Axelera AI Metis Compute Board: RK3588-Based SBC with Metis AI Accelerator for Edge AI Applications

Axelera AI has introduced the Metis Compute Board, a compact SBC designed for edge AI applications. It features the Rockchip RK3588 processor, combining Cortex-A76 cores for high performance and Cortex-A55 cores for power efficiency.

The board integrates the Metis AI accelerator with a four-lane PCIe connection to the CPU, enabling high-speed AI inference. With 16GB of LPDDR4 memory, it supports multi-stream and multi-model workloads, making it suitable for computationally intensive AI tasks at the edge.

Previously, we covered SBCs designed for edge AI applications, including the iBASE IB962, OrangePi 5 Ultra, and Avnet MaaXBoard OSM93. Feel free to check them out if you’re interested in similar products.

Metis Compute Board Specifications:

• Processor: Rockchip RK3588 with 4x Cortex-A76, 4x Cortex-A55, and 3x Cortex-M0 cores; integrated GPU, NPU, and video codec
• AI Acceleration: Axelera AI Metis AIPU (up to 120 TOPS) with up to 16GB LPDDR4X RAM
• Storage: 64GB onboard eMMC, MicroSD card slot, 2x SATA ports
• Video Outputs: HDMI 2.0, DisplayPort (up to 4096 x 2160 @ 24Hz)
• USB Connectivity: 4x USB 3.1 Gen1, 1x USB-C, 3x USB 2.0 headers
• Networking: 2x Gigabit Ethernet ports
• Expansion: 1x M.2 E-key, 1x M.2 B-key, GPIOs, UART, ADC, I2C, SPI
• Power: 12V DC input
• Operating Temperature: -20°C to 70°C
• Form Factor: Mini-ITX (170 x 170 mm)

The Metis Compute Board is designed for energy-efficient operation and offers multiple connectivity options. It features 64GB of onboard eMMC storage, two SATA ports, and a microSD slot for expanded storage. Networking is handled by two Gigabit Ethernet ports, while peripheral connectivity includes four USB 3.1 Gen1 ports, one USB-C port, and three USB 2.0 headers. Additional interfaces such as UART, I2C, and GPIOs enable easy integration with sensors and other peripherals.

The board comes with a pre-configured Board Support Package (BSP) based on Yocto, providing support for industry-standard frameworks and APIs. This allows for seamless deployment and development, ensuring a ready-to-use AI computing platform.

The Metis Compute Board is a complete AI acceleration system featuring the Voyager SDK, which streamlines application development. The SDK includes tools for AI model deployment and inference optimization, with support for TensorFlow, PyTorch, and ONNX. It also offers compiler features for automatic quantization and graph optimization to improve performance.

Built on a RISC-V-based architecture with D-IMC (digital in-memory compute), the Metis platform is designed for efficient AI processing. Axelera AI provides a model zoo with pre-trained models for image classification, object detection, segmentation, and pose estimation, enabling support for complex neural networks.

Pricing and availability details for the Metis Compute Board have not been disclosed. More information can be found on the product page.

Please follow and like us: