Uploaded November 2026 | Updated November 2026, 12 hours ago
Forlinx builds embedded computing platforms and edge AI hardware around silicon from Rockchip, Qualcomm, Texas Instruments, NXP and Allwinner. At electronica 2026, the demonstrations include local language model inference, multi-camera video analytics and depth perception across several processor and accelerator combinations. forlinx.net
A Rockchip RK3588 platform paired with RK1828 accelerators runs language model workloads locally, with PCIe connectivity for the accelerators; the interview describes configurations supporting up to five and models up to 26 billion parameters. Another RK3588 and RK1828 box handles up to 36 camera streams, with different AI models running across feeds simultaneously.
Qualcomm IQ8 and IQ9 platforms are shown running real-time video workloads. A multi-camera demo processes 36 channels, while another IQ9 setup detects and tracks people, extracts figure features and stores vectors for later search by tags. The interview cites 24 GB RAM for the IQ8 board and 36 GB for the IQ9 board, along with a high-speed connector on the board shown.
Other demonstrations cover a depth-perception camera combining radar and depth sensing, and Forlinx’s range of embedded platforms using Rockchip, TI AM62 and AM335x, NXP i.MX and Allwinner processors. The company describes use across robotics, industrial automation, smart city, energy storage and transportation, and mentions support for customers pursuing EU Cyber Resilience Act compliance and long-term product support.
The examples show how processor choice and accelerator expansion shape edge deployments, from offline language models to concurrent vision inference and searchable people tracking. Forlinx also highlights Rockchip NPU platforms and a newer RK3565-class option with an integrated NPU, aimed at bringing local inference to embedded systems.
Forlinx builds embedded computing platforms and edge AI hardware around silicon from Rockchip, Qualcomm, Texas Instruments, NXP and Allwinner. At electronica 2026, the demonstrations include local language model inference, multi-camera video analytics and depth perception across several processor and accelerator combinations. forlinx.net
A Rockchip RK3588 platform paired with RK1828 accelerators runs language model workloads locally, with PCIe connectivity for the accelerators; the interview describes configurations supporting up to five and models up to 26 billion parameters. Another RK3588 and RK1828 box handles up to 36 camera streams, with different AI models running across feeds simultaneously.
Qualcomm IQ8 and IQ9 platforms are shown running real-time video workloads. A multi-camera demo processes 36 channels, while another IQ9 setup detects and tracks people, extracts figure features and stores vectors for later search by tags. The interview cites 24 GB RAM for the IQ8 board and 36 GB for the IQ9 board, along with a high-speed connector on the board shown.
Other demonstrations cover a depth-perception camera combining radar and depth sensing, and Forlinx’s range of embedded platforms using Rockchip, TI AM62 and AM335x, NXP i.MX and Allwinner processors. The company describes use across robotics, industrial automation, smart city, energy storage and transportation, and mentions support for customers pursuing EU Cyber Resilience Act compliance and long-term product support.
The examples show how processor choice and accelerator expansion shape edge deployments, from offline language models to concurrent vision inference and searchable people tracking. Forlinx also highlights Rockchip NPU platforms and a newer RK3565-class option with an integrated NPU, aimed at bringing local inference to embedded systems.










