Forlinx Embedded

Powering Intelligence at the Edge.

Visit at Hall C5, Booth 331 atelectronicaFair 2026 to explore our latest , development boards, FCUs, and embedded solutions.

We’ll showcase ™ IQ8 and IQ9 platforms, live Edge AI demos featuring multimodal AI, Digital Human, and multi-channel real-time recognition, alongside our broad portfolio of embedded products powered by NXP Semiconductors , Rockchip , Texas Instruments, and Allwinner.

We look forward to meeting you and discussing how Forlinx can support your next embedded project.

Book a meeting in advance
https://www.forlinx.net/article-contact.html

AGV to AMR: Powering the Next Generation of Mobile Robotics.

As industrial logistics shift from fixed-path AGVs (magnetic/QR code) to fully autonomous AMRs using 2D/3D SLAM and multi-sensor fusion, selecting the right computing architecture is critical. High-precision positioning, obstacle avoidance, and dynamic path adjustment demand cost-effective, high-compute platforms that minimize heat and power consumption.

Why ARM & RK Processors are leading the AMR transition:
🧠 RK3588 (8-Core A76+A55): High performance tailored for complex 3D SLAM, outdoor robotics, and unmanned forklifts (FMR).
💰 RK3576 (6-Core A72+A53): High cost-performance ratio for indoor warehouse AMRs and logistics automation.
🔀 Dual-Board Functional Split: Pair FET3588-C + FET3568-C/FET3576-C to separate high-level perception from real-time motion control.

🔗 https://www.forlinx.net/article-contact.html

🧠 Can 27B/31B LLMs Run Locally at the Edge?
Yes — with the right architecture.

Forlinx combines the OK3588-C with four RK1828 AI accelerator through PCIe, creating a compact edge inference platform for larger local language models.
https://www.forlinx.net/industrial-news/ok3588-c-rk1828-pcie-cascade-edge-llm-inference-846.html

With approximately 40W total accelerator power, the platform supports:
🔹 Qwen3.8-27B / Qwen3.5-27B
🔹 Gemma-4-31B-it
🔹 4-card PCIe cascading & pipeline parallelism
🔹 Hybrid quantization for reduced memory requirements
🔹 Local RAG and LoRA fine-tuning support

In practical testing, the 27B model handled embedded code generation and local document RAG, achieving generation speeds of around 13 tokens/s while keeping data entirely on-device.