A fanless box built for harsh industrial sight
Forlinx has rolled out the FCU3101, a compact fanless edge AI system that wraps Rockchip’s RV1126B/RV1126BJ system-on-chip in a rugged gateway form factor. CNX Software reports this is the first complete industrial edge box based on that silicon the publication has featured, moving beyond the usual SBC and module designs that have used the same NPU.
At its core, the RV1126B drives a quad-core Arm Cortex-A53 processor up to 1.6 GHz and a Rockchip NPU rated at 3 TOPS for INT8 inference. It supports mixed precision—INT4, INT8, INT16, and FP16—and common frameworks including TensorFlow, ONNX, PyTorch, and Caffe. Video encoding and decoding capabilities include H.265/H.264 up to 4K at 30 fps decode and 12 Mbps encode at 30 fps, which is plenty for inspection cameras.
The box itself measures 202 x 123 x 42 mm and accepts 9V to 36V DC via a terminal block, with reverse polarity and overcurrent protection. Two temperature grades are offered: extended commercial (-20°C to +85°C) and full industrial (-40°C to +85°C). Memory options reach 4GB DDR4, storage goes up to 64GB eMMC with a microSD slot supporting up to 128GB.
I/O that speaks factory and security languages
Connectivity is the real differentiator. The FCU3101 includes dual Ethernet (one Gigabit, one 10/100 Mbps), Wi-Fi 4, Bluetooth 4.2, and optional 4G LTE Cat 1 through a Nano SIM slot. Serial interfaces are plentiful: two RS485 ports with selectable 120Ω termination via internal DIP switch and two RS232 ports. Industrial controls appear as two digital inputs for dry contacts and two relay outputs with 5A/30VDC normally open contacts. One USB 3.0 host and one USB 2.0 host, HDMI up to 1080p, and a 3.5mm audio jack round out the panel. A built-in RTC with battery, hardware watchdog, reset button, and a programmable light bar add practical touches.
Pre-trained vision models and the path to deployment
Forlinx ships the system with Linux and a set of ready-to-use AI algorithms aimed squarely at vision-heavy IIoT and smart security roles. The pre-loaded models cover:
- Personnel and safety detection
- Smoke and fire recognition
- Intrusion detection
- License plate recognition
- Defect inspection
- Material counting
Custom algorithm development is also available, and the company points to previous experience with a Linux 6.1.141 BSP on related RV1126 modules, though an exact kernel version for the FCU3101 isn’t published.
Pricing is not listed; Forlinx asks interested customers to contact them directly. Sample lead time is 5 working days, bulk orders about 6 weeks, and the company expects CE, FCC, and RoHS certifications by Q4 2026. For portable AI projects that need a hardened, mountable inference node rather than a bare board, the FCU3101 fills a useful gap.