toggle

Installing and Configuring GPU on AWS EC2 Instance

Setting up a GPU on an Amazon EC2 instance can significantly enhance your computational power, especially for tasks involving machine learning, data analysis, and graphical processing. This interactive guide will walk you through the process of launching an EC2 instance with a GPU, installing the necessary drivers, and configuring the instance for optimal performance.

author-logo

Apoorva

Aug 20, 2024 |

8 mins

Installing and Configuring GPU on AWS EC2 Instance

Watch the video instead if you want quick instructions and resolution.

Step 1 : Increase the quota of EC2 Instance for GPU

1. Open the AWS Management Console and log in with your credentials.

aws management console

2. Before we launch an EC2 instance, we need to increase the quota of whichever EC2 instance we aim to create. For that, we search and go to “Service Quotas”.

EC2 instance

3. In the Service Quotas Dashboard, click on “Amazon Elastic Compute Cloud(Amazon EC2)”.

Amazon Elastic Compute Cloud

4. Search for “Running Dedicated p2 Hosts”.

Running Dedicated p2 Hosts

5. Since we are going to run “p2.xlarge” instance , click on “Request quota increase”.

p2.xlarge

6. A “p2.xlarge” needs at least 4 CPUs to work. So, we’ll change the quota value to 4 and then press “Request”. This might take a day or two, so make sure you do it early. 

request quota

Step 2 : Launch an EC2 Instance

1. In the console, search for "EC2" and click on it to open the EC2 Dashboard.

EC2 Dashboard

2. Click on the “Launch Instance” button

Launch Instance

3. Provide a name to your server, for this blog, we’re keeping it instance-example. 

instance-example

4. For the operating system, we’ll choose Ubuntu. We’ll choose the instance type as “p2xlarge” for this blog, since it’s the cheapest instance with GPU. If you prefer other instance types with GPU, feel free to go ahead with them. 

Ubuntu

p2xlarge

5. We’ll choose a key pair.Create one if you don’t have it already. 

key pair

6. We’ll increase the storage a bit, since GPU heavy programs will be of large sizes usually. 

GPU heavy programs

7. And let’s launch the instance now. We will have to wait for the instance to launch and then let’s connect to it. 

instance to launch

launch the instance

8. Now, go back to the instances page and connect the instance-example instance as shown.

instance-example instance

Step 3 : Install NVIDIA Drivers

1. Now, we’ll update the packages currently installed in the system by upgrading to their latest versions by typing sudo apt -y update. 

sudo apt -y update

2. Next we need to find a compatible driver for the GPU. For this, we need to follow this link and open this website : 

compatible driver for the GPU

3. We expanded “Option 2 : Public NVIDIA drivers” to find the specifications we needed. 

Public NVIDIA drivers

4. After this, we open this link : 

to open the nvidia driver downloads page. Our instance is P2, and our product type is Tesla, product series is K series, and the product is K80. Choose the operating system as Linux 64-bit and search.

Linux 64-bit

5. Now we get the version of the driver we need. Copy or make a note of this version.

get the version of the driver

6. Open the instance terminal, and type : “sudo apt install nvidia-driver-460 nvidia-dkms-460”, 460 being the driver version that we got. We let it finish installing. 

sudo apt install nvidia-driver-460 nvidia-dkms-460

7. Now, we restart our instance. For that, go back to the instance page, stop the selected instance and connect it again from the “Instance State” menu.

Instance State

8. Now, when we execute “nvidia-smi”, we can get details of the Graphic card. If the graphic card displays, that means that we have successfully installed and now we can use any application. 

nvidia-smi

Conclusion

You have successfully launched an Amazon EC2 instance with GPU capabilities, installed the necessary NVIDIA drivers.This setup will allow you to leverage the power of GPU for your computational tasks. If you encounter any issues, refer to the official AWS documentation or the NVIDIA documentation for further assistance.