Using AWS Panorama after update 8.0.29 - AWS Panorama
Application Base Image ChangesRunning Applications on an OpenGPU framework

Using AWS Panorama after update 8.0.29

AWS Panorama version 8.0.29 ends support for SageMaker Neo models. This is because SageMaker Neo for Edge Devices has pulled support for the NVIDIA Jetpack SDK. Jetpack 4.6 is not supported on 8.0.29, although you can still deploy Jetpack 4.6 Neo models on older images. If your application is currently using SageMaker Neo model nodes, you have two choices for how to proceed:

  1. Rebuild your application without the Neo models for use with Jetpack 5.1.1. The recommended strategy is to run models directly on the GPU instead. See the following section on Running apps on OpenGPU for more information.

  2. Decline the update to 8.0.29 and keep using Neo model nodes.

While the 8.0.29 image is marked as "Mandatory", you will not be forced to upgrade to it. While using the old image, all regular functions are available including managing (deploying) applications and OTA from one pre-8.0.29 image to another pre-8.0.29 image. However, you will not be able to OTA from a post-8.0.29 image to a previous, older image.

The following are examples of which version changes are allowed and not allowed:

  • Allowed: 6.0.8 -> 6.2.1

  • Allowed: 6.2.1 -> 7.0.13

  • Allowed: 7.0.13 -> 8.0.29

  • Not allowed: 8.0.29 -> 7.0.13

Application Base Image Changes

The updated base Docker images can be at the following link, for both 1.2.1-arm64v8 and latest_jetpack5 versions:

Amazon ECR Public Gallery

The application base image changes for 8.0.29 are as follows:

  • Change base image from Ubuntu 18 to Ubuntu 20

  • Software versions:

    1. Python - 3.8.10

    2. NumPy - 1.24.4

Running Applications on an OpenGPU framework

If you would like to update to a version of AWS Panorama that has ended support for SageMaker Neo, you can migrate your SageMaker Neo apps to a framework that uses OpenGPU. The advantage of using OpenGPU is that your application has direct GPU and hardware access, giving you greater flexibility and performance.

The following links are some sample applications that demonstrate how OpenGPU can be used to run Jetpack 5 apps:

Run Yolov5s with Tensor RT in Panorama

Convert YoloV5s to ONNX and run on Panorma

Run YoloV5s on Panorama with Torch and TorchVision