Setting up a MSI laptop with GPU (gtx1060), Installing Ubuntu 18.04, CUDA, CDNN, Pytorch and TensorFlow

msi-gtx1060-ubuntu-18.04-deeplearning.md

Setting up a MSI laptop with GPU (gtx1060)

Installing Ubuntu 18.04, CUDA, CDNN, Pytorch and TensorFlow

Installing Ubuntu 18.04

Get Image

https://tutorials.ubuntu.com/tutorial/tutorial-create-a-usb-stick-on-ubuntu#0

Install (solving issues)

• Security boot disable (change this in BIOS, for MSI computers reboot system and press delete a couple of times before the OS is loaded)
• Possible problem: Screen freezing while installing
  • Reboot system
  • Go to the Install Ubuntu option (BUT DONT PRESS ENTER)
  • Press e
  • Find the line that starts with linux then add modprobe.blacklist=nouveau after quiet splash.
• Choose a Wifi Network and install third party software and updates
• Once the installation is done, remove the pendrive and reboot system

Installing Nvidia Drivers

• Select Ubuntu at GRUB (you may need to repeat the modprobe.blacklist=nouveau step if the screen freezes again).
• Make sure security boot is disable
• Access command line (ctrl+alt+ F1,F2,F3...F??)

  sudo add-apt-repository ppa:graphics-drivers/ppa
  sudo apt update
  # You can chenge this driver to whatever version suits you best :)
  sudo apt install nvidia-driver-396
  sudo reboot
  

Installing Conda

https://docs.anaconda.com/anaconda/install/linux

Installing CUDA 9.0 and CUDNN 7.0

https://medium.com/@taylordenouden/installing-tensorflow-gpu-on-ubuntu-18-04-89a142325138

Installing Pytorch 0.4

create conda env

conda create -n torch python=3.6

install torch (cuda 9)

conda install -y pytorch=0.4 torchvision cuda90 -c pytorch

if cuda 9 fails, install this way using cuda 8

conda install -y pytorch=0.4 torchvision -c pytorch conda install cuda80 -c soumith

test gpu install

python -c 'import torch; print(torch.rand(2,3).cuda())'

Installing TensorFlow (1.8)

https://www.tensorflow.org/install/install_linux#installing_with_anaconda

REMEMBER

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

test gpu install

python -c 'from tensorflow.python.client import device_lib; device_lib.list_local_devices()'

output:

2018-06-11 11:32:18.954198: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/device:GPU:0 with ...)

Also works with Pytorch 1.0

(torch1) ~ ❯❯❯ conda list pytorch
# packages in environment at /home/nick/anaconda3/envs/torch1:
#
# Name                    Version                   Build  Channel
pytorch                   1.0.0           py3.7_cuda9.0.176_cudnn7.4.1_1    pytorch