Firefly’s Core-3576JD4: An AIoT Board capable of vision computing

AIoT, which integrates artificial intelligence directly into IoT devices using edge computing, opens doors to innovation in applications of limited connectivity and systems requiring local data processing. The Firefly Core-3576JD4 Low-Power Large-Model AIOT Core Board is one example of such a development.

The core board has an octa-core 64-bit processor–the Rockchip RK3576, a Mali C52 MC3 CPU providing 145G FLOPS. It also supports heterogeneous computing along with having an NPU performing up to 6-TOPS. Furthermore, this NPU supports mixed-precision operations like INT4/8/16, FP16, BF16, and TF32 and can run AI models having Transformer-based architectures. It also supports the following frameworks: TensorFlow, PyTorch, MXNet, and Docker container management technology. The Core-3576JD4 has LPDDR4/LPDDR4x RAM. For storage, the board supports eMMC with options ranging from 16GB to 256GB and UFS 2.0 for better data transfer rates. There is also an option for PCIe NVMe/SATA SSD expansion.

Special Processors and Accessories:

The Core-3576JD4 supports 8K@30fps and 4K@120fps video decoding for formats H.265/HEVC, VP9, AVS2, and AVI. The board is equipped with HDMI 2.1, eDP 1.3, and DP 1.4 interfaces, which enable high-resolution video output at up to 4K@120fps. These features allow it to help in digital signage, media streaming, and surveillance systems.

The Core-3576JD4 also has a 16-megapixel Image Signal Processor. It supports low-light noise reduction, an RGBIR sensor, and up to 120dB of HDR. The MIPI-CSI interfaces of the board can handle multiple camera inputs, allowing it to manage different types of camera setups. Also, the Core-3576JD4 has high-resolution E-Ink displays (1920 x 2560). The Core-3576JD4 uses RK-PQ algorithms to improve image quality. It also supports multiple refresh modes and has a latency of 20 milliseconds.

This board features a camera module that supports an 8-megapixel HD camera and a 2-megapixel dual-lens camera. It also includes a wireless module compatible with the EC20 4G-LTE module, EG25-G 4G LTE module, and RM500Q-GL 5G module, as well as GPS and GLONASS capabilities with the RG4538.

Connectivity options for Core-3576JD4:

The Core-3576JD4 has a number of connectivity options through a 260-pin SODIMM interface. Other than this, the supported interfaces include the I2C, UART, CAN, SPI, and GPIO. This kind of variety in connectivity allows the board to support a wide range of applications.

The RK3576 consumes 1.2 watts, making the Core-3576JD4 suitable for battery-powered devices and applications. This helps where long-term energy efficiency is required, particularly in continuous operation, such as remote monitoring systems and mobile applications.

The Core-3576JD4 Low-Power Large-Model AIOT Core Board is powered by the Rockchip RK3576 octa-core 64-bit processor. It supports high-resolution video processing (8K decoding and 4K encoding) and features a 16-megapixel Image Signal Processor. The board is available on Firefly, with prices ranging from $119 to $199. For more detailed specifications, please refer to the datasheet.

Please follow and like us: