Uploaded June 2026 | Updated September 2026, 2 weeks ago
Radiantek, a recently launched subsidiary of Geniatech, demonstrated their local edge AI hardware platforms and custom engineering services at Computex 2026. The company focuses on ARM-based architectures and modular computing designs, offering custom hardware development, Linux optimization, and manufacturing services to support edge applications. Their showcase highlighted system-on-modules and single-board computers that enable local artificial intelligence workload processing without relying on cloud services.
Among the new hardware integrations is a Rockchip M.2 module featuring the RK3588 and RK1828 accelerators. This module supports local deployment of vision-language models, such as the Qwen 3.0 parameter model, performing inference locally with a response time of approximately two seconds. Because the model runs entirely offline, it eliminates internet dependency, protects data security, and removes token-based operating costs.
For larger deployments, the APC880 and APC880 Mini edge computing boxes pair the NXP i.MX 8M Plus system-on-chip with the Kinara Ara-2 (Ara240) coprocessor, delivering up to 42.3 TOPS of AI performance. The standard APC880 model features modular expansion and a comprehensive selection of interfaces, including HDMI output, USB Type-C, and dual Gigabit Ethernet ports, while the APC880 Mini offers a more compact form factor with similar processing capabilities. Additionally, the planned EdgeClaw agent boxes can scale performance by combining up to four M.2 modules to achieve 128 TOPS, enabling local inference for models up to 35 billion parameters such as Gemma or Nemotron.
The company's product roadmap supports a wide range of hardware platforms and NPU architectures, allowing developers to choose from different accelerators based on project requirements. Supported modules include Hailo NPUs, DeepX accelerators noted for temperature control, and MemryX systems. The modular designs conform to standard SMARC specifications, utilizing system-on-chips from NXP (i.MX 95), Qualcomm, MediaTek (Genio 520 and 720), and Renesas.
These edge computing systems are designed for real-world deployment in areas like smart healthcare and intelligent security. In nursing home environments, cameras connected to these edge boxes can monitor patient rooms, identifying incidents such as falls and generating immediate alerts for nursing staff. In security applications, the local processing allows cameras to identify threats, such as a weapon, and notify security personnel instantly to protect public safety.
Radiantek, a recently launched subsidiary of Geniatech, demonstrated their local edge AI hardware platforms and custom engineering services at Computex 2026. The company focuses on ARM-based architectures and modular computing designs, offering custom hardware development, Linux optimization, and manufacturing services to support edge applications. Their showcase highlighted system-on-modules and single-board computers that enable local artificial intelligence workload processing without relying on cloud services.
Among the new hardware integrations is a Rockchip M.2 module featuring the RK3588 and RK1828 accelerators. This module supports local deployment of vision-language models, such as the Qwen 3.0 parameter model, performing inference locally with a response time of approximately two seconds. Because the model runs entirely offline, it eliminates internet dependency, protects data security, and removes token-based operating costs.
For larger deployments, the APC880 and APC880 Mini edge computing boxes pair the NXP i.MX 8M Plus system-on-chip with the Kinara Ara-2 (Ara240) coprocessor, delivering up to 42.3 TOPS of AI performance. The standard APC880 model features modular expansion and a comprehensive selection of interfaces, including HDMI output, USB Type-C, and dual Gigabit Ethernet ports, while the APC880 Mini offers a more compact form factor with similar processing capabilities. Additionally, the planned EdgeClaw agent boxes can scale performance by combining up to four M.2 modules to achieve 128 TOPS, enabling local inference for models up to 35 billion parameters such as Gemma or Nemotron.
The company's product roadmap supports a wide range of hardware platforms and NPU architectures, allowing developers to choose from different accelerators based on project requirements. Supported modules include Hailo NPUs, DeepX accelerators noted for temperature control, and MemryX systems. The modular designs conform to standard SMARC specifications, utilizing system-on-chips from NXP (i.MX 95), Qualcomm, MediaTek (Genio 520 and 720), and Renesas.
These edge computing systems are designed for real-world deployment in areas like smart healthcare and intelligent security. In nursing home environments, cameras connected to these edge boxes can monitor patient rooms, identifying incidents such as falls and generating immediate alerts for nursing staff. In security applications, the local processing allows cameras to identify threats, such as a weapon, and notify security personnel instantly to protect public safety.
