AWS IoT Greengrass Version 2 core device requirements - Amazon Lookout for Vision

AWS IoT Greengrass Version 2 core device requirements

To use an Amazon Lookout for Vision model on an AWS IoT Greengrass Version 2 core device, your model has various requirements of the core device.

Tested devices, chip architectures, and operating systems

We expect Amazon Lookout for Vision to work on the following hardware:

  • CPU architectures

    • X86_64 (64-bit version of the x86 instruction set)

    • ARM_64 (ARMv8 64-bit CPU)

  • NVIDIA GPU Accelerator with sufficient memory capacity (At least 6.0 GB for a running model).

The Amazon Lookout for Vision team has tested Lookout for Vision models on the following devices, chip architectures, and operating systems.

Devices

Device Operating system Architecture Accelerator Compiler options

jetson_xavier (NVIDIA® Jetson AGX Xavier)

Linux

ARM_64

NVIDIA

{'gpu-code': 'sm_72', 'trt-ver': '7.1.3', 'cuda-ver': '10.2'}

g4dn.xlarge (EC2 Instances (G4) with NVIDIA T4 Tensor Core GPUs)

Linux

X86_64/X86-64

NVIDIA

{'gpu-code': 'sm_75', 'trt-ver': '7.1.3', 'cuda-ver': '10.2'}

g5.xlarge (EC2 Instances (G5) with NVIDIA A10G Tensor Core GPUs)

Linux

X86_64/X86-64

NVIDIA

{'gpu-code': 'sm_80', 'trt-ver': '8.2.0', 'cuda-ver': '11.2'}

Core device memory and storage

To run a single model and the Amazon Lookout for Vision Edge Agent, your core device has the following memory and storage requirements. You might need more memory and storage for your client application component.

  • Storage – At least 1.5 GB.

  • Memory – At least 6.0 GB for a running model.

Required software

A core device requires the following software.

Jetson Devices

If your core device is a Jetson device, you need the following software installed on the core device.

Software Supported versions

Jetpack SDK

4.4, 4.5, or 4.5.1

Python and Python virtual environment for Lookout for Vision Edge Agent version 0.x

3.7

Python and Python virtual environment for Lookout for Vision Edge Agent version 1.x

3.8 or 3.9

X86 hardware

If your core device uses x86 hardware, you need the following software installed on the core device. Software versions vary depending on the microarchitecture of the NVIDIA GPU that you use.

NVIDIA GPU with microarchitecture prior to Ampere (compute capability is less than 8.0)

Required software for an NVIDIA GPU with a microarchitecture prior to Ampere (compute capability that is less than 8.0). The gpu-code must be less than sm_80.

Software Supported versions

NVIDIA CUDA

10.2

NVIDIA TensorRT

At least 7.1.3 and less than 8.0.0

Python and Python virtual environment for Lookout for Vision Edge Agent version 0.x

3.7

Python and Python virtual environment for Lookout for Vision Edge Agent version 1.x

3.8 or 3.9

NVIDIA GPU with Ampere microarchitecture (compute capability 8.0)

Required software for an NVIDIA GPU with the Ampere microarchitecture (compute capability is 8.0). The gpu-code must be sm_80).

Software Supported versions

NVIDIA CUDA

11.2

NVIDIA TensorRT

8.2.0

Python and Python virtual environment for Lookout for Vision Edge Agent version 0.x

3.7

Python and Python virtual environment for Lookout for Vision Edge Agent version 1.x

3.8 or 3.9