Which Linux Graphics Driver Bugs Do You Hate?

[phoronix]

Phoronix: Which Linux Graphics Driver Bugs Do You Hate?

This week we received a note from Matthias Dahl, a Phoronix reader, who wanted to remind us about current problems plaguing the NVIDIA 180.xx driver series. Using any of the newer NVIDIA Linux drivers can cause graphics corruption followed by the system locking up...

http://www.phoronix.com/vr.php?view=NzEwMQ

[Vadi]

Wow, I consider myself lucky then - 8600gt here and it's all working out just fine.

No real driver bugs either, save for S2 getting a drop in fps sometimes after switching focus. But I can't complain - it's 1% of the linux games that has the basic capability to alt+tab in fullscreen.

[MùPùF]

I noticed this bug too on Ubuntu. But when I switched to ArchLinux, the problem didn't occur any-more. Maybe it has to do with Ubuntu.

By the way, I've just registered, this is my first message. Long live to Phoronix !

[ethana2]

Oh, X freezes up every once in a while and I have to Ctrl+Alt+Backspace..
Aside from that it's hard to know which packages bugs I encounter are actually in; I don't know how polished compiz is right now.. I'm just looking forward to nouveau maturing so when a problem does show up, I know what the heck is going on.

Third party proprietary drivers are a huge part of the reason Windows sucks so hard, but they'll affect any system that depends on them, really, and it makes me sad to see that be the case..

[deanjo]

7300GS, 8800GT's, 9600GT, and 8600GT no issues here on openSUSE 11.1. At work, 8200/8300 chipsets and some 7600GT's, running combo of Bluewhite and opensuse 11.1/11.0 and not one report of lock up there (50+ stations all using nvidia chipset boards).

[Remco]

The absence of VTs.
The absence of kernel panic notifications.
The absence of monitor hot-plugging.

Proprietary NVIDIA driver, obviously.

[Ant P.]

I haven't had many bugs to complain about, more just lack of features. But then... I stopped using nVidia a long, long time ago.

[BassKnight]

I haven't had problems with the latest drivers in my 8500GT, the annoying 2D acceleration issues were resolved starting with the 180.xx drivers, so I'm happy overall.

I'm using Xubuntu 8.10, but with a custom-compiled 2.6.28 kernel.

[beecher]

The bug that I hate the most is tearing. I see it on videos (with or without Compiz enabled) with my intel 4500MHD. It makes watching video useless on Ubuntu for the moment.
The worst part is I also have an ATI 3650HD on my laptop, which show the exact same problem with the fglrx driver.

I don't know maybe I should have bought a laptop with an Nvidia card :rolleyes:, well maybe in a few months this problem will be history (either with the intel driver or maybe the radeonhd driver).

[sreyan]

with NVIDIA drivers i've had stability issues with my 8600GTS but not many of my other cards. with fglrx i've had tearing problems and unusable video for a year on my 2600xt. Till recently the 2600xt had unusable video with radeonhd since it didnt have xvideo support for r600 chips.

sadly the graphics cards that work best are a number of GMA950 based machines (mac mini, acer aspire one) and one ati rs200 based laptop. One uses the open source intel drivers and the other has used the open source ati drivers.

[Saist]

I'm just going to go ahead and say it : Welcome to 2 years ago

Nvidia's driver issues are hardly new, and hardly contained to the 180 branch. Case in point is my own report comparing Nvidia's 100.14 64bit drivers to ATi's 8.39 64bit release : http://mepisguides.com/Mepis-7/hardware-reports/nvidia-ati-64bit/nv-ati-64bit.html

My biggest beef with Nvidia though isn't the instability and X-server crashes I've lived through. It's the fan control. I literally had to underclock two 6600 GT cards I had because the fans wouldn't shut up, even when the system was at idle. While I've never had to underclock my 7900 GT KO, it's got an obnoxious habit at running the fan at full tilt... ALL the time.

Of course, I could rant about the lack of an Open-Source Strategy from Nvidia, and while I hope that Red Hat's forcing of the issue with NV / Nouveau will result in positive change, I just don't see it happening.

***

Of course, that doesn't mean I'm looking at AMD/ATi with positive thoughts either. Crossfire support has arrived for the RadeonHD 4x00 series... but what about all the PREVIOUS graphics cards? Crossfire has been around since what.. the x800? Where is that support? Although that's not actually a bug. Although the X-server locking up on certain drag-and-drops on a pair of Crossfired 3870's could possibly be attributed to a bug.

Well, then there's the packaging scripts, with the Debian packaging scripts appearing to be broken. Well, all insult intended, last I heard a Ubuntu guy was handling the Debian scripts, and Ubuntu has no interest what-so-ever in compatibility, a lesson Mepis Linux learned the hardway after finding there was no upgrade path out of Dapper. However, people who seem to be qualified in handling the scripts on debian, such as the creator of the popular SMXI / SGFXI scripts, seem to have an anti-AMD hard-on or something. The list of people available who want to handle Debian packaging, and are qualified, seems to be extremely small. Again though, that's not really a bug.

Truth be told, AMD/ATi solved most of the big bugs I had in mind. One of the driver sets last year broke Cedega / W.I.N.E. support, an issue that has since been resolved... which is good for my City of Heroes addition.

[bridgman]

beecher, I believe both the radeon and radeonhd drivers have tear-free Xv support today. You'll need to pick up drm from either drm- next or the 6xx-7xx git branch of drm, but the X driver code has been merged to master on both radeon and radonhd.

[Nille_kungen]

I can't use ati catalyst driver after 8.12
I can't get opengl to work on slackware 12.2
And many slackware users have the same problem with 9.1 and 9.2
http://www.linuxquestions.org/questions/slackware-14/ati-9.2-still-didnt-work-708209/?highlight=ati
http://www.linuxquestions.org/questions/slackware-14/catalyst-9.1-and-pcimsi-703142/?highlight=ati

I think that not being able to use the driver at all is an greater bug then an minor graphics corruption.

[macemoneta]

I have a Supermicro C2SEA with the Intel G45/X4500HD, and graphics freezes are common too (about once a day), even without Compiz. Between that and the video tearing, I can't wait for the new graphics subsystem to stabilize.

[oblivious_maximus]

The 180.xx drivers were a lot faster with KDE4 than the 177s and earlier versions, but it was still a pretty slow experience trying to use KDE4 with my 8600GT. I'm sure KDE4(and/or its dependencies) is partly to blame, but Nvidia's blob surely is also. One eight oh - STILL TOO SLOW. Ultimately, the slight performance increase was not enough to offset the bugs in the 180.xx series - loved that flashing red checkerboard on top of every non-fullscreen video I played. Oh and broken suspend too, that was so awesome.

So I'm back to KDE3 (again, not 100% thanks to Nvidia - but still, in good part), 177.82, and the same old never-gonna-get-fixed, totally-absent-for-no-good-reason TV out overscan control. If that had been fixed in 180.xx I would have put up with the other bugs. It's definitely the bug I hate the most, only it's not a bug so much as it was a decision some fucker at Nvidia made. The new Windows release doesn't need that anymore - so fuck the users of every other OS in existence right?

Does fglrx support running multiple X servers yet? Totally despise that one too.

In summary then, what I hate most is having Windows paradigms forced upon me. I shouldn't be forced to choose between Big Desktop and whatever other mode they offer when X.org offers the ability to run more than one Xserver. The fact that Vista doesn't need overscan controls shouldn't mean that users of every other fucking OS (other Windows versions too!) get the shaft.

There's 10s of millions of Linux users! Isn't that enough to get some fucking respect already?? Sadly I know the answer, and it's no.

[JeanPaul145]

What bothers me most is the fglrx driver locking up the whole INtrepid amd64 system (not just X) when I watch a video with compiz enabled. Radeonhd is no solution (yet) either, since the support for my card (a Radeon HD4870) is limited to EXA and Xv. In any case, no useable 3D support yet.

To sum it up: if I forget to switch back to metacity before watching a video, I can reboot my pc.

*Starting to reminisce about the good ol' windows days*

[clarious]

From what I understand, the bug which cause corrupted/not updated graphic is due to a change in Nvidia driver.

Prio to 180.*, the driver used shared memory but not the card memory for pixmap, that cause serious performance problem with 2D graphic (you can test this by scrolling in Firefox). In 180.*, they fixed this by using a cache for Pixmap, but the problem is that when this cache get full, the windows may get corrupted when it was first painted. And this is only happen if compiz is in use, of course.

Aside from the problem with Compiz, nvidia driver work fine with my laptop 8600M card :)

[hax0r]

I don't have any problems with latest nvidia driver nor did with earlier versions, I use Arch Linux with latest vanilla kernel compiled, 8800GT - 9600GT - 260GTX, all run fine :confused:. 2D is a bit too slow, it's much better with nouevau. I have tried KDE 4.2 but it's choppy on both nvidia and nouveau drivers, other DE and games run fine.

[Saist]

I can't use ati catalyst driver after 8.12
I can't get opengl to work on slackware 12.2
And many slackware users have the same problem with 9.1 and 9.2
http://www.linuxquestions.org/questions/slackware-14/ati-9.2-still-didnt-work-708209/?highlight=ati
http://www.linuxquestions.org/questions/slackware-14/catalyst-9.1-and-pcimsi-703142/?highlight=ati

I think that not being able to use the driver at all is an greater bug then an minor graphics corruption.

I tend not to get hung-up on a graphics driver not installing.

Case in point for me is Mepis 7 and Mepis 8. The install script that I had been using up until Catalyst 9.1 suddenly started failing with Catalyst 9.2, and the 64bit install of Fglrx drivers continually failed during the entire Mepis 8 beta series.

However, I would be hard-pressed to refer to an installation problem as an error with the driver itself. Several different factors can affect whether or not a particular driver will run, such as the system architecture (i386 / i586 / i86 / x86-64 / ppc and so on), the kernel version, and the x-server version.

While I think it's useful for driver vendors to be aware of issues with their installers, I think AMD's collaboration with Phoronix (Phorogit) is the right way to go about solving installer issues.

The biggest fly in the ointment, at least on the Debian side, is that Ubuntu has a proven track record of ignore Debian compatibility, and many software vendors seem to get hung up on providing Ubuntu support, rather than simply providing DEBIAN support and making sure Ubuntu stays compatible with Debian.

It's of little surprise to me then that I run into installation problems with Debian (Pure) based distributions when the maintainer of the packages or scripts is working on a deliberate fork.

Now, maybe the Debian problems could be solved if Ubuntu developers focused on maintaining Debian compatibility.

However, I'm not sure bringing what amounts to be political and personal issues into a discussion involving whether or not an installed driver functions as expected in an installed enviroment.

[Nille_kungen]

Well if it was an packager bug then it would be ok.
But even when i try to install it without building packages it doesn't work

[Kano]

Do you really think the driver is different when you use Ubuntu instead of Debian packageing option? Just a little hint: the result will be the SAME. As the package names are different you just have to completely uninstall the other packages before. Ubuntu packageing needs dkms which is not in Debian by default. No big deal, because that package is for all archs and can be easyly fetched from Ubuntu:

http://packages.ubuntu.com/jaunty/all/dkms/download

[Redeeman]

The 180.xx drivers were a lot faster with KDE4 than the 177s and earlier versions, but it was still a pretty slow experience trying to use KDE4 with my 8600GT. I'm sure KDE4(and/or its dependencies) is partly to blame, but Nvidia's blob surely is also. One eight oh - STILL TOO SLOW. Ultimately, the slight performance increase was not enough to offset the bugs in the 180.xx series - loved that flashing red checkerboard on top of every non-fullscreen video I played. Oh and broken suspend too, that was so awesome.

So I'm back to KDE3 (again, not 100% thanks to Nvidia - but still, in good part), 177.82, and the same old never-gonna-get-fixed, totally-absent-for-no-good-reason TV out overscan control. If that had been fixed in 180.xx I would have put up with the other bugs. It's definitely the bug I hate the most, only it's not a bug so much as it was a decision some fucker at Nvidia made. The new Windows release doesn't need that anymore - so fuck the users of every other OS in existence right?

Does fglrx support running multiple X servers yet? Totally despise that one too.

In summary then, what I hate most is having Windows paradigms forced upon me. I shouldn't be forced to choose between Big Desktop and whatever other mode they offer when X.org offers the ability to run more than one Xserver. The fact that Vista doesn't need overscan controls shouldn't mean that users of every other fucking OS (other Windows versions too!) get the shaft.

There's 10s of millions of Linux users! Isn't that enough to get some fucking respect already?? Sadly I know the answer, and it's no.

actually, the blame is practically solely on drivers..

while 180.xx does make 2d for KDE4 almost good, its still far from even aproaching the speed of software rendering..

I tried to install a VM with KDE 4.2, and to my (not really so great) surprise, with KVM + debian + X with vesa driver + KDE 4.2, it was 100% smooth, something which nvidia cannot do even on their most highend card, and this was simple software rendering..

[cruiseoveride]

Nvidia 7600 - Nvidia Driver works beautifully, couldnt ask for a better driver

ATi HD4870 - ATi driver doesnt work for shit. Doesnt even load. Just get a blank screen and the system is frozen. Worst driver in the world, but then again, you all knew that.

[perpetualrabbit]

I like to have the `blur´ plugin in compiz enabled for translucent terminal windows. This gaussian blur makes text in the terminal much more readable than without because it makes it so much easier to tell foreground text from background text.

However, on my Eeepc 1000H it results in immediate screen corruption or even total lockup. This laptop has a gma 945 chipset I think. I think Michael Larabel has a 1000H too, am I right? He sure does a lot of testing on it.

Anyway, blur results in screen corruption and even locking up X in the Intel driver. It took me a little while to figure out which of the many compiz I had enabled was causing the problem.

[Saist]

Nvidia 7600 - Nvidia Driver works beautifully, couldnt ask for a better driver

ATi HD4870 - ATi driver doesnt work for shit. Doesnt even load. Just get a blank screen and the system is frozen. Worst driver in the world, but then again, you all knew that.

nope. don't know that. I do know that a 4650 of mine perked right up with the 9.2 drivers, and I do know that a 7900 GT of mine required running dpkg-reconfigure xserver-xorg every single time after installing the nvidia driver to set the correct resolution and refresh rate (although Debian has for some bone-headed reason removed one of the best tools ever created from Lenny)

[Kano]

The refreshrate options are not used by default, only screen res. Nvidia uses by default EDID infos from monitor.

[tils]

Hello,

I have 4 machines with opensuse 11.1 installed (2.6.27.19 and KDE 4.2).
MSI Wind (intel GME 450): Pretty good performance for that type of computer. Compositing works awesome. But if I shut it down, the vt switch doesn't work and I see only fullscreen graphic corruption.
While system runs I can switch vts, the problem only appears if I shut it down.

Acer Extensa Notebook (intel GMA): Good performance too. I had very bad opengl system lockups but after a driver update they were gone.

MSI X48C (Nvdia GF 8800GTS with 180.35): No problems at all. GL, Compositing, VT switching. No lockups or something like this. Window resizing is slow. If I logout and relogin GTK apps won't work and I have to reboot. But I don't know if this is an error of nvidia.

Abit P45 (GF 9600 with 180.35): No problems. I had lockups on video playback and Xid errors but there gone.

So overall I'm quite happy with linux graphix at the moment. The qt4.5 dependency from opensuse are making plasma ugly on my intel notebooks. On my nvidia PCs nearly all looks fine under qt4.5.
But thats opensuses fault...

[homerhomer]

My laptop has an ATI 200m card built in. And it seems that about every 3 months either a new driver will come out or a new kernel that breaks ACPI suspend or Hibernate. This holds true to the latest fglrx driver 9.2



sigh

oh and honorable mention goes to misc opengl extensions that are missing or broken (PBO)

Thanks

[Troggy]

I own a GTX 260. As you might know, the card is really power-hungry and because of that, NVidia added two powersaving modes. Problem is: As soon as you configure it with two output-monitors, it won't ever enter any of the powersaving modes.
NVidia is aware, promises a solution "some time in the future".

[kraftman]

actually, the blame is practically solely on drivers..

while 180.xx does make 2d for KDE4 almost good, its still far from even aproaching the speed of software rendering..

I tried to install a VM with KDE 4.2, and to my (not really so great) surprise, with KVM + debian + X with vesa driver + KDE 4.2, it was 100% smooth, something which nvidia cannot do even on their most highend card, and this was simple software rendering..

Yes, those performance problems aren't fixed for about one year! Nvidia really sucks like no one in this case.

Fglrx crashed X sometimes onmy PC, but nvidia blobs crashed entire system... No more nvidia for me.

[sabriah]

What bugs me the most are not the graphics drivers per se, but the audio drivers.

I have to, and far too often, kill X in order to restore the sound. And, logging out and logging in as another user often produce a mute computer.

I have still not understood the pattern but I think the graphics drivers are interlinked to tis issue, in some way. Maybe someone knows better.

I run Debain Sid/Experimental with KDE4 and the latest nvidia drivers on a 8800GT.

[Chewi]

I don't seem to have as many problems with fglrx as most people. The video tearing is annoying but I can live with it until the open driver supports 3D. The one problem that really bugs me is that logging out or trying to restart X in any way sends the card into a spasm. The display freezes, the graphics fan starts going full pelt and the only thing I can do is hit the power button, which at least shuts down the machine cleanly. If it was a hard lockup then the file corruption probably would have made me chuck my card by now.

[djtm]

My intel card loves to stutter:
E.g. I play a video or scroll down a web page and nothing moves for half a second or so, then more quickly for .1 second or so. It can get extremely annoying... I checked it out in latencytop and found nothing there. I'm using the most up to date drivers.

The current driver in Ubuntu 8.10 also often doesn't recover some suspend correctly. I have to switch back and forth between X and console to get a working screen. But that's not as bad.

[psycho_driver]

All's peachy with my 9800gt and the 5900fx before that. Also the HTPC with the 7050 chugs right along.

[Dragoran]

Currently NVidia focues on implementing new features but seems to ignore serious bugs which are affecting a lot of people.

The people writing the new features (VDPAU is meant here) aren't the same who are responsible for the X11 or kernel parts where the bugs are. So adding this new features does not take man power away from the other parts.

In fact, just while typing this news post a Pidgin window became unusable on NVIDIA hardware.

I suppose you where using compiz, this is not a bug in the nvidia driver but a race condition in the way the XDamage extension work.

The reason why it does not happen with older drivers was that not all pixmaps where stored in video memory and therefore it was unlikely to trigger. Now pixmaps are stored in video memory by default which improves performance a lot but causes this.

This corruption does not happen if you use strict bindings (which is used in all open source drivers and fglrx) but on G80+ hardware it is slow due to a bug, that's why people/some distro scripts enable loose-binding for nvidia hardware.

Plagman (an nvidia dev) investigated the issue and told me that it should be fixed in the next driver series. (18x ; where x > 0 )
It cannot be backported to 180.xx because it was a rather huge change.

There might be problems and bugs in the nvidia drivers, but they aren't that bad as this article tries to show it.

In fact the 180.xx drivers work very well for me (compiz works fine with good performance, 2d and 2d is fast, video playback works fine, suspend/resume works fine.)

[unix_epoch]

The issue that annoys me the most is that when there is more than one video card present (such as in an SLI system) I have to enter PCI IDs in the xorg.conf manually. This may have been fixed, but why can't X just detect which video card was enabled by the BIOS and use it?

[Hephasteus]

The issue that annoys me the most is that when there is more than one video card present (such as in an SLI system) I have to enter PCI IDs in the xorg.conf manually. This may have been fixed, but why can't X just detect which video card was enabled by the BIOS and use it?

That's what everyone seems to be working on. Not just for making sytems easier to set up or install with less things to do but also for virtualization. It just seems that it's a long boring process of going through all the bios's and cataloging what and how they do everything.

I just reset up windows XP and fedora 10 and hands down fedora 10 beat windows to a pulp. Configuring repositories, installing few drivers, getting flash working versus hours of installing tons of drivers and driver packages under windows. I couldn't believe how things have turned so ugly for microsoft on hardware support. Granted the end result under windows is bit more polished but at this rate in 2 years nobod will want to set up windows because it will be considered HARD versus linux.

Nouveou is giving Nvidia a kick in the pants but everyone seems to be waking up to fact that the number of linux systems keeps growing rapidly. Fedora broke 12million, ubuntu has god knows how many, and if they are going to use video cards as processors it's all get done for servers and workstations. Everyone is climbing all over each other to get it done. It's hard to believe i have supported opengl 3.0 on linux and only 2.1 on windows. :D

[NeoBrain]

- Creating a second instance of the X server on another display hardlocks the system
- Resuming after S2RAM results in corruption and a freezing X server (mouse movable but hardlocks if you try to kill X)

Apart from that, I have not been able to use Compiz's "Blur windows" effect since 9 months or so.

[.CME.]

let's see
geforce4 420 go: 96.43.10, the nvidia drivers only detects a 969x768 display and there is a black bar in the right (this is also in the windows driver since detonator 28.xx!!) with a custom EDID it works ok, but i can not use anymore a second display (ist just says that it doesn't detect any second monitor in the xorg.0.log)

geforce 7800GTX: well, overall it works good, but xvideo is borked. when i use my TV at 50hz and play some 50hz material, X gets very slow and then there is a bad judder in the video (mplayer, with OpenGL there is no problem, and when playing 25fps videos, it doesn't display everytime every picture twice, but sometimes the first picture 1 time, the second 3 times (xvideo AND OpenGL) which ends also in a bad judder

ati rage 128 xpert 128: r128:
short: a MESS
long: xvideo on interlaced modelines is broken
OpenGL is slow, some application (blender, google earth) has graphic errors and after 1 minute the complete system freezes (no ping response, no sysrq etc.)
on some OpenGL apps it just crashs with "r128 timed out" and when using OpenGL the mouse cursor gets a random color and when i move a windows fast while running an OpenGL app, X freezes for 20-60sec etc.
ah, and DMA for xvideo is also broken and since debian lenny Xorg is slow as hell on this machine :\

[Na-Fiann]

I guess that tearing in xv is one that bugs me a lot, since I watch all my video on my pc.. especially how it even crashes my desktop when compiz is running.
Also, not really a bug, but in ut2004 I can get 90 fps easily on my monitor's native resolution in windows. On linux however, it's having trouble getting to 25 fps just rendering the wall, and this is on an ati 4830 which is capable of much better!
Other than that, things work fine here :)

[bugmenot]

I hate fglrx, because there are lockups from time to time in different scenarios. compiz, videos, vt switching or with full screen games while pressing alt-tab.
I'm hoping for the free driver. :)

[beecher]

beecher, I believe both the radeon and radeonhd drivers have tear-free Xv support today. You'll need to pick up drm from either drm- next or the 6xx-7xx git branch of drm, but the X driver code has been merged to master on both radeon and radonhd.

I read that but I think it wouldn't be wise to use the free driver now without any kind of powermanagement implemented. Wouldn't my card run a little too hot after a while on my laptop ?

[lordmozilla]

fglrx catalyst 9.2 (or 9.1 same behaviour), with compiz + emerald - crashes to nothing during fullscreen video playback after 1 second.

Apart from that on my hd4850 everything is peachy and i dont' mind fglrx too much, nvidia drivers had their problems too.

[kernelOfTruth]

fglrx | AMD/ATI catalyst ! :mad:

they:
- don't support proper video or opengl playback during composited desktop (flickering)
- don't support preemptive rcu
- don't properly support fully preemptive kernel
- don't support rt-kernel
- often lead to black screen which can't be restored
mostly when running composited desktop after some time
and several users logging in and out of their account (running gdm)
only MagicSysRQ Key works in those cases
- STILL hardlocks kernel when logging out of the composited desktop session (using gdm) OR when being forced to restart gdm/kdm several time OR remove the fglrx kernel-module and restarting gdm
- providing bad performance compared to an 7600 GT (I'm using an Radeon HD 4850 !)

I'm really looking forward to the xf86-video-radeonhd when it will be support powermanagement, 3D and other stuff properly my desktop mostly will be free :)

I don't get it why everybody is ranting about nvidia (I had problems with my nvidia card and that was the reason why I bought my first radeon card after a long period of time (4-5 years). Before that I had sworn not to buy any AMD/ATI card anymore because of their abysmal linux support which has improved somewhat but I'm still far from satisfied)
ATI/AMD's drivers have much improved but they're still MUCH worse than nvidia (unfortunately)

[mitrol]

radeon and radeonhd - X freeze with EXA, no reliable resume from suspend

fglrx - black screen on user switch on Ubuntu 8.10 and sometimes on logout with openSUSE (catalyst 8.11), garbled screen on logout from driver 8.12 till 9.2 with openSUSE 11.1 and no EXA support.

[kernelOfTruth]

I hate fglrx, because there are lockups from time to time in different scenarios. compiz, videos, vt switching or with full screen games while pressing alt-tab.
I'm hoping for the free driver. :)

exactly !


fglrx | AMD/ATI catalyst ! :mad:

they:
- don't support proper video or opengl playback during composited desktop (flickering)
- don't support preemptive rcu
- don't properly support fully preemptive kernel
- don't support rt-kernel
- often lead to black screen which can't be restored
mostly when running composited desktop after some time
and several users logging in and out of their account (running gdm)
only MagicSysRQ Key works in those cases
- STILL hardlocks kernel when logging out of the composited desktop session (using gdm) OR when being forced to restart gdm/kdm several time OR remove the fglrx kernel-module and restarting gdm
- providing bad performance compared to an 7600 GT (I'm using an Radeon HD 4850 !)
- their support for new linux kernel versions is ridiculous !

I'm really looking forward to the xf86-video-radeonhd when it will be support powermanagement, 3D and other stuff properly my desktop mostly will be free :)

[wswartzendruber]

Well now that we have X Server 1.6, a vast majority of my complaints (two years now) have gone out the window. Now all I wish for is proper VSYNC and VT switching (without freezing).

[deanjo]

I just reset up windows XP and fedora 10 and hands down fedora 10 beat windows to a pulp. Configuring repositories, installing few drivers, getting flash working versus hours of installing tons of drivers and driver packages under windows. I couldn't believe how things have turned so ugly for microsoft on hardware support. Granted the end result under windows is bit more polished but at this rate in 2 years nobod will want to set up windows because it will be considered HARD versus linux.


Take a linux distro from the 2001 era and see how well it runs out of the box on your modern system. Slipstreaming and integration of XP patches and drivers can relieve the issues you have with windows XP. nLite makes the whole process painless. In contrast if you downloaded the Windows 7 Beta you would see that on modern hardware even things like tv cards, bluetooth, sli, fakeraids, etc are supported on a vast majority of hardware out of the box, what isn't is usually easily found with a windows update. Comparing a OS that was put together 7+ years ago to one that was released a few days ago isn't exactly a fair comparison when it comes to what it supports out of the box.

Let's be fair, if someone was to compare Windows 7 to Redhat 7.2 there would be a lot of screaming.

It's hard to believe i have supported opengl 3.0 on linux and only 2.1 on windows. :D
Not sure what video card you have but both ATI and Nvidia have ogl3 window drivers.

[RobotMarvin]

The people writing the new features (VDPAU is meant here) aren't the same who are responsible for the X11 or kernel parts where the bugs are. So adding this new features does not take man power away from the other parts.


You sure about that...? It doesn't look that way if you consider some postings over at nvnews.net where some of the devs state that they had a look at issue x or issue z but don't have the time to fix it just yet. Or similar things.


There might be problems and bugs in the nvidia drivers, but they aren't that bad as this article tries to show it.


I really beg to differ. My 8600 GTS card died on me in January and I had to get a replacement fast because I need my computer for work. Since I already tried an AMD 4870 in early Q4/2008, I decided to go with NVidia again because the experience I made with the AMD one wasn't too good, to say the least. So after driving around in my city, I ended up with a Palit 9800 GTX+ which is actually overkill for me but at least silent and was the only thing I could get which wasn't a total vacuum cleaner soundwise.

To make a long story short: with the 8600 GTS, my system was rock solid, no freezes or anything like it, no matter what 180.xx release I used. Once I got the 9800 GTX+, I had total freezes with graphical corruption. Totally erradic. Working was impossible that way because you could get everything from zero to 10 freezes a day. Disabling the Composite Extension and similar things naturally didn't fix the issue. Only reverting back to 177.82 "fixed" it for me... well... more a workaround actually.

Head over to nvnews.net and have a look at that lengthy thread. You will soon realize that a lot of people are bitten by this bug and there are a total of <5 dev postings there if I remember correctly which are more or less useless to be honest. And this is still ongoing since Nov 08.

So... I'd say, and many from nvnews.net would agree, that depending on your GFX card, at least the 180.xx series is a total mess and nvidia is unable to fix it or keep their users in the loop about what's actually going on.

Best regards,
matthias.

PS. I don't use Ubuntu but Gentoo on x86_64 and also tried X 1.6.0 snapshots, different kernels and what else. The bug is _clearly_ with nvidia.

[december]

I hate any sort of graphic card bugs, really. :rolleyes:

I can understand all the latest multi-gpu stuff is still rather buggy, but there really shouldn't be any excuse for basic 2D and 3D functionality to be unstable and riddled with bugs. As I don't really play games, I put my money on any card that I assume will guarantee the most trouble-free graphical experience. If ATI and NVidia don't get their act together, I can only hope one of those open-source graphical cards will see the light of day, or a third manufacturer (Intel? S3?) to gobble up some of the ATI/NVidia market share by having superior drivers for products with reasonable enough performance.

I don't care for half-working products, be it due to hardware or software bugs. There just is no excuse for selling something that doesn't work as advertised.

[deanjo]

I hate any sort of graphic card bugs, really. :rolleyes:

I can understand all the latest multi-gpu stuff is still rather buggy, but there really shouldn't be any excuse for basic 2D and 3D functionality to be unstable and riddled with bugs. As I don't really play games, I put my money on any card that I assume will guarantee the most trouble-free graphical experience. If ATI and NVidia don't get their act together, I can only hope one of those open-source graphical cards will see the light of day, or a third manufacturer (Intel? S3?) to gobble up some of the ATI/NVidia market share by having superior drivers for products with reasonable enough performance.

I don't care for half-working products, be it due to hardware or software bugs. There just is no excuse for selling something that doesn't work as advertised.


In all fairness there really isn't a game for linux that requires multicard rendering. Even the most demanding ( which I would have to guess is ET:QW ) runs smooth as butter on semi modern hardware. The most demanding games out there usually require the use of wine, native ports are pretty easy on the requirements.

[TheK]

The _by far_ biggest problem is the miss of Hybrid Power in the nVidia-driver. Bad enough, that such a feature is needed - GPU makers (ATI isn't better) should be able at least not to increase their idle power consumption over the time (they did by factor 3 over the last 3 years), as CPU makers did - but as they were unable to solve the problem the easy way, at least the complicate way should be allowed for non-Vista-users. In some parts of the world electrical power is expensive enough to need more money to run a modern GPU for 1 1/2 years in idle mode than to buy it!

Another issue is the intel 3D driver. Yes, their hardware sucks in 3D performance; it sucks deadly - but their crappy drivers make this suck even more. Their 3D performance is _4 times_ better on MacOS and I'm sure, the driver isn't perfect there too...

[deanjo]

The _by far_ biggest problem is the miss of Hybrid Power in the nVidia-driver. Bad enough, that such a feature is needed - GPU makers (ATI isn't better) should be able at least not to increase their idle power consumption over the time (they did by factor 3 over the last 3 years), as CPU makers did - but as they were unable to solve the problem the easy way, at least the complicate way should be allowed for non-Vista-users. In some parts of the world electrical power is expensive enough to need more money to run a modern GPU for 1 1/2 years in idle mode than to buy it!

Another issue is the intel 3D driver. Yes, their hardware sucks in 3D performance; it sucks deadly - but their crappy drivers make this suck even more. Their 3D performance is _4 times_ better on MacOS and I'm sure, the driver isn't perfect there too...

On the newer nvidia cards hybrid power really isn't needed. a GTX 280 for example uses 25 watts when not running in premium 3d mode. ATI on the other hand will gobble the power just sitting idle. It's even less with the new die shrinks.

Now lets put some real world numbers behind the "green" concern

25 watts = 25 watts per hour.
25 watts per hour = .025 kWh usage in an hour.
.025 kWh x .12 cents = .003 cents per hour
16 hours a day x 365 = 5840
5840 hours x .003 cents per hour = 17.52 per year.
17.52 per year divided by 12 months = $1.46 a month.

Even with running that card in premium mode for a while (say 10-20%) it still amounts to a paultry amount. Chances are you waste more money on leaving some lights on at home through out the year.


http://ixbtlabs.com/articles3/video/spravka-gt2xx-p4.html

[jonnycat26]

The absence of VTs.
The absence of kernel panic notifications.
The absence of monitor hot-plugging.

Proprietary NVIDIA driver, obviously.

Yeah, I have none of those issues with the Proprietary NVIDIA driver. Not a single one of those has affected me, other than kernel panics, but then, I only had kernel panics with fglrx.

I had, until last week, a radeon HD 4650 in my system. After 3 catalyst releases, each of which fixed old bugs while introducing new bugs, I just ripped the card out of the system and have it sitting in the closet.

I replaced it with a NVidia 9400GT. Cheap card, not nearly as fast as my 4650, but it works.

I had been a NVidia user for years prior to the Radeon purchase, and I had always taken for granted the fact that NVidia drivers more or less work under Linux, no questions asked. The list of problems I had with fglrx was nothing short of mind boggling, and it astounds me that ATI can foist fglrx upon the general public. It's barely alpha quality.

[wfeltmate]

I keep reading all these issue people have with the ati drivers and how they are just so horrid and widespread. Besides an incomplete or missing feature, none of which have ever been more than a 'oh hey, it would be nice if they would add this', it has always worked for me. I've used various version of Ubuntu, Mandrake and Suse and no issues.

Point is, both nvidia and ati have issues with their drivers. You know this as well as the rest of us, so please stop being so immature as to trash talk one company and make it seem like the other is perfect.

[JeanPaul145]

What bugs me the most are not the graphics drivers per se, but the audio drivers.

I have to, and far too often, kill X in order to restore the sound. And, logging out and logging in as another user often produce a mute computer.

I have still not understood the pattern but I think the graphics drivers are interlinked to tis issue, in some way. Maybe someone knows better.

I run Debain Sid/Experimental with KDE4 and the latest nvidia drivers on a 8800GT.

If you use Sid you can't expect anything to work well. In fact, the official stance is that you should _expect_ frequent breakage.
So here's my advice to you: switch to Debian testing or another distro, and see what happens.

[SyXbiT]

I own computers with intel, ATI and Nvidia graphics cards.
Sure, they all have bugs, but I'd definitely rate them in this order

Intel
Nvidia
RadeonHD
...
...
...
...
Catalyst

[bridgman]

25 watts = 25 watts per hour.
25 watts per hour = .025 kWh usage in an hour.
.025 kWh x .12 cents = .003 cents per hour
16 hours a day x 365 = 5840
5840 hours x .003 cents per hour = 17.52 per year.
17.52 per year divided by 12 months = $1.46 a month.

Something seemed odd about your numbers. The result is right but...

25 watts = 25 watt-hours per hour.
25 watt-hours per hour = .025 kWh usage in an hour.
.025 kWh x 12 cents = 0.3 cents per hour (or 0.003 dollars/hour)
16 hours a day x 365 = 5840
5840 hours x 0.3 cents per hour = 17.52 per year.
17.52 per year divided by 12 months = $1.46 a month.

Things like that catch my eye. It's what happens when your father is an accountant :D

[curaga]

Whoa, 6 pages already :o

Guess people really hate graphics driver bugs. Me too.

In no particular order:

Neomagic - XV has moving colorful pixels. Understandable as they never released docs.

Ati (radeon driver) - VT switch sometimes hangs, missing 3d features (comparison: q3-based games run great natively, 100+ fps, but the same games run like crap in wine, with corruption, assuming because Wine maps some features to OGL 2.0 things), sucky 2d performance
Ati (radeonhd): You cause a black screen and a complete hang for me.
Early ati (mach64): Nice otherwise, but the full potential of the cards is not used. There could be both full Render accel and better 3d, but, meh, not popular anymore, nobody capable willing to help.

Intel - You were great hon, but a bit too slow on the 3d side.

Sis - you should be eaten alive. Not only is your driver slower than vesa, it also does not have 3d accel and produces corruption. Often there's more corruption on screen than content, measured in pixels.

S3 - you did release some docs early. However your drivers were left rotting in Xfree 3, because you weren't popular.

Matrox - I was really in love with your image quality and 2d performance. You even had acceptable 3d in some games. The only thing I can whine about is the fact the drivers never used up the full potential - partly due to different microcode than in the windows drivers, partly due to something else, perhaps just nobody had interest to implement opengl 1.4, multitexturing, and other niceties.

[TheK]

25 watts = 25 watts per hour.
25 watts per hour = .025 kWh usage in an hour.
.025 kWh x .12 cents = .003 cents per hour
16 hours a day x 365 = 5840
5840 hours x .003 cents per hour = 17.52 per year.
17.52 per year divided by 12 months = $1.46 a month.

yes, US-pricing... We pay about 0.25€ per kWh here in Germany. In other words: those 25W are close to 1€ _per Week_ or about 36€ per Year. From that money I can eat for 2-3 Weeks or buy a new graphics card every 2 years.

The other changes I currently plan for my PC would save 25W (+12,5W for the old card), but those "great innovative GPU makers" manage to reduce this to those 12,5W in best case; several cards have even much more...

[Ant P.]

Something seemed odd about your numbers. The result is right but...

25 watts = 25 watt-hours per hour.
25 watt-hours per hour = .025 kWh usage in an hour.
.025 kWh x 12 cents = 0.3 cents per hour (or 0.003 dollars/hour)
16 hours a day x 365 = 5840
5840 hours x 0.3 cents per hour = 17.52 per year.
17.52 per year divided by 12 months = $1.46 a month.

Things like that catch my eye. It's what happens when your father is an accountant :D

Heh, I was about to ask how the US gets electricity 200 times cheaper than the UK...

VerizonMath strikes again.

[Aradreth]

I have a 8400M GS and 8800GTX and have only had one problem the nvidia driver (I think it is due to the nvidia driver at least...) when using suspend2disk (never get my screen back) which is mildly annoying. Although I do wish I had a intel card instead of the nvidia card in the lappy though. (Longer battery life...)

[deanjo]

yes, US-pricing... We pay about 0.25€ per kWh here in Germany. In other words: those 25W are close to 1€ _per Week_ or about 36€ per Year. From that money I can eat for 2-3 Weeks or buy a new graphics card every 2 years.

The other changes I currently plan for my PC would save 25W (+12,5W for the old card), but those "great innovative GPU makers" manage to reduce this to those 12,5W in best case; several cards have even much more...

Actually it the average Canadian rate in Canadian dollars. Here it's actually 8 cents/kWh. Going @ US prices with the exchange rate that would be about 6.5 cents /kWh.

[some-guy]

Well, it was compiz not redrawing properly, but autostarting 'nvidia-settings -a InitialPixmapPlacement=1' fixed it :)

[jhansonxi]

My nVidia GeForce 7300GT works without problems on Ubuntu 8.04.2 (Hardy Heron) x86_64 with the closed-source 173.14.12 driver. I'm using dual-head without Compiz. UT2004, Doom3, and Quake4 all perform flawlessly at 1280x1024. My Asus M3A78-EM motherboard uses the 780G with (HD 3200) integrated video. It also worked without problems using fglrx but I don't use it because the second port doesn't support analog monitors.

My Toshiba M35X-S114 laptop has the 855GME chipset which had very broken video support in Hardy. With Intrepid (8.10) it's stable.

I have a couple of boards with the Intel 875P AGP chipset (82875P) that when coupled with PowerColor HD2600 cards are completely unstable. As soon as I try anything that uses OpenGL the system freezes within a few seconds. No keyboard lights flashing panic, no kernel messages on a serial console, nothing. This occurs with the fglrx default driver in Hardy and anything newer. Others with a similar setup have found that increasing the AGP aperture size to match the card's memory can make it work but my cards are 512MB and the BIOS maxes out at 256MB.

[TheK]

Actually it the average Canadian rate in Canadian dollars. Here it's actually 8 cents/kWh. Going @ US prices with the exchange rate that would be about 6.5 cents /kWh.

..no wonder, that ATI and nVidia don't care for power consumption :/

So I guess, I can be happy, that "only" 1/3 of the next PC's idle power will go to the GPU.. If they'd at least make a slower (no, not the 9400 GT, then I can stay with my current one) GPU with those old "low" values of 15/45W...

OK, back on topic: I want VDPAU on PureVideo Generation 1 ;p

[domicius]

Not too much to complain about except for one that I'm not really sure where it lies because I've experienced it with open source and proprietary driver for my ATI Mobility Radeon X1300. Maybe somebody here knows something about it because I don't know even to whom I should report it.

The bug I'm talking about is not being able to keep the system running for more than 10 days (of uptime) with periodic suspends to RAM and resumes from suspend. What I'm saying is that after an x number of days (really can't say how many days but I've never seen an uptime value of 10 days) of using laptop and constantly using suspend to RAM, once it just doesn't wake up and I have to shut it down hard.

What may be related is that during any resume it often goes back to suspend while trying to resume. I lift the lid of a suspended laptop and I see it resuming but when it should have resumed it has in fact gone back to suspend to RAM. Simply trying to resume it once again is successful in probably 99.99% cases - I might be wrong but I think it never did this twice in a row, although it happens almost every other time I try to resume it from suspend. Meaning: at least once a day which is rather annoying.

I'm running Debian unstable always with the latest fgrlx driver and this is the only bug that doesn't allow me to say I have a stable system. Yes, even though it's Debian "unstable", it's almost rock solid except for this annoying bug.

[nightalon]

I hope no one has posted this one so far, but I really don't have time to read the whole thread...sorry...please don't flame me.

Having to reboot or restart X to change monitor setups is a pain, but I'm guessing new revisions of Xorg and RandR will help resolve this.


However, using Compiz full time, (which I think is akin to using the 2D and 3D acceleration features in OS X and Vista) one notices that 3D bits are not morphed and modified with the Compiz animation using the ATI and Intel drivers.


NVidia drivers do not seem to have this issue.


ATI 3D acceleration just flickers in Compiz, and using composite/redirection moves the background, but not the 3D accelerated frame. Hence Compiz and Google Earth don't work together.

Intel 3D acceleration (12 months ago anyway) didn't have the flicker problem, but it did have the issue where the 3D zone wasn't morphed with the rest of the desktop.

Of course the open source NVidia drivers (though maybe not Nouveau) don't have 3D acceleration, and Google Earth simply crashes with the open source ati driver. Annoying. Ditto for hi-res video and Compiz sumultaneously.


My experiences stem from using the latest ATI and NVidia binary drivers as of 3 days ago, and whatever the 8.10 open source driver packages are. Of course multi-desktop setups seem to work a lot better with these open source drivers. That is, of course, understandable.


Point of information: given the redraw issues, when a program is locked up, for example, the technology behind Compiz-less Metacity in Ubuntu seems to be a lot like XP's rendering system, while Compiz seems to be a lot like Vista's rendering system. Mac's rendering system seems to accel at 2D acceleration and doesn't have the redraw problems. I have no idea if Mac's window acceleration are somehow also done in 3D. (as 3D textures)

Enlighten me if I am grossly misinformed!

My experiences are with NVidia Quadro NVS110M, NVidia GeForce 9800GTX, Intel GMA950, and ATI Radeon Mobility X1400. Maybe my experience doesn't suffice.

[Saist]

wanted to bring up something that crossed my mind after chewing out the ArchLinux guys on another site.

How many times have you come across a bug and tried to do something about it? I mean beyond just making a forum post somewhere and complaining about it? How many times have you stepped back and actually LOOKED to see if a Vendor has a support method to connect with other users and developers?

The thing is, ATi's linked to a bug reporting system for literal years now : http://ati.cchtml.com/ : and it's linked right from the driver download page. The actual driver develoeprs also are known to browse the bugs, and I'm aware of various readme's included in driver releases referring back issues in the bug reports.

While I see that some people here have filed grievances with the ATi drivers, such as the inability to install, or encountering black screens, I'm not seeing corresponding bug reports actually... you know... FILED.

By comparison, in the whole issue of "Nvidia doesn't have an open-source strategy" there is no way there to connect with the developers or file bugs, so maybe many of the people who have had problems with ATi and use Nvidia cards think that lacking such a feedback mechanism is well.. normal.

***

Basically, I would encourage readers of this thread then: if you have a bug with an ATi card using FGLRX, FILE A BUG REPORT.

If you happen to have an Nvidia card... um... I dunno? Company hasn't responded well to letters or common sense? Anybody got a baseball bat handy?

[deanjo]

By comparison, in the whole issue of "Nvidia doesn't have an open-source strategy" there is no way there to connect with the developers or file bugs, so maybe many of the people who have had problems with ATi and use Nvidia cards think that lacking such a feedback mechanism is well.. normal. ?

I really wish people would read the README (or even what is plainly printed on the driver download page)

If you have any questions or problems, please check the NVIDIA Linux discussion forum (http://www.nvnews.net/vbulletin/forumdisplay.php?s=&forumid=14). If you don't find an answer to your question there, you can send email (in English) to linux-bugs@nvidia.com.

[Kano]

Did you ever look at

http://www.nvnews.net/vbulletin/forumdisplay.php?f=14

That's a forum for Nvidia users.

[Saist]

I really wish people would read the README (or even what is plainly printed on the driver download page)

If you have any questions or problems, please check the NVIDIA Linux discussion forum. If you don't find an answer to your question there, you can send email (in English) to linux-bugs@nvidia.com.

and you miss the point. There's no tracking record with that. You send off an email, and you have nothing to link to, nothing to point, no reference.

That's not support.

[deanjo]

Did you ever look at

http://www.nvnews.net/vbulletin/forumdisplay.php?f=14

That's a forum for Nvidia users.


There is also the #nvidia channel on freenode and Plagman is constantly on there getting and giving feedback as well as a few other nvidia guru's.

[Saist]

Did you ever look at

http://www.nvnews.net/vbulletin/forumdisplay.php?f=14

That's a forum for Nvidia users.

and you get the same chew-out as well for missing the point.

did you ever look at Driverheaven.net ? Rage3d.com ? Phoronix.com ?

Yes, CatalystMaker reads Driverheaven.net and Rage3D. Yes, Mr. Bridgman and Alex read Phoronix. That doesn't make them feedback for reporting bugs.

[deanjo]

and you miss the point. There's no tracking record with that. You send off an email, and you have nothing to link to, nothing to point, no reference.

That's not support.

You fire off a email you get a case number and a direct contact and email of the person looking into it. Trust me, been doing it for years and they are very good with responding and keeping you up to date with issues. That is better then most bugtrackers out there where crap sometimes goes unlooked at for sometimes months and years.

[Kano]

@Saist

Bugtrackers can be useful for huge projects, but even if you track em offically then it does not lead to faster solutions. If there is a direct way to get in contact with a developer I definitly prefer that over using a bugtracker. Some distro devs only do something when there is an open bug, maybe because they have to justify the time they spent on bug x to solve it and have got stats who did solve the most of em. Many huger problems you can found repeatedly on a forum, so it should be possible for dev to figure out where to look first too. But if they are unable to fix bugs then the most informative bugtracker will not help em.

[deemz]

Chrome9!! Chrome9!

Random screen corruption that can only be fixed, usually through a restart

Chrome9 Chrome9

Help VIA

[Hephasteus]

Take a linux distro from the 2001 era and see how well it runs out of the box on your modern system. Slipstreaming and integration of XP patches and drivers can relieve the issues you have with windows XP. nLite makes the whole process painless. In contrast if you downloaded the Windows 7 Beta you would see that on modern hardware even things like tv cards, bluetooth, sli, fakeraids, etc are supported on a vast majority of hardware out of the box, what isn't is usually easily found with a windows update. Comparing a OS that was put together 7+ years ago to one that was released a few days ago isn't exactly a fair comparison when it comes to what it supports out of the box.

Let's be fair, if someone was to compare Windows 7 to Redhat 7.2 there would be a lot of screaming.


Not sure what video card you have but both ATI and Nvidia have ogl3 window drivers.

Pretty fair comparison as windows vista would require the same amount of hardware messing. Every package I set up with xp i would have had to set up with vista. I didn't even count having to set up firewall and spyware blockers and everything.

Have to check it again but think Furmark was reporting GL 2.1 on lastest driver under windows.

[kraftman]

@deanjo

Take a linux distro from the 2001 era and see how well it runs out of the box on your modern system.

Windows has even problems with current hardware. I had xp and GF2 MX and I had to replace my card, because xp refused to install (sp1 fixed this after long time waiting). An only reason why windows 'supports' many devices is third party members which provide drivers for it.

[curaga]

If they'd at least make a slower (no, not the 9400 GT, then I can stay with my current one) GPU with those old "low" values of 15/45W...But they do? The low end usually has those values or less. My X1300 uses 30W max, reportedly 10W idle (possibly more under Linux, since not all power saving stuff is supported). The HD4350 should have similar values.

[Wyatt]

I think my anger for the week is squarely directed at fglrx.ko, though that's more because it's more fresh than what I hold for nvidia.ko (which is easier on the kernel, but has other problems).

fglrx.ko has some annoying kernel requirements.
fglrx is not compatible with CONFIG_PREEMPT_RCU.
fglrx requires CONFIG_UNUSED_SYMBOLS.
fglrx requires CONFIG_PCI_LEGACY (c'mon, guys, pci_find_slot() was deprecated two years ago at this point...).
fglrx is not compatible with CONFIG_PARAVIRT.
And then, even after I got it installed, I was getting "BUG: using smp_processor_id() in preemptible [00000000] code: X/26646" in my dmesg and the system would restart. It would be nice if we could get a performant driver that isn't in danger of being deprecated into inoperability.

With an FX5200, nvidia.ko was giving me strange green corruptions and would cause a crash in the app or an oops if I was unlucky. Highly uncool. Before that, with the 8800 I was borrowing, it was very bad at doing multiple monitors: Twinview is an ugly, ugly hack and nVidia should be ashamed of themselves.

radeon.ko on my r300 chipsets has performed swimmingly. Suspend on my laptop is faster and more reliable, now.

If I can offer any real complaint, it's that the documentation for the whole open stack could really use some amount of cleanup, pruning, updating, etc.

For example...well, I guess the X.org wiki is the "official" documentation for radeon and radeonhd. Is it? I would never be able to tell from the main page. Were it not for Google, I don't know that I would ever have made it to the radeonhd page. And were it not for radeonhd, I'd never be able to find the radeon page for the place. Ahh. this gem caught my eye, too:
"I have an ATI graphics card, which driver should I use ?
The GATOS project..."
Yes, the same GATOS that hasn't seen a commit in four years or so. This is straight from the VideoDriverFAQ at the X.org wiki. If you don't just stop reading after three words, it also states that r300 and up is still highly experimental and doesn't even mention radeonhd.

So I think something definitely needs to be done about documentation management for the open drivers and acceleration: clearly leaving it under the banner of X.org is not working out so well and I feel it ends up being misleading to the end user.

[deanjo]

@deanjo



Windows has even problems with current hardware. I had xp and GF2 MX and I had to replace my card, because xp refused to install (sp1 fixed this after long time waiting). An only reason why windows 'supports' many devices is third party members which provide drivers for it.

I had the EXACT same card when XP came out and it installed fine.

[szczerb]

You could have had the a card based on the same exact model of GPU but you have no idea what card he had.

[deanjo]

Pretty fair comparison as windows vista would require the same amount of hardware messing. Every package I set up with xp i would have had to set up with vista. I didn't even count having to set up firewall and spyware blockers and everything.

Have to check it again but think Furmark was reporting GL 2.1 on lastest driver under windows.

And how well did Fedora 5 install your dmraid, tvcard, free video, printer, soundcard etc on that? On both those systems a firewall would have been enabled by default. If your going to compare linux to windows at least use the same generation of OS. As I said before Windows 7 has improved a lot of your concerns and slipstreaming a updated disk is childs play now days.

And where does furmark report GL level?

[deanjo]

You could have had the a card based on the same exact model of GPU but you have no idea what card he had.

Every card manufacturer of that era followed the reference design. Even hard mods like doing the Quadro mod was universal on the GF mx2 (which I did later on and worked on XP as well).

[szczerb]

Every card manufacturer of that era followed the reference design. Even hard mods like doing the Quadro mod was universal on the GF mx2 (which I did later on and worked on XP as well).ok, then - my bad ;] I don't really remember, so I assumed :)

[deanjo]

ok, then - my bad ;] I don't really remember, so I assumed :)

No worries, because they all followed the reference design the Quadro mod was a piece of cake to do. Did many of them back in the day for friends and family. Even the resistor numbers were consistent across brands.

PS: Just for FYI in that era Leadtek was the designer of the reference boards.

[kraftman]

I had the EXACT same card when XP came out and it installed fine.

Strange, because my friend had the same problem. I had Inno 3D Geforce 2mx 200 and xp without sp1 just hung when I tried to install it. I had to replace my card, install xp and then replace card again. As I said windows has rather good hardware support just, because third party members. The same problem here:

http://www.techimo.com/forum/graphics-cards-displays/26385-video-card-causes-system-freeze-wont-install-xp.html

Hey Deanjo! I don't believe you:

There is a known problem with installing WinXP on a system with a GeForce 2 MX video card. Borrow another type of video card from another PC if you can. Install WinXP and then you'll be able to put your GeForce 2 MX video card back in the system after the operating system is installed.

http://www.viaarena.com/default.aspx?PageID=3&FCat=15

What I noticed is you trying to 'whitewash' proprietary crap.

Oh, and I had to find some exotic patch, because xp hanged after few minutes playing in some games. It looks patch was needed to proper support Abit KT7A motherboard. That's windows hardware 'support'.

And how well did Fedora 5 install your dmraid, tvcard, free video, printer, soundcard etc on that?

Why Fedora? I ran Aurox 9 as far as I remember (distro based probably on red hat 7; xp times) and my hardware was fully supported. It installed without problems 'even' with gf2 mx.

@szczerb

ok, then - my bad ;] I don't really remember, so I assumed

No, not your bad, because sp1 fixed my and my friend issue. I bet you can find something about this problem on ms site (if they don't hide some issues...).

[deanjo]

Strange, because my friend had the same problem. I had Inno 3D Geforce 2mx 200 and xp without sp1 just hung when I tried to install it. I had to replace my card, install xp and then replace card again. As I said windows has rather good hardware support just, because third party members. The same problem here:

http://www.techimo.com/forum/graphics-cards-displays/26385-video-card-causes-system-freeze-wont-install-xp.html

Hey Deanjo! I don't believe you:



http://www.viaarena.com/default.aspx?PageID=3&FCat=15

What I noticed is you trying to 'whitewash' proprietary crap.

Oh, and I had to find some exotic patch, because xp hanged after few minutes playing in some games. It looks patch was needed to proper support Abit KT7A motherboard. That's windows hardware 'support'.


LOL, again exact same hardware, The system that I was using at the time was a KT7A-Raid and a Inno3d card. It was a board with a chipset riddled with bugs (KT133A) and had nothing to do with the OS. That chipset was riddle with bugs and if you knew your Abit history you would also know that the solution to the issue was to increase the AGP driving strength as boards were way out of spec when a more power hungry card was put in due to the bad cap issue ( which by the way Abit at the time would warranty your board even though it may have been expired and send you a new one). Abit later increased the default AGP driving strength to compensate for it with a BIOS update. It had NOTHING to do with SP1. The same exact issues plagued other boards of that era as well, but the KT133a is one of the worst. If you slapped that card in any other chipset with good caps it would have ran fine. The same issues were seen with linux as well.

[kraftman]

LOL, again exact same hardware, The system that I was using at the time was a KT7A-Raid and a Inno3d card. It was a board with a chipset riddled with bugs (KT133A) and had nothing to do with the OS. That chipset was riddle with bugs and if you knew your Abit history you would also know that the solution to the issue was to increase the AGP driving strength as boards were way out of spec when a more power hungry card was put in due to the bad cap issue ( which by the way Abit at the time would warranty your board even though it may have been expired and send you a new one). Abit later increased the default AGP driving strength to compensate for it with a BIOS update. It had NOTHING to do with SP1. The same exact issues plagued other boards of that era as well, but the KT133a is one of the worst. If you slapped that card in any other chipset with good caps it would have ran fine. The same issues were seen with linux as well.

Ok, I believe you :) The difference is Linux wasn't affected in my case (maybe they implemented some workaround in my distro). And sp1 solved my issue with gf2. I updated bios, but it didn't help with hanging games - I still had to use patch which I found on some forum (the funny thing is I forgot to backup it and I didn't find mentioned patch second time ^^). Yeah, KT133a is a nightmare.


Btw. Which Linux Graphic Driver Bugs Do You Hate?

I hate Catalyst bugs like: problems with video playback when using compiz, black screen bug and others Catalyst related.

[unix_epoch]

What bugs me the most are not the graphics drivers per se, but the audio drivers.

I have to, and far too often, kill X in order to restore the sound. And, logging out and logging in as another user often produce a mute computer.

I have still not understood the pattern but I think the graphics drivers are interlinked to tis issue, in some way. Maybe someone knows better.

I run Debain Sid/Experimental with KDE4 and the latest nvidia drivers on a 8800GT.

Could it be that pulseaudio has control of the sound card, which is preventing normal alsa applications from using it? On cards with hardware mixing (like the SoundBlaster Live-Audigy line), this isn't a problem, but with motherboard audio it is. Try killing pulseaudio, and if that works, disable the sound server in the GNOME settings.

[Jimmy]

What gets me is what people are using to set the bar for drivers.

Compiz: experimental software known to break shit. My shit is broken, NVidia and ATI need to get their act to geather?

KDE4: Known to underperform its predecessor KDE 3.5. KDE 3.5.x works just fine and is lightning fast with the same drivers and hardware. KDE4 does not. Therefore, its NVidia's fault?

Same hardware, same distro, same drivers, same game, same settings (everything turned up), desktop effects disabled:


Prey under KDE 3.5.9
3545 frames rendered in 101.2 seconds = 35.0 fps
3545 frames rendered in 101.2 seconds = 35.0 fps
3545 frames rendered in 101.2 seconds = 35.0 fps

Prey under KDE 4.2.0
3545 frames rendered in 125.3 seconds = 28.3 fps
3545 frames rendered in 122.4 seconds = 29.0 fps
3545 frames rendered in 122.4 seconds = 29.0 fps


Again, Nvidia or KDE4?

I'm sure Nvidia and ATI could do a lot better but lets be fair. They aren't the only ball players on the field.

[unix_epoch]

Same hardware, same distro, same drivers, same game, same settings (everything turned up), desktop effects disabled:


Prey under KDE 3.5.9
3545 frames rendered in 101.2 seconds = 35.0 fps
3545 frames rendered in 101.2 seconds = 35.0 fps
3545 frames rendered in 101.2 seconds = 35.0 fps

Prey under KDE 4.2.0
3545 frames rendered in 125.3 seconds = 28.3 fps
3545 frames rendered in 122.4 seconds = 29.0 fps
3545 frames rendered in 122.4 seconds = 29.0 fps


Again, Nvidia or KDE4?


I'm curious to know what happens if you use nice or chrt to run Prey with realtime priority (i.e. chrt -f 25 prey or nice -n -20 prey).

[kraftman]

Again, Nvidia or KDE4?

I'm sure Nvidia and ATI could do a lot better but lets be fair. They aren't the only ball players on the field.

Nvidia and Ati. QT4 is rather new and they probably don't fully support it yet.

[uaaquarius]

I don't seem to have as many problems with fglrx as most people. The video tearing is annoying but I can live with it until the open driver supports 3D. The one problem that really bugs me is that logging out or trying to restart X in any way sends the card into a spasm. The display freezes, the graphics fan starts going full pelt and the only thing I can do is hit the power button, which at least shuts down the machine cleanly. If it was a hard lockup then the file corruption probably would have made me chuck my card by now.

I had the same problem on my Ubuntu 8.04. Then I've found a corresponding bug:
http://ati.cchtml.com/show_bug.cgi?id=992#c26
Bug is still not resolved, but there is a workaround which fixed my problem.

[Jimmy]

I'm curious to know what happens if you use nice or chrt to run Prey with realtime priority (i.e. chrt -f 25 prey or nice -n -20 prey).

KDE 3 still wins.


KDE 3.5.9, nice -n -20
3545 frames rendered in 100.9 seconds = 35.1 fps
3545 frames rendered in 100.8 seconds = 35.2 fps
3545 frames rendered in 100.9 seconds = 35.1 fps
3545 frames rendered in 100.7 seconds = 35.2 fps

KDE 4.2.0, nice -n 20
3545 frames rendered in 121.6 seconds = 29.2 fps
3545 frames rendered in 120.7 seconds = 29.4 fps
3545 frames rendered in 123.1 seconds = 28.8 fps
3545 frames rendered in 121.8 seconds = 29.1 fps

[Jimmy]

Nvidia and Ati. QT4 is rather new and they probably don't fully support it yet.

Is there a graphics card and driver set where KDE4 preforms better than KDE3? Or are all the hardware vendors just as inadequate next to the infallible QT?

If KDE 4.5 starts preforming better than 4.2 on the same drivers and hardware then what... you'll eat your socks?

[domicius]

It's of little surprise to me then that I run into installation problems with Debian (Pure) based distributions when the maintainer of the packages or scripts is working on a deliberate fork.

Now, maybe the Debian problems could be solved if Ubuntu developers focused on maintaining Debian compatibility.

However, I'm not sure bringing what amounts to be political and personal issues into a discussion involving whether or not an installed driver functions as expected in an installed enviroment.

Oh, you refer to Debian *based* distributions. Because on Debian unstable, I don't recall ever having problems installing fglrx drivers. I usually compile the modules with module-assistant if there's no matching package for my kernel at the moment in unstable repositories.

I really can't comment on the Ubuntu (not) maintaining Debian compatibility other than I'd like this doesn't stay our problem in future...

Everybody, sorry to get little OT...

[sreyan]

The 180.xx drivers were a lot faster with KDE4 than the 177s and earlier versions, but it was still a pretty slow experience trying to use KDE4 with my 8600GT. I'm sure KDE4(and/or its dependencies) is partly to blame, but Nvidia's blob surely is also.


QT 4.5 is much, much faster IMO. Might be interesting to see some benchmarks there, Michael.