Re: MSI RS482M4-ILD (Radeon Xpress 200 chipset)

Vladimir Florinski <vflorins@xxxxxxx> wrote:
That machine often gives some strange segfaults when trying to run KDE.
This happens both on the local console and when logging in with xdmcp.

It must be a KDE bug (and you shouldn't be using xdm anyway).

I have 30+ machines where KDE works without any problems at all. This is
the only machine with those segfaults. My guess is that the driver, or
rather the nvidia libraries are involved in the bug as that is the only
thing that makes this machine differ from others.

I am aware that xdm logins send clear text passwords. However, this is on
a secure net with trusted users and the network doesn't even have a
connection to internet.

In my department, where I set hardware policies, use of opensource video
drivers is strongly discouraged (with the exception of laptops, where
Nvidia doesn't have a large presence).

Does this policy apply also to drivers for other peripherals like SCSI
cards and network cards?

Opensource drivers are amateurish, of poor quality and performance,
crash often, break on updates, and it's hard to get bugs fixed or
receive help from developers.

If that is your experience it might explain your policy, but I have the
opposite experience.

Nvidia drivers are written by professionals, virtually never crash (and
when there is a problem, it is easy to report to NVidia and the fix is
usually available in the next update), and have excellent performance.

I agree that the drivers have very good performance. However, I find it
much easier to report or even fix bugs myself in an opensource driver.
With access to the driver source it is possible to add trace printouts to
find out where the driver hangs. How useful is a bugreport which only says:

"A machine with distro X and kernel Y with your driver version Z hanged
last night. After reboot there were no traces in the logs, probably
because the file system wasn't flushed at the hang. The screen was blank
when I came, the hang might have been caused by an OpenGL screen saver".

I have never even bothered to send such a bug report.

Being a developer myself, I know that almost every developer wants to
write perfect software. However, working as a professional developer means
that you have some kind of boss which means that your time means money and
his task is to stop your work when the software is good enough. Fixing the
last rare bugs costs too much money.

That has been my experience with dozens of systems. The only ones that
crashed had ATI, Matrox, and Intel cards running opensource drivers.

At work I have by default the environment variable LIBGL_ALWAYS_INDIRECT
set for all users. This disables DRI. The reason for this is that without
that I had at least one machine with a Matrox card crashing every month,
mostly because of some randomly selected screen saver, but it could also
be Matlab or some custom application. Users which need DRI are able to
temporary unset the environment variable for their application.

So I share your experience that the Matrox opensource driver isn't really
stable. However, in my experience also the nVidia binary driver crashes
about as often as the Matrox cards. At home I have a low end ATI card. I
don't have that environment variable set as I feel no need for it. I have
seen maybe one or two crashes with this card in about two years, but thats
OK for a home system. Current uptime on my home system is more than 100
days.

You would not be able to debug it even if you had the source (and nobody
would except Nvidia engineers). The driver is more complex than the
kernel and there is simply no talent or experience within the opensource
community to understand the code (much less develop an equivalent
driver).

By adding some trace printouts it would be rather easy to find where it
hangs. However, once the location of the crash is found it might not be as
easy to produce a fix, but at least it would be possible to produce an
informative bug report.

The binary drivers are good enough for a home gaming system. However,
in my experience they are not good enough for a server which is
supposed to be up 24/7 with local and remote users logged in from the
network. It is not OK if such a machine hangs once a month.

I am not talking about servers here. These can be run without video
cards at all.

This depends upon what kind of server you mean. Servers for web, mail and
NFS could be run without video card. However, we have some powerful
machines which are used both as login servers for users on network to run
simulations and also local users which need both CPU and GPU performance.

Thanks for your input. If it turns out that the Xpress 200 based board
from MSI is a bad choice I am now considering an AOpen i945GTm-VHL.
However, that motherboard costs about 3 times as much as the MSI board.
It also requires more expensive SO-DIMM and intel core (duo) CPU. Such
a system will probably end up at least twice as expensive and with less
performance than an Athlon 3000+ based system.

Well, if opensource is your religion, you should expect to pay more. I
do have the Intel 945 chipset in my laptop, and its 3D performance is
abysmal with the i810 driver. Actually, its software rendering speed
(tested with vesafb) was better than hardware rendering!

That was one interesting test. I also appreciate your advice on the
K8NGM2-FID, however, as you understand by now I prefer another chipset
than nVidia for graphics. If I would be building a machine with a high end
card with maximum graphics performance nothing would performace-wise be
able to match a high-end nVidia card like 7900 with binary drivers.
However, as I am now building a low-end machine I am able to choose among
cards with opensource drivers and with that option I prefer the opensource
alternative. If someone with own experiences of the RS482M4-ILD shared
them my decision would be easy. The RS482M4-ILD is even a little more
cheaper than the K8NGM2-FID, but I don't want to buy it unless I know it
works OK.

regards Henrik
--
The address in the header is only to prevent spam. My real address is:
hc8(at)uthyres.com Examples of addresses which go to spammers:
root@xxxxxxxxxxxxx root@localhost

.

Relevant Pages

  • Re: [kde-linux] To NV or not to NV?
    ... capabilies or an old Nvidia TNT marsupial using the nv driver ... As it happens, I'm looking at video cards ATM, and the AMD/ATI Radeon ... current support, but the r600 and r700 based driver development is coming ...
    (KDE)
  • Fwd: [Bug 469470] Re: proprietary nvidia driver does not work under karmic
    ... I would appreciate it very much if those Karmic and Nvidia proprietary ... driver usershaving X issues of any ... kind would visit this bug report and add any appropriate comments. ... of Hal and DKMS according to error messages and logs. ...
    (Ubuntu)
  • Re: How to remove a service from the system?
    ... Your mistake, in my opinion, was that you should have swapped cards before ... > Perhaps the Nvidia uninstall just failed to remove the Driver Helper ... > Would it be safe to just delete the Driver Helper Service and the DLLS ... Yet the Driver Helper Service remains in the Services ...
    (microsoft.public.windowsxp.general)
  • Re: Help Setting up Dual Monitors
    ... using an NVIDIA Quadro FX video card please? ... One monitor is fine but ... I don't think the nv driver that comes with SUSE supports dual monitors. ... Radeon 8500 but Nvidia cards are the best supported PCI-E cards ...
    (alt.os.linux.suse)
  • Re: Graphics card for laptop: nVidia, ATI or Intel?
    ... nVidia GeForce Go 7300/7600 ... ATI Mobility Radeon X1300/X1400/X1600 ... anything about the advantages of this type of graphics cards. ... The nv driver should support the G70 cards, ...
    (alt.os.linux)