Using the Fedora rpmfusion nvidia driver vs the internal Nvidia driver

Accelerated Computing CUDA CUDA Programming and Performance
1

You can get CUDA functionality in Fedora Linux by either installing the nvidia driver that’s in the rpmfusion repository or by installing the Nvidia closed driver (for example out of: cudadriver_2.3_linux_64_190.18.run). For various administration reasons I’d much prefer to use the rpmfusion one but I wanted to find out what functionality if any would be lost. I think the ‘run’ style driver from Nvidia allows CUDA profiling to work but this isn’t supported with the rpmfusion driver? Is there anything else missing?

Mark

2

Yes, I’d like to know that too. I was unable to install the proprietary nVidia driver for the GTX295 on my Fedora 11 box, but somehow managed to get hold of this package

nvidia-x11-drv-190.18-1.fc11.i586

so as to be able to configure the GTX295. But now, Cuda [post=“0”]doesn’t even compile[/post]. Mark, which distro are you on?

3

At work I’m using Fedora 10 because I knew of the Fedora 11 gcc 4.4 problem with the weak pointers

I’m surprised it’s taking Nvidia this long to support Fedora 11, it’s been out a long time (June 9)

UPDATE: I found this work-around in the forum for the weak pointer situation

http://forums.nvidia.com/index.php?showtop…st&p=587018

and then I just needed to make some links for libXi.so, libXmu.so, libglut.so because Fedora doesn’t have generic links for those but puts a specific

version number in the link. A good number of the SDK example projects will compile but a few fail with:

I think this has something to do with a GL incompatibility with the Mesa stuff on Fedora and the Nvidia GL files that

got put on my machine when I did the run style Nvidia CUDA driver (my machine is pretty polluted now) so I did a

make -k

to continue past those and compile the ones that work.

Makr

4

New CUDA releases tend to come slowly, which means that, at any given time, the last CUDA Toolkit has only been tested with distributions that are at least 6-12 months old.

Usually, the biggest obstacle to forward compatibility of the CUDA Toolkit is a new gcc (as you found). New Linux distribution releases which include a new major release of GCC are far more likely to have compilation problems with CUDA.

5

I resolved my compilation issues with CUDA by removing rpm of the nvidia driver and by installing the official one [1].
Weird…

[1] [url=“http://forums.nvidia.com/index.php?showtopic=108032”]http://forums.nvidia.com/index.php?showtopic=108032[/url]