Why AI at the Edge Wins the Race for Real-Time Decisions

Artificial intelligence (AI) has transformed numerous industries, but its true potential lies at the “edge” of the network – where data originates from sensors or devices.

While cloud-based AI is powerful, issues with latency and bandwidth can hamper its efficiency for edge applications.

This is where edge AI shines, offering clear advantages for real-time decision-making and intelligent automation.

The Cloud vs. The Edge: Key Differences

Cloud AI excels at processing massive datasets for complex tasks like training AI models or large-scale analytics. However, relying solely on the cloud introduces latency – the time required for data to travel between the edge device and the cloud server. This delay can be problematic in scenarios demanding immediate action. Cloud processing also often requires substantial bandwidth, which can challenge resource-constrained edge devices, especially in remote areas.
In contrast, edge AI empowers devices to process data locally, eliminating latency for real-time decision-making based on the live data stream. Edge AI also reduces reliance on constant connectivity, suiting scenarios with unreliable internet access. For example, an industrial system using edge AI can analyze machinery sensor data in real-time to detect potential failures, preventing costly downtime proactively.

Edge AI in Action, use cases:

Real-Time Video Analytics

  • Security and Surveillance
  • Retail Analytics


Autonomous Vehicles

  • Navigation
  • Object Detection

Industrial IoT (IIoT)

  • Predictive Maintenance
  • Quality Control



  • Wearable Devices
  • Medical Imaging

Smart Home and Cities

  • Energy Management
  • Traffic Management



  • Personalized Recommendations
  • Inventory Management


  • Crop Monitoring
  • Animal Health



  • Network Management
  • Content Delivery

Augmented Reality (AR) and Virtual Reality (VR)

  • Real-Time Processing
  • Interactive Applications


Energy Sector

  • Smart Grids
  • Renewable Energy Management

VCS³ – Single-board computer with AMD ZYNQ device

Due to its physical dimensions, power, and versatility, the VCS³ is ideal for deployment for Edge AI duties. It is a small Single-Board Computer with an AMD© ZYNQ™ device with integrated ARM CPUs and FPGA fabric. Measuring just 30mm x 50mm, this tiny workhorse can be placed almost anywhere, opening up the benefits of FPGAs to many more applications.

The VCS³ utilizes an AMD UltraScale+ MPSoC coupled with high-speed LP-DDR4 memory to produce a highly compact evaluation platform. Together with four digital camera interfaces, a 9-axis IMU, and a CAN-Bus interface, this platform is ideally suited for autonomous machines, cameras or automation.

Device booting can be from SPI ROMs or eMMC flash, with no bulky, fiddly or unreliable SD cards.

Numerous onboard power rails are generated from a single external 5V supply, or via a USB3 Type-C interface.

Several LEDs indicate board functionality and numerous test points allow access to the various interfaces.

Vitis AI: Porting AI to FPGAs

  • Vitis AI Model Compiler (VAI-C): Transforms pre-trained AI models from frameworks like TensorFlow and PyTorch for FPGA execution. It optimizes models for efficiency using techniques like quantization and pipelining, then translates them into FPGA-compatible hardware representations.
  • Vitis AI Library (XLA): Provides pre-built, optimized hardware kernels for common AI operations, enhancing performance and reducing development time. It also allows for custom hardware accelerator creation.

Broader Ecosystem Components:

  • DPUCZDX8G: A high-performance AI engine for AMD Zynq UltraScale+ MPSoC devices, ideal for deep learning tasks like image recognition and video analytics with low latency.
    • Optimized for convolutional neural networks (CNNs), supporting various architectures and operations essential for CNN-based workloads.
    • Features up to four homogeneous cores for efficient parallel processing, ensuring high performance for real-time AI inference.
    • Ensures predictable, consistent latency, crucial for applications requiring immediate response, such as medical imaging and autonomous vehicles.
    • Excels at local edge processing, reducing reliance on cloud communication, enhancing privacy, and improving data security.
  • DPUv3: Versatile AI engine supporting various AMD FPGA families.
  • Pre-built Software Libraries: Offer solutions for tasks like image pre-processing and data movement, saving development time.
  • Accelerator Development Kits (ADKs): Provide tools for designing custom hardware accelerators for specific AI model functionalities.

Sundance transforms AMD’s core technology into production-ready edge AI hardware and develops high-performance solutions leveraging AMD FPGAs, addressing real-world needs for low latency, ruggedness, and reliability in AI deployments.


At the forefront of Sundance’s offerings is the PCIe104Z board. Adhering to the industry-standard PCIe104 form factor, this compact board is ideal for space-constrained environments often encountered in industrial settings or embedded systems. Built around the powerful AMD Zynq UltraScale+ MPSoC family, the PCIe104Z boasts exceptional processing power with its dual-core ARM processing system and robust FPGA fabric.

Critically, the PCIe104Z is specifically engineered for harsh environments, ensuring reliable operation in scenarios with extreme temperatures, vibration, or shock. This ruggedness makes it a perfect choice for applications like industrial automation, smart grid monitoring, or even in-vehicle AI deployments where environmental conditions can be challenging.

Solar Express 125

For applications where the device is not used in harsh environments, Sundance offers the Solar Express 125 (SE125) board. This half-size PCIe board utilizes the same AMD Zynq UltraScale+ MPSoC architecture as the PCIe104Z, providing a significant boost in processing power for complex AI workloads such as object detection, image recognition, or natural language processing.

The processing capabilities of the SE125 cater to more demanding edge applications like video analytics for security systems or real-time traffic monitoring in smart cities. Importantly, the SE125 offers deployment flexibility – it can operate as a standalone dedicated processing unit or integrate seamlessly within a host PC for scenarios requiring additional system resources.


Another option for developers working within the PXIe ecosystem is the PXIe800Z board. This board integrates the AMD Zynq UltraScale+ MPSoC within the industry-standard PXIe form factor, offering seamless integration with existing PXI test and measurement setups commonly found in research labs or validation environments.

Additionally, the PXIe800Z features built-in support for industry-standard FMC (FPGA Mezzanine Card) modules. These modules provide further customization and expansion of functionality, allowing developers to add specific features like high-speed analog-to-digital converters or custom I/O interfaces tailored to their unique application needs.