RADEON(0): MC_FB_LOCATION was: 0xd3ffd000 is: 0xd3ffd000
(WW) RADEON(0): MC_AGP_LOCATION was: 0xffffffc0 is: 0xe07fe000
---------
(II) RADEON(0): Using 8 MB GART aperture
(II) RADEON(0): Using 5 MB for GART textures
(isn't 8MB a bit low?)
---------
glxinfo reports
libGL warning: 3D driver claims to not support visual 0x4b
OpenGL renderer string: Mesa DRI R300 20060815 AGP 1x x86/MMX/SSE2 TCL
(why agp 1x? I can handle 4x!)
--------

I've tried the disable low impact fallback some time ago. It improbed
google earch performance to quite acceptable (though the cpu was at
100%, but the images were fluid) but it caused instability and system
hangup.
I will try this option soon again

[Nicolò Chieffo (yelo3)]

I've just tried to play smc (super mario chronicles) and I got this error:

*********************************WARN_ONCE*********************************
File r300_render.c function r300Fallback line 441
Software fallback:ctx->Line.SmoothFlag
***************************************************************************
Try R300_SPAN_DISABLE_LOCKING env var if this hangs.
Last known Error : Failed loading DPMSDisable: /usr/lib/libX11.so.6:
undefined symbol: DPMSDisable

the performance is near to null

[Nicolò Chieffo (yelo3)]

I've done some tests with nexuiz (a full 3d game, quite intensive for
my video card)

without the option disable low impact fallback it is not playable,
after enabling it I got nearly 60fps, sometimes with 1 second hangs,
but the situation is not so bad...
I think the key for 3d games is this option! I think that I should see
the fps when using fglrx, but now it can't be used due to a X version
mismatch

After launching google earth with the option enabled, the system
hangs, as I reported before (a hard reboot is needed)

compiz does not take any advantage from this option :( it is sad
because compiz performance is quite bad!
scrolling is very slow (though improved) the animations are quite
huggy, and everything is highly dependent on the cpu usage of the
moment

also take a look ad this Xorg.0.log line
(II) RADEON(0): Render acceleration unsupported on Radeon 9500/9700 and newer.
(II) RADEON(0): Render acceleration disabled

Tormod, what do you think?

[Tormod Volden (tormodvolden)]

Nicolo, please file a new upstream bug, the old one had too many different issues mixed in, and then the developers get confused. Attach your log files and other results that you have obtained. From what I know you might have two separate issues, one for 3D and one more general 2D (that might have been fixed in git), and in that case they should be treated separately. I think "Render acceleration" is for 2D only.

[Nicolò Chieffo (yelo3)]

xserver-xorg-video-ati version 6.6.3-2ubuntu6

This bug is about playing 3d games/apps with a Mobility Radeon 9700.
I've done tests with nexuiz (a complex 3d game like quake), smc (a simple super mario clone), google earth and compiz (git version 0.5.1)

- nexuiz: the game runs kindly fluid (though the cpu is used at 95%), with 60fps. l see this message in the console
*********************************WARN_ONCE*********************************
File radeon_mm.c function radeon_mm_alloc line 216
Ran out of GART memory (for 1048576)!
Please consider adjusting GARTSize option.
***************************************************************************
In fact in xorg.0.log I see this:
(II) RADEON(0): Using 8 MB GART aperture
(II) RADEON(0): Using 5 MB for GART textures

I tried to manually set gart to 128MB with the option GARTSize, but the situation didn't improve (except that the debug message is gone)
The result with fglrx is worse: good job!

- smc and google earth: The performance is near to 1 fps and the cpu goes immediately to 100% of use (centrino 1.6ghz), and I see these messages:
*********************************WARN_ONCE*********************************
File r300_render.c function r300Fallback line 441
Software fallback:ctx->Line.SmoothFlag
***************************************************************************
Try R300_SPAN_DISABLE_LOCKING env var if this hangs.

if I use driconf to enable the option "Disable low impact fallback" smc runs smooth (though with 70% cpu usage) but google earth hangs the whole system.
the result with fglrx is: smc has no cpu usage, google-earth has low cpu usage

- compiz: using compiz git version 0.5.1+git20070621-0ubuntu1 I see some problems in scrolling, animations and when changing to a virtual desktop that has some open windows. Scrolling is really slow and not acceptable, animations have some hungs that don't let them be fluid, and when I change desktop there is a 2 seconds laf before the windows are drawed
Of course I cannot test the performance with fglrx

[Nicolò Chieffo (yelo3)]

Created an attachment (id=10418)
xorg.conf

[Nicolò Chieffo (yelo3)]

Created an attachment (id=10419)
Xorg.0.log

revelant parts:
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is -1, (No such device or address)
drmOpenDevice: open result is -1, (No such device or address)
drmOpenDevice: Open failed
drmOpenByBusid: Searching for BusID pci:0000:01:00.0
drmOpenDevice: node name is /dev/dri/card0
drmOpenDevice: open result is 7, (OK)
drmOpenByBusid: drmOpenMinor returns 7
drmOpenByBusid: drmGetBusid reports pci:0000:01:00.0
---------
(WW) RADEON(0): DRI init changed memory map, adjusting ...
(WW) RADEON(0): MC_FB_LOCATION was: 0xd3ffd000 is: 0xd3ffd000
(WW) RADEON(0): MC_AGP_LOCATION was: 0xffffffc0 is: 0xe07fe000
---------
(II) RADEON(0): Using 8 MB GART aperture
(II) RADEON(0): Using 5 MB for GART textures
(isn't 8MB a bit low?)
---------
(II) RADEON(0): Direct rendering enabled
(II) RADEON(0): Render acceleration unsupported on Radeon 9500/9700 and newer.
(II) RADEON(0): Render acceleration disabled
---------
glxinfo reports
libGL warning: 3D driver claims to not support visual 0x4b
OpenGL renderer string: Mesa DRI R300 20060815 AGP 1x x86/MMX/SSE2 TCL
(why agp 1x? I can handle 4x!)

[Bugzi09-fdo-tormod (bugzi09-fdo-tormod)]

Can you please try 6.6.192 also? (Ubuntu packages on https://wiki.ubuntu.com/XorgOnTheEdge)

[Nicolò Chieffo (yelo3)]

I've tried them but the performance is not improved at all, and the error messages about r300Fallback are still here. I haven't tried google earth because now I don't want to crash the system again

[Sroland-vmware (sroland-vmware)]

New driver should use higher agp mode by default, you can enable it with "Option" "AGPMode" "4" manually, but it's unlikely to make a lot of performance difference. When you see those fallback messages for things not implemented, software rasterizer is used so it's expected that performance is very low and cpu usage at 100%. Not sure why cpu usage is quite high even when hw rendering is used, maybe it's just state overhead or vertex submitting or something not very well optimized - you could use oprofile to get a rough idea.
compiz should run better with current ddx and dri drivers.
The google earth gpu lockup (maybe - you didn't say if the box is truly dead or it just appears to be dead) could (or not, I don't know but it would be worth a try) be fixed by a newer dri (mesa) driver.

[Nicolò Chieffo (yelo3)]

You're right for the AGP part, I didn't notice it! I hoped that also GARTSize could be right detected!
Will the not implemented parts developed sooner or later, or are they discontinued now?
I will try oprofile, though I'm not sure how to use it...
When you say that compiz should run better with new drivers you mean GIT version or the one I just installed? I hope the GIT version, since the one I am using does not improve anything...
google earth with the option "disable low impact fallback" hangs the whole system: I even tried sysrq keys but nothing happened!

[Nicolò Chieffo (yelo3)]

I've tried google earth with ati 6.6.192. The situation is a bit different:
I can move the mouse, but I cannot do anything. again keyboart shorcuts to recover the system are not working.

[Michel-tungstengraphics (michel-tungstengraphics)]

(In reply to comment #7)
> I've tried google earth with ati 6.6.192. The situation is a bit different:
> I can move the mouse, but I cannot do anything. again keyboart shorcuts to
> recover the system are not working.

As Roland mentioned, the version of Mesa generally matters more for 3D apps than the version of xf86-video-ati. Have you tried Mesa 7.0 or GIT?

To get better compiz performance, you need to build and run the xserver GIT master branch (which also requires the Mesa GIT master branch) and the xf86-video-ati GIT master branch built against that with EXA.

Also, please don't mix in several issues into a single bug report.

[Nicolò Chieffo (yelo3)]

I'm not using git.

I will file new bugs for compiz and google earth, sorry.

Now this bug is about poor 3d performance and hi cpu usage for the missing implementations that fall back to software render.

Is there some developer still interested to implement them?

[Michel-tungstengraphics (michel-tungstengraphics)]

(In reply to comment #9)
> I will file new bugs for compiz and google earth, sorry.

Don't bother for compiz, or I'll just resolve it as fixed due to the improvements in the current GIT versions. Also please only file a new bug for Google Earth after trying Mesa GIT or at least the 7.0 release.

[Glisse (glisse)]

(In reply to comment #9)
> Is there some developer still interested to implement them?

It will get implemented once any one got time to do it. If
you got time and motivation you welcome to jump in.

[Nicolò Chieffo (yelo3)]

You were right. I now installed mesa 7.0 and google earth will not crash with the option disable low impact fallback! Thank you for the suggestion

Sorry I don't have the knowledge to help you writing missing functions, sorry!

In the end, this bug will be only to remember that there are missing functions in the r300 code.

And I will open a new bug about hi cpu usage in some 3d apps (such as glxgears). Can I open it now, or will I have to install a newer version of xorg or ati before?

[Ivan Ivanoff (spammeroff)]

Hello! I have got HP Compaq nc4000. And I can't get direct rendering support, despite it is written in xorg.log, where I can see clearly "Direct Rendering enabled". I have ATI Technologies Inc Radeon IGP 330M/340M/350M as lspci said. I have very poor performance on 3D - 300 fps in glxgears. In xorg.conf I wrote "radeon", but it does not help at all. I see clear, without any errors log, but in real I cannot get 3D-direct rendering.

[Johannes H. Jensen (joh)]

Hello,

I'm sorry but I don't know what your problem might be. Have you tried
posting it at the ubuntu forums?

Best regards,

Johannes H. Jensen
<email address hidden>

On Aug 5, 2007, at 8:45 :34, Ivan Ivanoff wrote:

> Hello! I have got HP Compaq nc4000. And I can't get direct rendering
> support, despite it is written in xorg.log, where I can see clearly
> "Direct Rendering enabled". I have ATI Technologies Inc Radeon IGP
> 330M/340M/350M as lspci said. I have very poor performance on 3D - 300
> fps in glxgears. In xorg.conf I wrote "radeon", but it does not
> help at
> all. I see clear, without any errors log, but in real I cannot get 3D-
> direct rendering.
>
> --
> ATI radeon: poor 3D performance
> https://bugs.launchpad.net/bugs/56692
> You received this bug notification because you are a direct subscriber
> of the bug.

[Johannes H. Jensen (joh)]

Whups, I mistakenly interpreted the mail from this list as a mail directly to me requesting support. Please ignore / delete my previous message...

[Nicolò Chieffo (yelo3)]

Ok I'll mark this as fixed. The 3D problems are gone away if the DRI option "DisableLowImpactFallback" is enabled