Tensorflow won't find libcublas.so.9.0 even though LD_LIBRARY_PATH is set

Question

I have installed Cuda Toolkit 9.0 and cuDNN v7.0 according to instrucitons on https://www.tensorflow.org/install/install_linux (and nvidia site) instruction. However, when I try to run some script using tensorflow it says it cannot find it

ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory

even though my LD_LIBRARY_PATH is set correctly: echo $LD_LIBRARY_PATH /usr/local/cuda-9.0/lib64:/usr/local/cuda-9.0/extras/CUPTI/lib64

is there something I'm missing?

I've set the variable only for current session in terminal so resetting it will unset the variable — Julius ShadowAspect Flimmel, Mar 21 '18 at 11:51
Maybe you just need `export LD_LIBRARY_PATH=...` rather than `LD_LIBRARY_PATH=...`? [Only the former will affect subprocesses started by the shell.](https://stackoverflow.com/questions/1158091/defining-a-variable-with-or-without-export) — Allen Lavoie, Mar 22 '18 at 19:31

score 1 · Answer 1 · edited Dec 19 '20 at 22:29

If you've installed cuda with the runfile installation method you have to set also the PATH variable according to the Nvidia documentation. I would recommend adding export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}} and export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}(64-bit) or export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}(32-bit) to your .bashrc (if you use bash).

Tensorflow won't find libcublas.so.9.0 even though LD_LIBRARY_PATH is set

1 Answers1