Edge areas where demand for high-performance AI applications is growing
Versal Edge delivers 4x the performance per watt of NVIDIA GPUs.
10x the compute density of Xilinx's previous generation MPSoCs.


On the 11th, Xilinx held an online press conference and launched the 'Versal™ AI Edge' series, which provides 4x the AI performance per watt compared to GPUs and 10x the computing density compared to the previous generation adaptive SoC to achieve edge AI innovation.

The Versal AI Edge Adaptive Compute Acceleration Platform (ACAP) brings intelligent capabilities to a wide range of applications, including autonomous and collaborative driving robots that require the highest levels of functional safety, predictive maintenance systems in factories, and multi-purpose payloads for healthcare, aerospace, and defense markets.

The new adaptable platform is based on 'AI Engine-ML', which delivers four times the machine learning (ML) compute of the existing AI Engine architecture and integrates an improved memory hierarchy and new accelerator RAM for evolving AI algorithms. These architectural innovations enable high-performance edge devices, delivering up to four times the performance per watt and lower latency compared to NVIDIA's Jetson AGX Xavier GPU.

High computing density is essential for AI-based automation systems to accelerate applications ranging from sensors to AI and real-time control. The Versal AI Edge series delivers 10x the compute density of Xilinx's Zynq® UltraScale+™ MPSoCs, enabling intelligent autonomous systems.

Additionally, it supports safety standards for various markets, including automotive (ISO 26262), industrial (IEC 61508), and aviation (DO-254/178), enabling suppliers to achieve ASIL C/D compliance.

“AI chipsets supporting edge applications will more than double from 2021 to 2025,” said Dan Mandell, senior analyst for IoT and embedded technologies at VDC Research. “The Versal AI Edge series addresses this need with a purpose-built design for AI tasks while maintaining scalability and low power.”

The Versal AI Edge series is built on the 7nm Versal architecture and is miniaturized to perform ultra-low latency AI computing, while delivering power efficiency of less than 6W and the safety and security requirements required for edge applications.

◇ Accelerating the development of AI-based edge applications through various hardware and software.

Users can optimize algorithm-tailored engines using the heterogeneous platform Versal AI Edge Series, including: Scalar Engine for embedded computing, Adaptable Engine for sensor fusion and hardware adaptability, ASSP dedicated to edge applications, AI inference performance scalable up to 479 TOPS of INT4, and Intelligent Engine for advanced signal processing workloads such as vision, radar, lidar, and SDR (Software Defined Radio).

The connectivity block supports native MIPI for vision sensors with resolutions up to 8 megapixels and higher, critical for Level 2 and higher ADAS, from LPDDR-4266 and 32 Gb/s transceivers. It also supports all the protocols required by edge applications, from 40G multi-rate Ethernet to PCIe Gen 4 with CCIX. It also features an enhanced memory hierarchy, including an AI engine and accelerator RAM, making it suitable for a wide range of applications.

The Versal AI Edge ACAP is accessible to both hardware and software developers. It provides an easy design start with the Vivado® design tool for hardware developers, the Vitis™ unified software platform for software developers, Vitis AI for data scientists, and acceleration libraries and domain-specific operating systems and frameworks for the platform's target applications.

The Versal AI Edge series is the latest addition to the Versal ACAP portfolio, a software-programmable, adaptable SoC that delivers higher performance and flexibility than traditional CPUs, GPUs, and FPGAs. Modifiable at both the hardware and software levels, the ACAP dynamically adapts to a wide range of workloads and application requirements, from edge to cloud. The Versal AI Core and Versal Prime series are in mass production, and the Versal Premium is sampling now.

The Versal AI Edge series, scheduled to ship in the first half of 2022, is available with design documentation and support for early access customers. This will also include a roadmap for automotive and defense-grade devices. More information about the series can be found on the website .