Software Setup

Software Setup#

This section describes how to set up your AMD Vitis™ AI development environment to perform the following tasks:

  1. Offline model quantization and compilation — Use the Quark and Vitis AI Compiler Docker image.

  2. Cross-compilation of Arm application software — Use the Vitis AI SDK.

  3. VEK385 AI Reference Design — Use the Vitis AI GitHub repository for software examples, tutorials, and VEK385 platform source code. Use the repo to extract the prebuilt boot images needed for VEK385 board bring-up. An additional MIPI streaming example is also available; see MIPI Streaming Example.

  4. Adaptive SoC development tools — Use AMD Vivado™ Design Suite 2026.1.

Follow these steps to create a reproducible, ready-to-use environment for evaluating example models and developing custom applications.

AMD Quark and Vitis AI Compiler Docker Image#

This Docker image contains the development tools required to quantize and compile your AI models. To set up your environment and start using AMD Quark and the Vitis AI offline compiler, see Docker setup.

Vitis AI SDK#

The Vitis AI SDK / sysroot provides the cross-compilation environment and libraries required to build applications for the target device.

If the sysroot is not installed, follow these steps:

  1. Download it from here: SDK for cross-compilation (vai6.3_vek385_sdk.sh).

  2. Complete the following steps on the host machine before you cross-compile applications for the target:

    1. Copy the SDK installer to a temporary location:

      cp vai6.3_vek385_sdk.sh /tmp/sdk.sh

    2. Make the installer executable:

      chmod +x /tmp/sdk.sh

    3. Clear the library path to avoid conflicts:

      unset LD_LIBRARY_PATH

    4. Install the SDK to the specified path:

      /tmp/sdk.sh -y -d <SYSROOT_PATH>

    5. Source the sysroot environment for cross-compilation:

      source <SYSROOT_PATH>/environment-setup-cortexa72-cortexa53-amd-linux

VEK385 AI Reference Design (Source Code and Pre-built Images)#

This section provides instructions on how to download and install the VEK385 AI Reference Design and the associated software examples and tutorials.

The Vitis AI GitHub repository provides access to the software examples, tutorials, and the VEK385 platform source code.

The following TAR files contain the prebuilt boot images for VEK385 board setup that allow you to run these software examples on hardware. Download the archive that matches your board revision:

The release package includes:

  • Boot images

  • ONNX model

  • VART-based and ONNX Runtime-based applications

Extract the package under /home on the host machine. You can extract the package to another path, but you must update commands to match your path.

The following commands show how to extract the prebuilt images and clone the Vitis AI GitHub repository. The examples use the Rev-B archive; substitute vai6.3_vek385_RevA_ai_ref_design_boot_images.tar if you are using a Rev-A board.

Extract To /home#

sudo mkdir -p /home/amd/boot_images
sudo tar -xvf vai6.3_vek385_RevB_ai_ref_design_boot_images.tar -C /home/amd/boot_images --strip-components=1
sudo git clone -b release/6.3 https://github.com/amd/Vitis-AI.git /home/Vitis-AI
cd /home

Extract To A User-Owned Directory#

mkdir -p <your_preferred_path>/amd/boot_images
tar -xvf vai6.3_vek385_RevB_ai_ref_design_boot_images.tar -C <your_preferred_path>/amd/boot_images --strip-components=1
git clone -b release/6.3 https://github.com/amd/Vitis-AI.git <your_preferred_path>/Vitis-AI
cd <your_preferred_path>

Note

Verify directory ownership and permissions before extraction. Use sudo when the extraction directory belongs to root (for example, /home). For a user-owned directory, you might not need sudo.

Directory Structure#

This topic assumes that you extract vai6.3_vek385_RevB_ai_ref_design_boot_images.tar to /home/amd/boot_images. The following directory structure shows the extracted contents:

/home/
|-- amd/
|   |-- boot_images/
|   |      |-- amd-end-user-license-agreement.pdf
|   |      |-- vitis-ai-vaiml-versal-2ve-vai-6.3-third-party-notices.pdf

AMD End User License Agreement#

Review the licensing files under /home/amd/boot_images:

  • amd-end-user-license-agreement.pdf

  • vitis-ai-vaiml-versal-2ve-vai-6.3-third-party-notices.pdf

Running and Building the VEK385 AI Reference Design#

After the environment is set up, continue as follows:

MIPI Streaming Example#

The VEK385 MIPI Streaming AI Reference Design is an additional example that uses this same host environment but requires a Rev-B board and extra camera and display hardware. See that section for its full requirements, build steps, and run procedure.

The Vitis AI GitHub repository provides access to the VEK385 MIPI design platform source code.

The following TAR file contains the prebuilt boot images for VEK385 board setup that allow you to run software examples on MIPI streaming enabled hardware. Download the prebuilt boot image archive for the RevB board:

Vivado Installation#

For instructions on installing Vivado 2026.1, see UG973.

Contents