Can't use nvidia GPU optimization (libcublas.so)

Question

I have a Nvidia 750Ti on a dedicated server

I'm trying to install cuda in order to use GPU optimization in Theano.

I did several cuda 7 installations (on Ubuntu server x86_64 12.04, Ubuntu server x86_64 14.04, installing Nvidia drivers manually...).

The current installation is: - Ubuntu 14.04 (updated) - cuda 6.5 (via .run)

After each installation I try the first GPU example in this tutorial but I keep getting the same response:

THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python check1.py
ERROR (theano.sandbox.cuda): Failed to compile cuda_ndarray.cu: libcublas.so.6.5: cannot open shared object file: No such file or directory
WARNING (theano.sandbox.cuda): CUDA is installed, but device gpu is not available  (error: cuda unavilable)
[Elemwise{exp,no_inplace}(<TensorType(float32, vector)>)]
Looping 1000 times took 1.9148170948 seconds
Result is [ 1.23178029  1.61879337  1.52278066 ...,  2.20771813  2.29967761
  1.62323284]
Used the cpu


$ ls -la /usr/local/cuda-6.5/lib64/libcublas.so
lrwxrwxrwx 1 root root 16 Jun 14 18:23 /usr/local/cuda-6.5/lib64/libcublas.so -> libcublas.so.6.5

I compiled the nvidia cuda examples. I can execute the matrixMul example. But executing matrixMulCUBLAS gives me a similar error related to libcublas.so shared object

Also, when I execute NVIDIA X Server Settings I receive:

You do not appear to be using the NVIDIA X driver. Please edit your X configuration file (just run nvidia-xconfig as root), and restart X

I execute:

root> nvidia-xconfig 

And get the following message in return:

WARNING: Unable to locate/open X configuration file.
New X configuration file written to '/etc/X11/xorg.conf'

But in spite of nvidia-xconfig, the problem in NVIDIA X Server Settings still remains.

In one of the intents, I solved this problem with NVIDIA X Server Settings. But it doesn't solved the problem with libcublas.

I have set related variables via /etc/environment

CUDA_HOME=/usr/local/cuda-6.5
LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib64
PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/opt/anaconda/bin:/usr/local/cuda-6.5/bin"

I'm a newbie in GPUs and its like a big mess!

Answer 1

I solved the same problem by putting the following two lines in the file $HOME/.bashrc

export PATH=/usr/local/cuda/bin:/usr/local/cuda/lib64:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

This will set some environment variables in order to run and write CUDA enabled programs.

Answer 2

Solved.

• The server had two graphic cards. One GPU card, and one on board card. I asked the hosting company to turn off the on board card

• I was worried about NVIDIA X Server Settings, when I was connecting via VNC. nvidia-smi informed me that all was ok.

• I needed to set LD_LIBRARY_PATH using ldconfig, not in /etc/environment, as explained here.

  sudo echo "/usr/local/cuda-7.0/lib64" > /etc/ld.so.conf.d/cuda.conf
  sudo ldconfig
  

Thanks everybody!