2

I've followed the NVIDIA guide to installing CUDA on Ubuntu, using the .deb file and then calling sudo aptitude install cuda.

dpkg --list | grep nvidia now shows :

ii  nvidia-367                                 367.57-0ubuntu0.16.04.1                                     amd64        NVIDIA binary driver - version 367.57
ii  nvidia-367-dev                             367.57-0ubuntu0.16.04.1                                     amd64        NVIDIA binary Xorg driver development files
ii  nvidia-modprobe                            367.48-0ubuntu1                                             amd64        Load the NVIDIA kernel driver and create device files
ii  nvidia-opencl-icd-367                      367.57-0ubuntu0.16.04.1                                     amd64        NVIDIA OpenCL ICD
rc  nvidia-prime                               0.8.2                                                       amd64        Tools to enable NVIDIA's Prime
ii  nvidia-settings                            367.48-0ubuntu1                                             amd64        Tool for configuring the NVIDIA graphics driver

As you can see, it looks like I have 367.57 and 367.48 running concurrently. As a result, I get errors like

modprobe: ERROR: could not insert 'nvidia_367_uvm': Unknown symbol in module, or unknown parameter (see dmesg)

I can't figure out how to fix this issue. Any help would be much appreciated !

Quentin
  • 31
  • 1
    This is a Ubuntu package management issue, not a programming question. You world be better served asking it somewhere else. –  Nov 26 '16 at 08:06
  • The nvidia guide doesn't have sudo aptitude install cuda anywhere in it. It does however explain how to remove old installations. You might try doing that and try again. –  Nov 26 '16 at 14:25
  • show me where aptitude appears in that document. Section 2.7 of that document is pretty clear on how to remove old installations. And as already pointed out, this question is off-topic for SO. –  Nov 26 '16 at 17:54
  • I used aptitude instead of apt-get, which is generally not an issue. As for being off-topic, I've asked it elsewhere, as was recommended. I was merely answering your comment. – Quentin Nov 26 '16 at 18:05
  • @RobertCrovella : http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#ubuntu-installation

    It isn't very clear on removing old installations. That being said, this isn't an old installation - this is a fresh Ubuntu install. I just followed the directions in this guide, using the package manager version, as the generic installer ( .run ) kept failing to disable Nouveau drivers, and when I tried manually, I got no display whatsoever and had to reinstall the OS again.

     – Quentin Nov 26 '16 at 17:52
  • Just install drivers and Cuda with .run option. Download drivers from Nvidia page and run it in command line, and do the same for Cuda, choose, .run file installation. I can confirm this setup, at 2016.12.26 I am running Ubuntu 16.04, Cuda 8.0.44, NVIDIA x86_64-375.26 drivers With Dual GTX 980Ti GPUs, CuDNN 5105, CNMeM. Gcc version 5.4.0, Using it with Keras on both backends TensorFlow and Theano. – Edmhs Dec 29 '16 at 08:10

1 Answers1

0

I am running a setup identical with yours. I think you misnamed the module, it should be nvidia_uvm. I used the Ubuntu supplied Nvidia drivers (same versions as yours, and have an identical mix of 367.48 and 57 versions.

ubfan1
  • 17,838