FC3: Mesa and 'nvidia' driver

[Kevin J. Cummings]

Alexander Volovics wrote:
> On Tue, Nov 23, 2004 at 01:01:17PM +0000, Colin J Thomson wrote:
> 
> 
>>>>>>Is it still necessary to remove the (xorg-x11) Mesa-libGL and
>>>>>>Mesa-libGLU rpm's before installing the nvidia driver.
> 
> 
>>>>>The NVidia driver package comes with it's own GL libraries. I would
>>>>>physically uninstall the Mesa-libGL RPM.
> 
> 
>>I find this interesting, I have never had to remove *any* libGL rpm to get 
>>nvidia's driver to run on my system rh8/fc1/2 and the driver seems to do a 
>>good job backing up the original libGL files..
> 
> 
> Indeed the nvidia driver install in the past (and presumably now too)
> did(does) a good job of removing/backing up the original libGl files.
> 
> But I asked the question because I seem to remember that about
> a year (or longer ?) ago (FC1/FC2) there were problems with this and 
> advice was given on the lists to first remove the original libGl files first.

The technical answer here is that the nVidia GL libraries are put in 
/usr/lib, and the Mesa-libGL library is put in /usr/X11R6/lib (like Neal 
eluded to, though its *not* supposed to be a matter or luck!).  The 
trick is to ensure that your ld.so.conf file looks in /usr/lib before it 
looks in /usr/X11R6/lib.  You can tell by looking at the ldd output of 
some GL X11 program (like glxgears).  If it links to the nVidia library 
in /usr/lib, you've done it right.  If not, you may need to re-order the 
libraries in your /etc/ld.so.conf file.

> I can't remember what caused the problem but I can remember a lot
> of discussion about this and Mike Harris was discussing packaging
> the Fedora file so that the problem wouldn't occur anymore.

If the libary order is correct, then all that *should* be needed is to 
run ldconfig after each installation of the nVidia modules.  The 
packaging problem is complicated by the fact that xorg (and XFree86) and 
nVidia put their GL libraries in different places.  It would help if 
both libraries provided the same "capability" and could be used to 
"upgrade" one another.  Then, we wouldn't be trying to re-install 
Mesa-libGL after the nVidia modules are installed in order to satisfy 
dependancies that nVidia provides, but since the nVidia modules aren't 
RPMs (at least the installation from nVidia isn't), we have this problem.

I know that on my measley MX-400 video card, I see about 46-48 fps when 
my libraries are mis-configured, and about 92-96 fps when I "clean 
things up".  That's nowheres near the 300-400 fps I think I should be 
getting with this card at 4x AGP?

> I can't remember what happened after that and I stopped using the
> nvidia driver because the nv driver was good enough for my purposes.
> 
> (But now I have 'ugly' font rendering with the nv driver on an
>  LCD DVI flatpanel and I would like to see if the nvidia driver does
>  any better).

Is the font rendering the driver's fault, or the font server's 
configuration's fault?  My font rendering didn't change when I switched 
from the "nv" driver to the "nvidia" driver.

-- 
Kevin J. Cummings
kjchome at rcn.com
cummings at kjchome.homeip.net
cummings at kjc386.framingham.ma.us