Use Amazon's cloud server EC2 to do deep learning (i) apply for a spot instance

This is the first article in the series "Using Amazon's cloud server EC2 to do deep learning".

(i) Application for spot instances (ii) configuration Jupyter notebook Server (iii) configuration TensorFlow

It is well known that deep learning has high demands on computers, and a deep learning machine that configures a thousands of-dollar GPU, 8GB of memory, HDD hard drives is expensive. However, you do not need to specifically configure a computer to do deep learning.

Amazon cloud Services is widely used to deploy Web services on AWS, and most people do not know that AWS also has a server with a GPU. The low-edition server-hugging (g2.2xlarge) has 8-core CPU,15GB memory, 2 nvidia GRID GPUs (Kepler GK104), a high-edition server (G2.8xlarge) with 32-core CPU,60GB of memory, 4 nvidia GRID GPU (Kepler GK104).

In addition to the very powerful configuration, the price of the server is also very pro-people. Amazon offers two ways of pricing, common instances and spot request instances. Spot request instance bidding to lease the server, the price according to the server usage in a certain interval floating, usually the price is much lower than the normal instance. Servers with low edition are usually between $0.1 and $0.3 per hour.

One prerequisite for using AWS is that you have a credit card that supports US dollar settlement. Fees are automatically deducted every month.

The specific process of operation is as follows:

(1) Enter the EC2 interface

(2) Select a spot instance

(3) Select the instance type and operating system type

Use Amazon's cloud server EC2 to do deep learning (i) apply for a spot instance