Because of their graphics processing unit (GPU) resources, Amazon EC2 Elastic GPU instances are a good fit to run applications that render images, such as those for gaming, 3-D modeling or virtual desktops.
To use an Elastic GPU, a developer attaches one to an on-demand Elastic Compute Cloud (EC2) instance when he or she first launches the instance via the AWS Management Console, AWS Command Line Interface or AWS software development kits.
Elastic GPUs support a range of EC2 instance types, enabling a developer to choose the optimal amount of memory, compute and storage resources for an application. A developer can also choose between four partial GPU sizes -- 1, 2, 4 or 8 gibibytes of GPU memory -- instead of a full physical GPU to best fit a workload's needs and save money.
An EC2 instance connects to its attached Elastic GPU through an Elastic Network Interface in an Amazon Virtual Private Cloud. The developer must install an Elastic GPUs driver on an EC2 instance that acts as an OpenGL library to facilitate the connection between the instance and the Elastic GPU. AWS manages the underlying EC2 hardware, device drivers and command transportation to and from the instance.
Elastic GPUs, which support several operating systems and desktop streaming protocols, detach from an EC2 instance when the instance is terminated.
AWS offers a Graphics Certification Program for independent software vendors and developers to certify that their applications can run on the cloud provider's GPU-based offerings.