AMD VEK385: Versal AI Edge Gen 2 Platform for Embedded Vision and AI

Modern embedded vision systems increasingly need to process multiple sensors, run AI inference, and make real-time decisions while maintaining low latency and predictable performance. Traditional processors can handle many of these workloads, but applications such as autonomous machines, robotics, industrial vision, and advanced driver assistance often require a more flexible architecture.

The AMD VEK385 Evaluation Kit provides a development platform for the AMD Versal AI Edge Series Gen 2, combining general-purpose processing, AI acceleration, programmable logic, DSP processing, image signal processing, and high-speed connectivity in a single adaptive platform.

What Is the AMD VEK385?

The VEK385 is an evaluation platform built around the AMD Versal AI Edge Series Gen 2 2VE3858 adaptive SoC. It is designed to help developers evaluate and prototype embedded AI, vision, sensor-processing, and real-time control applications before developing a production system.

Unlike a conventional development board centered around a CPU or GPU, the VEK385 exposes the heterogeneous architecture of the Versal AI Edge Gen 2 device. This allows different parts of an application to run on the processing engine best suited to the workload.

The platform is particularly relevant to applications that need to move data through an entire pipeline—from sensors and preprocessing to AI inference and real-time control.

Inside the Versal AI Edge Gen 2 Architecture

The VEK385 is built around the 2VE3858 device, which combines several processing resources:

Arm Cortex-A78AE processors for high-level application processing
Arm Cortex-R52 processors for real-time workloads
AI Engine-ML v2 technology for machine-learning acceleration
Programmable logic for customized hardware acceleration and sensor processing
DSP Engines for signal-processing workloads
Image Signal Processors (ISPs) for camera and imaging applications
Video Codec Unit (VCU) for video processing
High-speed networking and programmable I/O

This heterogeneous architecture allows developers to partition an application instead of forcing every workload onto the CPU.

For example, Linux-based application software can run on the Cortex-A78AE processors while deterministic control tasks run on the real-time processors. Sensor preprocessing can be implemented in programmable logic, while AI inference can be accelerated using the AI Engines.

VEK385 for Embedded Vision

Vision is one of the most important applications for the VEK385 platform.

A typical embedded vision pipeline may look like:

Camera Sensors → Sensor Interface → ISP / Preprocessing → AI Inference → Decision Making → Control

The VEK385 can accelerate different portions of this pipeline using different processing resources.

The platform also supports image signal processing and camera-oriented interfaces, making it possible to build systems that connect cameras directly into the processing architecture. AMD's current VEK385 ISP documentation demonstrates camera development using a GMSL2 interface module and image sensors connected through the board's FMC interface.

This makes the platform useful for applications involving multiple cameras, sensor fusion, computer vision, and AI-based image analysis.

Camera Processing and ISP

One particularly important feature for embedded vision developers is the integrated ISP capability of the Versal AI Edge Series Gen 2.

An ISP can perform operations such as image preprocessing before frames are passed to computer vision or AI algorithms. This can include operations needed to transform raw sensor data into useful image information for downstream processing.

AMD provides a dedicated Versal AI Edge Series Gen 2 ISP development flow, including support for the VEK385 evaluation platform. The documented setup can use an FMC GMSL2 module and multiple camera sensors, demonstrating how the platform can be used to evaluate multi-camera imaging pipelines.

For systems using multiple cameras, this type of architecture can help move computationally intensive image processing away from the main application processor.

AI Acceleration

The VEK385 also incorporates AI Engine-ML v2 technology for machine-learning workloads.

AMD's platform comparison identifies the VEK385 as the Gen 2 successor to the VEK280, moving from the first-generation AI Engine-ML architecture to AIE-ML v2. The VEK385 platform uses the VE3858 device and provides 144 AI Engine-ML v2 tiles along with 2,064 DSP Engines.

This architecture is intended for applications where AI inference must operate alongside other real-time processing rather than functioning as an isolated accelerator.

For example, a vision system could perform image preprocessing in programmable logic and the ISP, execute a neural-network model using the AI Engines, and then use the Arm processors to coordinate the application and system-level response.

Connectivity for Multi-Sensor Systems

Embedded vision systems rarely consist of a single camera. Autonomous vehicles, robotics platforms, industrial inspection systems, and smart-city equipment can combine RGB cameras, stereo cameras, thermal sensors, LiDAR, and other inputs.

The Versal AI Edge Gen 2 architecture is designed around this type of sensor-fusion workload. AMD highlights the ability to connect standard and proprietary sensors and process their data through programmable logic, AI Engines, and other processing resources.

The VEK385 provides a range of board-level interfaces, including FMC+, QSFP28, SFP28, PCIe, DisplayPort, HDMI, and high-bandwidth LPDDR5X memory.

This connectivity gives developers flexibility when building prototypes involving cameras, networking, displays, storage, and external accelerators.

Software Development

Hardware flexibility is only useful when the software environment can take advantage of it.

The VEK385 development ecosystem includes AMD Vivado Design Suite and the Vitis unified software platform. AMD also provides Vitis AI tools, example designs, tutorials, and platform support for developing applications on the VEK385.

Current AMD Vitis documentation recommends the VEK385 base platform as a starting point for AI Engine development. The Gen 2 development flow also uses an embedded Linux/Yocto-based environment for applications running on the Cortex-A78 processing system.

This provides a path from hardware configuration and programmable logic through Linux software and AI application development.

Applications for the VEK385

The VEK385 and Versal AI Edge Series Gen 2 are targeted at applications where low latency, heterogeneous processing, and sensor integration are important.

Potential applications include:

Automotive: ADAS and autonomous driving
Robotics: autonomous mobile robots and machine vision
Industrial: industrial PCs, inspection, and edge AI
Smart cities: intelligent cameras and edge analytics
Aerospace and defense: avionics, UAS, detection, and tracking
Healthcare: ultrasound, endoscopy, and 3D imaging

AMD has also demonstrated automotive vision workloads on the VEK385, including perception software running on the platform for multi-camera automotive applications.

Why the VEK385 Matters for Embedded Vision

The main advantage of the VEK385 is that it is more than a conventional processor evaluation board. It provides a heterogeneous architecture where camera processing, programmable logic, AI inference, signal processing, application software, and real-time control can operate as parts of the same system.

For embedded vision developers, this can be particularly valuable when a product requires multiple cameras, custom sensor interfaces, real-time preprocessing, AI inference, and deterministic system behavior.

The VEK385 provides a development platform for exploring this architecture before translating the design into a production embedded system.

Building Camera Systems Around AMD Versal

Developing an embedded vision product around the AMD VEK385 or Versal AI Edge platform requires more than selecting a processor. Camera sensors, MIPI or GMSL interfaces, ISP configuration, sensor drivers, image processing, Linux integration, and AI pipelines all need to work together.

Turning an evaluation platform such as the VEK385 into a production-ready vision system requires careful integration of the camera hardware, high-speed interfaces, ISP, processing architecture, and software stack. Explore our hardware design services to learn how we support custom camera integration, schematic and PCB design, and complete embedded vision hardware development for advanced AI and imaging systems.