0

I updated tensorflow to version 1.8 when tensorflow-gpu was already installed.

but this exception has occur.

ImportError: libcublas.so.8.0: cannot open shared object file: No such file or directory

And I imported according to tensorflow's guidelines(42013316).

But in my case it was not solved.

I would appreciate it if you could give a resolution.

environment:

ubuntu -16.04.01

gpu-nvidia 1080ti *4

tensorflow-1.8

tensorflow-gpu-1.8

python-2.7

talonmies
  • 70,661
  • 34
  • 192
  • 269
송준석
  • 991
  • 1
  • 16
  • 32
  • have you installed CUDA and CUDNN?that shared file is for CUDA to be able to run on GPU – Eliethesaiyan Jun 25 '18 at 05:38
  • i installed cuda-9.0 , also installed cudnn. – 송준석 Jun 25 '18 at 05:48
  • is CUDA_HOME and LD_LIBRARY_PATH set in your bash_profile file? – Eliethesaiyan Jun 25 '18 at 05:51
  • i export this lines: export PATH=/usr/local/cuda-9.0/bin:$PATH export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH export CUDA_HOME=/usr/local/cuda – 송준석 Jun 25 '18 at 06:05
  • did you make symbolic link from that point to /usr/local/cuda-9.0 to /usr/local/cuda? it seems you re point to cuda-9 in path and to cuda in CUDA_HOME Path – Eliethesaiyan Jun 25 '18 at 06:09
  • yes, cuda-9.0's sybolic is in there. – 송준석 Jun 25 '18 at 06:11
  • 1
    not sure if it was not resolved by now...but it seems that cuda-9 wasn't supported in tensorflow few months ago.. you may consider to downgrade to 1.4 tensorflow and see how it goes ...here is a useful link about same problem...good luck https://github.com/tensorflow/tensorflow/issues/14622\ – Eliethesaiyan Jun 25 '18 at 06:21
  • Are there any issues about the simultaneous installation of tensorflow and tensorflow-gpu that you know? – 송준석 Jun 25 '18 at 06:49
  • I think you can't have both installed, the last installation will remove the previous one if you upgrade via pip, but you could have a native one and another in virtual environment – Eliethesaiyan Jun 25 '18 at 06:51
  • As long as they're not installed in the same python environment, they'll do just fine together on the same machine. Your error looks like you have the wrong version of CUDA/cuDNN installed – GPhilo Jun 25 '18 at 08:40
  • Should I lower the version of cuda? Or is there a version of cuda for tensorflow1.8? – 송준석 Jun 25 '18 at 08:44

0 Answers0