Firefly AIO-3576JD4 Mainboard: Powered by Rockchip RK3576 with ARM Mali-G52 MC3 and 6 TOPS NPU for AI Applications

The Firefly AIO-3576JD4 mainboard is powered by the Rockchip RK3576, an octa-core 64-bit AIOT processor featuring a big.LITTLE architecture (4×A72 + 4×A53) and a maximum frequency of 2.2 GHz. Built with an advanced lithography process, it offers a balance of high performance and low power consumption. Integrated with an ARM Mali G52 MC3 GPU and a 6 TOPS NPU, the mainboard supports AI applications, including the private deployment of ultra-large language models such as Gemma-2B, ChatGLM3-6B, Qwen-1.8B, and Phi-3-3.8B under the Transformer architecture. Docker container management is also supported for improved deployment flexibility.

The AIO-3576JD4 supports 4K video decoding at 120fps (H.265/HEVC, VP9, AVS2, AV1) and 4K video encoding and decoding at 60fps (H.264/AVC, H.265/HEVC). Its built-in 16MP ISP offers low-light noise reduction, support for RGB-IR sensors, and AI-ISP technology, enhancing image quality with up to 120dB HDR. The board also supports 3 MIPI-CSI D-PHY inputs for advanced image processing.

Designed for industrial applications, the AIO-3576JD4 includes an external watchdog for reliability and offers a wide range of expansion interfaces. These include MIPI-CSI, USB 3.0, USB 2.0, HDMI 2.1, Mini PCIe, M.2, Type-C, RS485, RS232, CAN, TF Card, and SIM Card, meeting diverse peripheral and application needs.

The Firefly AIO-3576JD4 board specifications:

• • SoC: Rockchip RK3576
    • CPU: Octa-core 64-bit processor (4×A72 + 4×A53), up to 2.2GHz
    • GPU: ARM Mali-G52 MC3 @ 1GHz
      • Supports OpenGL ES 1.1/2.0/3.2, OpenCL 2.0, Vulkan 1.1
      • Built-in high-performance 2D acceleration
    • NPU: 6 TOPS
      • Supports INT4/8/16/FP16/BF16/TF32 mixed operations
  • ISP: 16MP
    • Supports low-light noise reduction, RGB-IR sensor, up to 120dB HDR
    • AI-ISP for enhanced image quality and reduced noise
  • VPU:
    • Decoding: 4K@120fps (H.265/HEVC, VP9, AVS2, AV1), 4K@60fps (H.264/AVC)
    • Encoding: 4K@60fps (H.265/HEVC, H.264/AVC)
  • Memory: LPDDR4/LPDDR4x (4GB/8GB optional)
  • Storage:
    • eMMC (16GB/32GB/64GB/128GB/256GB optional), UFS 2.0 (optional)
    • 1x M.2 (SATA3.0/PCIe NVMe SSD, supports 2242/2260/2280)
  • Networking:
    • 2x RJ45 (1000Mbps) Ethernet
    • WiFi 6 and BT5.2 via M.2 E-Key (2230), 2.4GHz/5GHz dual-band
    • 4G LTE and 5G expansion via Mini PCIe and M.2 slots
  • Video Input:
    • 2x MIPI CSI DPHY (1x 4 lanes or 2x 2 lanes)
    • 1x MIPI CSI D/C PHY (MIPI DPHY or MIPI CPHY)
  • Video Output: 1x HDMI 2.1 (4K@120fps)
  • Audio: 1x 3.5mm Audio jack (MIC recording, CTIA standard)
  • USB:
    • 2x USB 3.0, 1x USB 2.0
    • 1x Type-C (USB 2.0/DEBUG)
  • Expansions:
    • 1x FAN header (4Pin)
    • 1x SIM card slot
    • 1x dual-row pin header (USB 2.0, SPI, I2C, Line In/Out, GPIO)
    • 1x Phoenix connector (RS485, RS232, CAN 2.0)
  • Power Supply: 12V DC input (12V~24V wide input voltage)
  • Dimensions: 122.89mm x 85.04mm x 22.7mm
  • Weight: ≈120g
  • Operating Temperature: -20°C to 60°C
  • Storage Humidity: 10% to 90%RH (non-condensing)

The AIO-3576JD4 mainboard supports Android 14, Linux OS, and Buildroot as its operating systems. It is designed for private deployment of large-scale parameter models under the Transformer architecture, including models such as Gemma-2B, LlaMa2-7B, ChatGLM3-6B, and Qwen1.5-1.8B. Additionally, it supports traditional AI models like CNN, RNN, and LSTM, and works with deep learning frameworks such as TensorFlow, PyTorch, MXNet, PaddlePaddle, ONNX, and Darknet. The mainboard also offers Docker container management and custom operator development for enhanced software versatility.

The AIO-3576JD4 mainboard is compatible with mainstream edge computing modules through its 260-pin standard SODIMM interface. This ensures flexibility in combining and replacing modules, meeting customization needs for various edge computing deployment scenarios. Supported core boards include the Core-1688JD4, Core-3576JD4, and Core-3588JD4, as well as NVIDIA Jetson Orin Nano and Jetson Orin NX modules.

Previously, we covered Rockchip RK3576-based SBCs, mini PCs, and development boards such as the Mekotronics R57 fanless AI mini-PC, Boardcom CM3576 SoM and EM3576, and the Banana Pi BPI-M5 Pro. Feel free to explore those if you’re interested in learning more about this product lineup.

The Firefly AIO-3576JD4 is available in two configurations: 4GB RAM + 32GB storage for $199 and 8GB RAM + 64GB storage for $229. For more details, visit the product page or check out their wiki page.

Please follow and like us: