• General

    Q: How can I attach and detach Elastic GPUs to my Amazon EC2 instances?

    Elastic GPUs can be attached to EC2 instances at creation. You can specify the size of Elastic GPUs through EC2 Console, AWS SDK and CLI. Elastic GPUs are detached when you terminate your EC2 instance.

    Q: Which instance types and sizes are supported by Elastic GPUs?

    Elastic GPUs support a wide range of current generation EC2 instances, including M5, M5d, M4, M3, T3 (t3.medium or larger), T2 (t2.medium or larger), C5, C5d, C4, C3, z1d, R5, R5d, R4, R3, X1e, X1, H1, I3, D2, P3, and P2.  










    T3 t3.medium or larger t3.medium or larger t3.medium or larger t3.medium or larger


    t2.medium or larger

    t2.medium or larger

    t2.medium or larger

    t2.medium or larger

















    G3 - - - -











    Q: Which operating system do you support?

    Elastic GPUs support Windows Server 2012 R2, and Windows Server 2016.

    Q: Can I reserve capacity for Elastic GPUs?

    No, Elastic GPUs can only be provisioned as on-demand.

    Q: How does my EC2 instances communicate with the attached Elastic GPUs?

    Elastic GPUs leverage the same bandwidth as instance network traffic. The channel is set up using an Elastic Network Interface (ENI) created in your VPC called the Elastic GPU endpoint.

    Q. Do I need a driver for Elastic GPUs?

    You need to install Elastic GPUs driver in your instance. However the driver is not a device driver. Instead, it is an Amazon-optimized OpenGL library that detects the presence of and connects to the attached Elastic GPU. When applications make OpenGL API calls for 3D operations, the Amazon OpenGL library re-directs those calls to the Elastic GPU over the network, where the operations are processed and results are returned back to the library. Amazon EC2 manages the physical hardware, GPU device drivers, and the transportation of commands to and from your EC2 instance. Elastic GPUs infrastructure is accessed through a network endpoint within an attached instance’s VPC, and OpenGL command traffic and results are transmitted using the instance’s provisioned networking capacity.

    Q: Can I see Elastic GPUs in my device manager?

    No, there is no additional device for Elastic GPUs in your device manager.

    Q: Can I attach multiple Elastic GPUs to a single instance, or attach the same Elastic GPU to multiple instances?

    No, you can only attach one Elastic GPU to one EC2 instance.

    Q: Can I attach/detach Elastic GPUs while the instance is running?

    No, Elastic GPUs can only be attached at instance creation and detached at instance termination.

    Q: Can I attach an on-premise machine to an Elastic GPU on EC2?

    No, Elastic GPUs can only be attached to EC2 instances.

    Q: Can an instance start or launch fail because of Elastic GPUs?

    If Elastic GPU capacity cannot be acquired due to insufficient capacity, an insufficient capacity error message is presented to you and the instance launch or start will fail. If the subnet presented has less IPs available than is required to launch all of the requested instances plus create the Elastic GPU endpoint, then you are presented with an exception with some additional information detailing that an extra IP is required for first launch with an Elastic GPU into an Availability Zone for this VPC.

    Q: How do I know Elastic GPUs, including whether necessary libraries are correctly installed, from within the instance?

    You can click on the Elastic GPU icon in the notification area of your taskbar and see whether Elastic GPU driver is installed correctly and Elastic GPU is functioning properly. If everything is working properly, the Elastic GPU UI will show healthy status. You need to re-install the Elastic GPU driver if the Elastic GPU icon is missing.

    Q: Which applications does AWS certify to work with Elastic GPUs?

    AWS does not offer application certification on behalf of the software vendors. However we provide a Graphics Certification Program to ISVs so that they can certify their applications on AWS.

    Q: What desktop streaming protocols do Elastic GPUs support?

    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).

    Q: Can I run a single application at a time or Elastic GPUs supports multiple applications running at the same time?

    Elastic GPUs do not impose any restriction on the number of applications running at the same time.

    Q: My application is reporting the system has 0 MB of GPU memory, why?

    If you are using Microsoft Basic Display Adaptor, your GPU memory query will return 0MB. However, you will still be able to use the GPU memory provided by Elastic GPUs through OpenGL API calls.

  • Performance

    Q: What factors can affect the performance of Elastic GPUs?

    There are many factors that may affect the performance of Elastic GPUs. This includes the available network bandwidth to your EC2 instance, the complexity of your 3D models, and the size of your OpenGL context.

    Q: What are the metrics I can get from Elastic GPUs?

    You can get GPU memory usage metrics of the Elastic GPUs through CloudWatch.

  • Networking

    Q: Can I attach Elastic GPUs to EC2 instances without VPCs?

    No, Elastic GPUs can only be attached to EC2 instances within a VPC.

    Q: Do Elastic GPUs consume network bandwidth of my EC2 instances?

    Yes, the communication channel between Elastic GPUs and EC2 instance is through an Elastic Network Interface. All bandwidth consumed by that Elastic Network Interface is counted towards the EC2 instance bandwidth limit.

    Q: How can I manage the Elastic GPU endpoint used by Elastic GPUs?

    Elastic GPU endpoint is an Elastic Network Interface associated with your Elastic GPUs. You can manage the security group associated to the Elastic Network Interface in the same way as you would for other Elastic Network Interface. However you are not able to detach the Elastic Network Interface from the EC2 instance it is attached to.

    Q: What is the security group requirement of Elastic GPUs?

    In order to let the network traffic flow between your EC2 instance and the Elastic GPU it attached to, the security group associated with the Elastic GPUs Elastic Network Interface should allow all TCP outbound traffic on port 2007.

    Q: Which kind of data is transferred on the network?

    We transfer OpenGL commands from your EC2 instances to Elastic GPU and rendered frames from Elastic GPU to EC2 instance. There are also some control information being transferred between your EC2 instances and Elastic GPU. However all data transferred never leaves your VPC.

    Q: Which network latency is guaranteed between the Windows instance and the Elastic GPU?

    Elastic GPU is subject to the same latency of the AWS network.


  • Billing & Metering

    Q: What is the lifecycle of an Elastic GPU and how will I be billed?

    Elastic GPUs are billed for by the hour, and the Elastic GPU lifecycle mirrors the lifecycle of the instance it is attached to.

    • Pending - When you launch an instance and Elastic GPU, the Elastic GPU enters the pending state. While in the pending state, Amazon EC2 finds Elastic GPU capacity and establishes a channel between your instance and Elastic GPU. You are not billed for time spent in the pending state.
    • Running - After the instance and Elastic GPU are both ready for you, the Elastic GPU enters the running state. As soon as your instance transitions to the running state, the instance is healthy. You are billed for Elastic GPUs for each hour or partial hour that you keep the instance running, if your Elastic GPUs are in healthy state.
    • Stopping and stopped - When you stop your instance (EBS backed instances only) with an Elastic GPU attached, the instance enters the stopping state and similarly, your Elastic GPU enters a stopped state. We don't charge for Elastic GPUs that are stopped.
    • Shutting-down and terminated - When you terminate the instance, as soon as the instance status changes to shutting-down or terminated, you stop incurring charges for that instance and Elastic GPU.


    Q: Will I get billed when my Elastic GPUs are impaired?

    No, your Elastic GPUs will only get billed if they are in healthy state.

    Q: Do I pay for the network bandwidth used by Elastic GPUs?

    No. Given the Elastic GPU and its associated EC2 instance is in the same Availability Zone, network transfer between Elastic GPU and EC2 instance is free.

    Q: Do your prices include taxes?

    Except as otherwise noted, our prices are exclusive of applicable taxes and duties, including VAT and applicable sales tax. If you have a Japanese billing address, use of AWS services is subject to Japanese Consumption Tax. Learn more.

  • API Support

    Q: Which graphics API do Elastic GPUs support?

    Elastic GPUs support OpenGL 4.3.

    Q: Which OpenGL version do Elastic GPUs support?

    Elastic GPUs support OpenGL 4.3.

    Q: Which OpenGL shader language version do Elastic GPUs support?

    Elastic GPUs support GLSL 4.30.

    Q: As an OpenGL developer, how can I detect my application is running on an Elastic GPUs implementation?

    You are able to use OpenGL API to detect the presence of Elastic GPU.