The quickest way to own a Azure Deep Learning VM with CUDA GPU support.
Pre-requisites: You will need to have a azure subscription
Firstly, create a NC6 VM with Ubuntu 16.4 LTS, make sure HDD is selected and port 80, 443 and 22 are open. This VM comes with a Tesla K80 GPU.
In my case, I have to select East US before I could find NC6 in the list.
SSH to the VM
Install docker
Don’t use the following command to install docker, as it’s an old version: Docker version 18.09.7, build 2d0083d
#don't run this sudo apt install docker.io
Follow this page to install the latest docker
https://docs.docker.com/install/linux/docker-ce/ubuntu/
sudo apt-get install apt-transport-https ca-certificates curl gnupg-agent software-properties-common sudo apt-key fingerprint 0EBFCD88 sudo apt-get update sudo apt-get install docker-ce docker-ce-cli containerd.io docker -v Docker version 19.03.4, build 9013bf583a
Then Install Nvidia Driver
The info from this page is useful, but don’t build the gpu docker using this docker file.
https://hub.docker.com/r/floydhub/dl-docker
sudo add-apt-repository ppa:graphics-drivers/ppa sudo apt-get update sudo apt-get install nvidia-352
Restart VM with following command
sudo shutdown -r now
check driver
cat /proc/driver/nvidia/version fennng@CUDA:~$ cat /proc/driver/nvidia/version NVRM version: NVIDIA UNIX x86_64 Kernel Module 384.130 Wed Mar 21 03:37:26 PDT 2018 GCC version: gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.11)
install nvidia docker
# Add the package repositories distribution=$(. /etc/os-release;echo $ID$VERSION_ID) curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit sudo systemctl restart docker
Test the enviroment
docker run --gpus all nvidia/cuda:9.0-base nvidia-smi
Don’t do this — Build the docker, this will take a while, about 20 minutes
#don't run this, as the built image doesn't work properly. git clone https://github.com/saiprashanths/dl-docker.git cd dl-docker docker build -t floydhub/dl-docker:gpu -f Dockerfile.gpu .
The reason of the failure as shown below.
So I switch to the following docker image.
https://hub.docker.com/r/jihong/keras-gpu/
sudo docker run --gpus all -it --rm -p 8888:8888 -v /sharefolder:/root/workspace jihong/keras-gpu bash jupyter notebook --allow-root
Now, the jupyter notebook is ready. We need to find a easy way to access it. So I use ngrok here. For proper setup, nginx-proxy with SSL is a better option.
Install ngrok following this instruction
https://dashboard.ngrok.com/get-started
#in my case wget https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip sudo apt install unzip unzip ngrok-stable-linux-amd64.zip mv ngrok /usr/bin/ngrok
Run ngrok
ngrok http 8888
Copy the jupyter notebook token from the linux bash to login
Then navigate to /sharefolder and run
git clone https://github.com/fengnz/deep-learning-with-python-notebooks.git
Then you will be able to see a bunch of notebook to start with
Pickup the first one and have a play.
