The following tutorial will teach you how to install the nvidia driver, docker, docker-compose and also nvidia docker plugin so you can easily use your dockerized projects on the cloud.
The GPU instances are billed even if stopped. Therefore there is no reason to stop the instances. Instead, it would be wise to shutdown the instance and keep the boot volume. The following chart afaik does not include the price for the CPU, but that's negligible.
| Instance | GPU | Price/hr |
|---|---|---|
| VM.GPU2.1 | 1x NVIDIA P100 | US$1.275 |
| BM.GPU2.2 | 2x NVIDIA P100 | US$2.550 |
If you have allocated boot volume, the pricing is around US$0.03/GB/month. So for 256GB drive it should be around 8 USD.
The console can be accessed at https://console.eu-frankfurt-1.oraclecloud.com/. Insert recombee into Cloud tenant field. On the next page, under Single Sign-On, click continue and sign in into the Oracle Cloud Console using the provided credentials.
Go to Compute -> Instances and click the blue Create instance button.
- Click the
Change byte copybutton and selectCanonical Ubuntu 18.04 - Under
ConfigurationclickChange configurationbutton.
If you have created an instance before and you have not deleted your statring volume, you can select it under Boot volumes.
Now we have to select which hardware is going to be used on our instance. There are 2 options, one with 1xP100 GPU and one with 2xP100 GPUs. There are also 3 availability domains. If you can't allocate a resource in one of them, try selecting a different one.
- 1xP100: Under
Virtual computer->Specializations and older versionsclickVM.GPU2.1. This instance has 12 cores and 72GB of RAM - 2xP100: Under
Bare metalclickBM.GPU2.2. This instance has 28 cores and 256GB of RAM.
A properly configured instance should look something like this. Beneath these settings, you can also select custom size of the starting volume. If you want to make a persistent volume for yourself, it's the best to create a larger than default volume.
Last thing that needs to be done is to add your ssh keys. Scroll down under Add SSH keys and select select public key radio button. Now select your public key (e.g. ~/.ssh/id_rsa.pub).
Now click create and your instance will be started!
By default, every instance gets a public IP. If you head to Compute -> Instances you can see the current state of the created instances and their respective IP's. To connect to an instance with ubuntu and your public key, use the following command:
ssh -i ~/.ssh/id_rsa -L 43603:127.0.0.1:43603 -L 43604:127.0.0.1:43604 ubuntu@130.61.254.173The -i flag uses the designed private key, and the -L flag defines the mapping of forwarded ports to your machine. In this case we use 43603 for a jupyter notebook and 43604 for Tensorboard.
In this step, we will install the NVIDIA drivers, docker, docker-compose and also the docker-compose nvidia plugin. The setup is described in the setup.sh file. You can either copy-paste the data, or clone this directory and run the setup.
git clone https://github.com/mskl/docker-deep-learning; sudo ./docker-deep-learning/setup.shFurther tutorial is described in the README.md If you want to start using the service with this docker-deep-learning repo, follow with setup the variables in .env and clone your project there. To build and run the image type
sudo make runWe are paying for the time that instance existst rerardless that it's stopped or running. Navigate to Compute -> Instances click the three dots on the right side of instance. Select terminate and if needed, select the checkbox to delete the created disc. Be carefull if you want to persist your disc!!!
