CUDA and Fedora 15 Can get dev drivers installed

I’ve been trying for a few days now to get the dev drivers installed for Fedora 15 with no luck. Can some one help?

nvidia-installer log file '/var/log/nvidia-installer.log'

creation time: Tue Aug 16 14:08:40 2011

installer version: 270.41.19

PATH:

/opt/gcc44/bin:/usr/lib64/ccache:/usr/local/bin:/bin:/usr/bin:/usr/local/sbin:/

usr/sbin:/sbin:/home/goddard/bin

option status:

  license pre-accepted               : false

  update                             : false

  force update                       : false

  expert                             : false

  uninstall                          : false

  driver info                        : false

  precompiled interfaces             : true

  no ncurses color                   : false

  query latest version               : false

  no questions                       : false

  silent                             : false

  no recursion                       : false

  no backup                          : false

  kernel module only                 : false

  sanity                             : false

  add this kernel                    : false

  no runlevel check                  : false

  no network                         : false

  no ABI note                        : false

  no RPMs                            : false

  no kernel module                   : false

  force SELinux                      : default

  no X server check                  : false

  no cc version check                : false

  run distro scripts                 : true

  no nouveau check                   : false

  run nvidia-xconfig                 : false

  sigwinch work around               : true

  force tls                          : (not specified)

  force compat32 tls                 : (not specified)

  X install prefix                   : (not specified)

  X library install path             : (not specified)

  X module install path              : (not specified)

  OpenGL install prefix              : (not specified)

  OpenGL install libdir              : (not specified)

  compat32 install chroot            : (not specified)

  compat32 install prefix            : (not specified)

  compat32 install libdir            : (not specified)

  utility install prefix             : (not specified)

  utility install libdir             : (not specified)

  installer prefix                   : (not specified)

  doc install prefix                 : (not specified)

  kernel name                        : (not specified)

  kernel include path                : (not specified)

  kernel source path                 : (not specified)

  kernel output path                 : (not specified)

  kernel install path                : (not specified)

  precompiled kernel interfaces path : (not specified)

  precompiled kernel interfaces url  : (not specified)

  proc mount point                   : /proc

  ui                                 : (not specified)

  tmpdir                             : /tmp

  ftp mirror                         : ftp://download.nvidia.com

  RPM file list                      : (not specified)

  selinux chcon type                 : (not specified)

Using: nvidia-installer ncurses user interface

-> License accepted.

-> Installing NVIDIA driver version 270.41.19.

-> Performing CC sanity check with CC="cc".

-> Performing CC version check with CC="cc".

-> Kernel source path: '/lib/modules/2.6.40-4.fc15.x86_64/source'

-> Kernel output path: '/lib/modules/2.6.40-4.fc15.x86_64/build'

ERROR: If you are using a Linux 2.4 kernel, please make sure

       you either have configured kernel sources matching your

       kernel or the correct set of kernel headers installed

       on your system.

If you are using a Linux 2.6 kernel, please make sure

       you have configured kernel sources matching your kernel

       installed on your system. If you specified a separate

       output directory using either the "KBUILD_OUTPUT" or

       the "O" KBUILD parameter, make sure to specify this

       directory with the SYSOUT environment variable or with

       the equivalent nvidia-installer command line option.

Depending on where and how the kernel sources (or the

       kernel headers) were installed, you may need to specify

       their location with the SYSSRC environment variable or

       the equivalent nvidia-installer command line option.

ERROR: Installation has failed.  Please see the file

       '/var/log/nvidia-installer.log' for details.  You may find suggestions

       on fixing installation problems in the README available on the Linux

       driver download page at www.nvidia.com.

Hi,

I gave up trying to use the nvidia CUDA driver installer and switched to the one supplied via yum (akmod-nvidia + xorg-x11-drv-nvidia-libs packages) as discussed at if !1 0 (Fedora 14 nVidia Drivers Installation instructions). Doing this installs the 280.13 driver which does support CUDA (at least I haven’t had any issues so far with the SDK examples). You will need to modify the common.mk files within the SDK to include /usr/lib64/nvidia as an additional directory to link against.

Given that Fedora 16(!) is slated to be released in ~2 months and to currently use CUDA with the latest Fedora distros requires one to compile and install a separate gcc4.4 compiler, it is an open question in my mind if nvidia is going to continue to support Fedora.

Does anyone at nvidia wish to comment on this?

dpe

Oh wow thanks. I asked this question like a hundred times over there with no answer. From now on I will just come to the Nvidia forums.

You’re welcome!

How can I confirm Cuda is working with the 280 driver? Because I got things to compile, but binaries are reporting it doesn’t work.

Any ideas on how I can confirm?

Hmmm … I don’t have any problems running deviceQuery, etc. Did you reboot your system and ensure that /etc/X11/xorg.conf is configured properly?

Performing ldd ./deviceQuery results in

ldd ./deviceQuery

	linux-vdso.so.1 =>  (0x00007fffaa3a7000)

	libcudart.so.4 => /usr/local/cuda/lib64/libcudart.so.4 (0x00007fd99a223000)

	libcuda.so.1 => /usr/lib64/nvidia/libcuda.so.1 (0x00007fd99987a000)

	libstdc++.so.6 => /usr/lib64/libstdc++.so.6 (0x0000003efe000000)

	libm.so.6 => /lib64/libm.so.6 (0x0000003ef4800000)

	libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x0000003ef5400000)

	libc.so.6 => /lib64/libc.so.6 (0x0000003ef4000000)

	libpthread.so.0 => /lib64/libpthread.so.0 (0x0000003ef4400000)

	libdl.so.2 => /lib64/libdl.so.2 (0x0000003ef4c00000)

	librt.so.1 => /lib64/librt.so.1 (0x0000003ef5000000)

	libz.so.1 => /lib64/libz.so.1 (0x0000003ef5800000)

	/lib64/ld-linux-x86-64.so.2 (0x0000003ef3c00000)

so the cuda library in /usr/lib64/nvidia is being used. libcuda.so.1 is a symbolic link to libcuda.so.280.13.

And /proc/driver/nvidia/version reports

NVRM version: NVIDIA UNIX x86_64 Kernel Module  280.13  Wed Jul 27 16:53:56 PDT 2011

GCC version:  gcc version 4.6.0 20110603 (Red Hat 4.6.0-10) (GCC)

so the 280.13 driver is being used.

I would check that the compiled binaries are using the libraries in /usr/lib64/nvidia and that /proc/driver/nvidia/version provides identical output to that listed above.

dpe

Thanks. I guess I will look into this error else where then.

Given that you reported that you were able to successfully compile the SDK examples (I’m assuming your gcc44 installation is correct) and we know that the SDK examples are not buggy, that only leaves the runtime system (libraries and drivers) which could be introducing the error for non-graphical SDK examples (such as deviceQuery). I always run deviceQuery to confirm correct CUDA installation. I also run additional non-graphical and graphical examples as well.

You also check if the nvidia driver is actually loaded into the kernel with lsmod | grep nvidia. Running this on my system results in

[root@jemez ~]# lsmod | grep nvidia

nvidia              11693958  32 

i2c_core               25712  2 i2c_i801,nvidia

There is also the updaed Fedora 15 nVidia driver installation guide at if !1 0.

I have the driver installed, but I’m trying to compile OpenCV with TBB and I’m getting an error. I’m guessing it was a problem with TBB, but not sure. I force compile and it claims I’m not using CUDA.

I read some where that the toolkit and the driver need to match, but I’m new to CUDA.

Hmmm … you’re out of my league with regards to CUDA integration into OpenCV. I don’t have a lot of experience with OpenCV.

If OpenCV is relying on libnvidia-tls (local thread storage libraries) for nvidia native OpenGL support then the location of these libraries has changed to /usr/lib64/nvidia/tls. See nvidia driver installed components writeup.

I do not think that you need to have toolkit and driver match with CUDA (at least for support for older features) Every driver should be backwards compatible with older toolkits. Just don’t expect to be able to run, say concurrent kernels, on a device when using an older toolkit since they may have needed to modify the toolkit to deal with this.

dpe

Nvidia is now supporting Fedora 14 with their 4.1RC2 release, but given that Fedora is now on 16 and working on 17, is nvidia always going to be two releases behind in their Fedora support? Does anyone at nvidia wish to comment on what their Fedora support policy is?

dpe

I don’t work for NVIDIA, but when this has come up in the past (typically regarding Ubuntu), the response has been that the QA process for CUDA takes long enough that they will probably not support a distribution release that happens after the middle the CUDA release cycle. Given that a CUDA release happens every 6-9 months, and Fedora and Ubuntu both use 6 month release cycles, CUDA is pretty much guaranteed to be one distribution release out of date. While this has not been stated as explicit policy, this has been the reality for several years of CUDA releases.

So the real question is why Fedora 15 is not (yet) supported in CUDA 4.1, as that was released only a few months after CUDA 4.0.

Hi seibert,

Yes, I agree that differing release cycles will cause CUDA to get out of sync with Fedora (and other open source) releases. I do think that it would be helpful to Fedora and Ubuntu users if nvidia would sync their CUDA release cycles to the Fedora/Ubuntu release cycles, but this is probably asking too much.

I also agree with your question asking why Fedora 15 is not yet supported by CUDA? It would be helpful to have someone at nvidia weigh in on this topic. I do believe that six month release cycles encourages open source software innovation as discussed on the Fedora Release Life Cycle page.

dpe