CUDA installation

NVIDIA-driver installation

If this is a re-install, do not forget to purge all previous NVIDIA and CUDA instalations. Follow the following instructions to purge:sudo apt-get purge nvidia*sudo apt-get autoremovesudo apt-get autocleansudo rm -rf /usr/local/cuda*

1. Stop X server:

sudo service lightdm stop

2. Uninstall any existing driver:

sudo nvidia-uninstall

3. Disable nouveau: open the config file: sudo vim /etc/modprobe.d/blacklist-nouveau.conf and copy

blacklist nouveau

blacklist lbm-nouveau

options nouveau modset=0

alias nouveau off

alias lbm-nouveau off

4. Update driver:

echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf

sudo update-initramfs -u

sudo reboot

sudo service lightdm stop

5. Download driver

wget http://us.download.nvidia.com/XFree86/Linux-x86_64/390.48/NVIDIA-Linux-x86_64-390.48.run

6. Install driver

chmod +x NVIDIA-Linux-x86_64-390.48.run

sudo ./NVIDIA-Linux-x86_64-390.48.run

sudo reboot

CUDA-9.0 installation for Ubuntu 18.04

The instructions are from this webpage which worked for me

1. Download CUDA 9.0 toolkits (and patches) from here. Run the installer and the patches. Do not re-install NVIDIA drivers

sudo chmod +x cuda_9.0.176_384.81_linux.run

./cuda_9.0.176_384.81_linux.run --override

2. Install CUDNN-7.0 from here. I downloaded 7.1.4 for CUDA 9.0

# Unpack the archive

tar -zxvf cudnn-9.0-linux-x64-v7.tgz

# Move the unpacked contents to your CUDA directory

sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda-9.0/lib64/

sudo cp  cuda/include/cudnn.h /usr/local/cuda-9.0/include/

# Give read access to all users

sudo chmod a+r /usr/local/cuda-9.0/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

3. Install libcutpi

sudo apt-get install libcupti-dev

4. Update the environmental variables

export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

PyTorch installation

Follow the instructions on this website. I installed it locally on a new Anaconda environmental variable which was cloned from base. I use python 2.7, but similar process for python 3 I believe.

pip install torch torchvision

# Check if pytorch is using cuda

In [1]: import torch

In [2]: torch.cuda.is_available()

Out[2]: True

In [3]: a = torch.rand(5,5)

In [4]: a

Out[4]:

tensor([[0.0941, 0.9865, 0.4618, 0.0944, 0.6364],

        [0.3783, 0.9316, 0.9661, 0.0253, 0.6062],

        [0.2219, 0.7945, 0.8306, 0.6733, 0.4095],

        [0.0396, 0.6149, 0.6605, 0.2157, 0.0762],

        [0.0193, 0.2906, 0.5172, 0.6390, 0.7789]])

Page updated

Google Sites

Report abuse