CUDA 10.0: cuInit Error [Ubuntu 14.04][GeForce GTX 1080]

Hi All,

I recently installed Cuda 10.0, as in the instructions,
when trying to run tensorflow 1.13.1 I get cuInit Error: CUDA_ERROR_UNKNOWN

after some googling and looking through forums, I didn’t find a solution yet.

nvcc --version produces correct version:

nvcc: NVIDIA ® Cuda compiler driver
Copyright © 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130

nvidia-smi shows Driver Version: 410.48 which is correct too, right?

my bashrc is updated with:
export PATH="/usr/local/cuda-10.0/bin:$PATH"
export LD_LIBRARY_PATH="/usr/local/cuda-10.0/lib64:$LD_LIBRARY_PATH"

suggested solutions I found in some forums:

  1. reboot - done that, didn’t change anything

  2. sudo apt-get install nvidia-modprobe
    but it returns:

    nvidia-modprobe is already the newest version.
    0 upgraded, 0 newly installed, 0 to remove and 251 not upgraded.

  3. only thing that bothers me is:
    for: find /lib/modules/ | grep -i nvidia
    I get:

    /lib/modules/4.4.0-31-generic/kernel/drivers/video/fbdev/nvidia
    /lib/modules/4.4.0-31-generic/kernel/drivers/video/fbdev/nvidia/nvidiafb.ko
    /lib/modules/4.4.0-31-generic/kernel/drivers/net/ethernet/nvidia
    /lib/modules/4.4.0-31-generic/kernel/drivers/net/ethernet/nvidia/forcedeth.ko
    /lib/modules/4.4.0-72-generic/kernel/drivers/video/fbdev/nvidia
    /lib/modules/4.4.0-72-generic/kernel/drivers/video/fbdev/nvidia/nvidiafb.ko
    /lib/modules/4.4.0-72-generic/kernel/drivers/net/ethernet/nvidia
    /lib/modules/4.4.0-72-generic/kernel/drivers/net/ethernet/nvidia/forcedeth.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia_384_modeset.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia_384_drm.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia-drm.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia_384.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia-uvm.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia_384_uvm.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia.ko
    /lib/modules/4.4.0-72-generic/updates/dkms/nvidia-modeset.ko

  • nvidia_384 refers to the old driver? is this supposed to be nvidia_410?
  • or having two modules versions: 4.4.0-31-generic and 4.4.0-72-generic is the problem?

Thanks a lot.
will supply any other information if needed.

with no other answer, I ended up removing CUDA and reinstalling, this time not with the runfile, but with the package [.deb].

so issue is resolved.