Amazon EC2 Elastic GPUs allow you to easily attach OpenGL 4.3 acceleration capabilities to your Amazon EC2 instances. Elastic GPUs function similarly to physical GPUs in a desktop or server in that they accelerate graphics operations by using special purpose hardware for graphics commands. Elastic GPUs are attached over the network and exist independently of Amazon EC2 instances. As a result they are supported by multiple instance types, including M5, M4, M3, C5, C4, C3, R4, R3, X1, I3, D2, and T2 instances. Once attached, you can use Elastic GPUs to accelerate your 3D applications in much the same way as if there’s a GPU local to your EC2 instance. At the moment, Elastic GPUs support Microsoft Windows Server 2012 R2 or above.

Amazon Elastic GPUs are created when you launch EC2 instances and specify the size of Elastic GPUs you want to use. Elastic GPUs are created in the same Availability Zone as your EC2 instances, and can be from 1 GiB to 8 GiB in GPU memory size. Once created and attached, the Elastic GPU driver will detect the presence of Elastic GPUs and establish the connection between your EC2 instances and Elastic GPUs. At that point, the instance can interact with the Elastic GPUs’ OpenGL library in much the same way as locally attached GPUs.

An Elastic GPU can only attach to one EC2 instance, and each EC2 instance can only have one Elastic GPU attached to it. Elastic GPUs can be used in conjunction with many desktop streaming protocols, such as Desktop Cloud Visualization (DCV), Remote Desktop Protocol (RDP), and Virtual Network Computing (VNC).

Launch an Amazon EC2 Instance for Free

Try Amazon EC2 for Free

AWS Free Tier includes 750 hours of Linux and Windows t2.micro instances each month for one year. To stay within the Free Tier, use only EC2 Micro instances.

View AWS Free Tier Details »

Elastic GPUs come with different sizes, from the smallest eg1.medium with 1GiB of GPU memory, to eg1.2xlarge with 8GiB of GPU memory. Therefore, Elastic GPUs are ideal for cost-effectively adding the right amount of GPU resource to your workload. This flexibility allows you to accelerate the graphics performance of applications such as virtual desktops, gaming, industrial design, or HPC visualization, for a fraction of the cost of stand-alone graphics instances.

Model GPU Memory
eg1.medium 1 GiB
eg1.large 2 GiB
eg1.xlarge 4 GiB
eg1.2xlarge 8 GiB

Elastic GPUs come with a driver, which is an OpenGL acceleration library in your EC2 instances. The driver intercepts OpenGL calls, and execute the graphics commands on GPUs that are remote to your EC2 instances. The frames rendered by graphics commands are passed back to Elastic GPU driver.