Enabling real-time AI inference on AMD adaptive SoCs

AMD Vitis AI

Overview

AMD Vitis™ AI software is an AI inference development platform for AMD adaptive SoC devices. The toolchain provides support for a rich set of AI models via the Vitis AI compiler, the neural processing unit (NPU) IP, runtime software, utilities and tools like the AMD Quark quantizer, libraries, and example designs. Vitis AI software is designed for high efficiency and ease of use, enabling developers to deploy optimized AI inference on embedded platforms.

The latest release of Vitis AI provides General Access support for two device series:

  • AMD Versal AI Edge Series
  • AMD Versal AI Edge Series Gen 2

Development Flow

  • Support for industry standard deep learning frameworks and models, including CNNs and select vision transformer architectures
  • Quantization tools to balance model accuracy, performance, and power consumption
  • Model compilation and runtime APIs that enable efficient deployment of custom AI inference workloads
  • NPU IP optimized for deterministic, low-latency, and scalable AI inference on embedded platforms
Image Zoom
AMD Vitis AI user application diagram

Get Started

Deployment

Deploy AI inference on AMD evaluation boards using Vitis AI software, enabling rapid prototyping and system bring‑up for embedded applications such as automotive, machine vision, industrial, and robotics.

Vitis AI Reference Designs target AMD Evaluation Kits:

Browse AMD and Partner Edge Platforms

Support

Looking to share your AI experiences, ask questions, and help solve issues? Check out the Vitis AI Community Forum.

For Vitis AI support of AMD Versal™ AI Core and Zynq™ UltraScale+™ MPSoC with the NPU technology, or any roadmap related inquiries please contact your AMD sales representative.

For legacy releases with the DPU technology: View Legacy Documentation >

Sign Up for Vitis AI News

Get the latest Vitis AI news and keep up to date on the latest release.

Sign Up