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 change 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 ...)

Getting my battery to last for longer

1. Open the app "Nvidia X Server Settings"
2. Go to "PRIME profiles" and select the Intel (Power Saving Mode).
3. Then restart or log out your user session and log in again.

This way you'll be using the Intel integrated graphics card and the battery will last way more. To use the Nvidia GPU, just repeat the process above but choose Nvidia (Performance Mode).

In a MSI Gs65 Stealth, with nvidia 1060GTX it lasts about 8h with the Intel GPU enabled. It's a minor issue while not using the nvidia graphics cards.