IoT Platforms For On-Device Generative AI

Designed for smart home, retail, and industrial applications with edge-AI performance, multimedia support, and low power consumption.

MediaTek has announced two new IoT platforms, the Genio 720 and Genio 520, designed to bring next-generation generative AI capabilities to edge devices. These platforms are ideal for smart homes, retail, industrial, and commercial applications, combining high performance, advanced multimedia features, and energy efficiency.

Built on a 6nm process, both chipsets feature an octa-core CPU with two Arm Cortex-A78 and six Cortex-A55 cores, optimized for edge-AI computing. The integrated 8th-gen MediaTek NPU delivers up to 10 TOPS, accelerating transformer and CNN-based AI models directly on-device for tasks such as natural language processing and computer vision. With support for up to 16GB of LPDDR5 memory, the platforms can run demanding edge-optimized LLMs like Llama, Gemini, and DeepSeek.

– Advertisement –

MediaTek has also integrated support for NVIDIA TAO and popular AI frameworks, allowing developers to scale applications across a wide range of devices. “Genio 720 and 520 bring powerful and efficient AI to the edge, offering real-time performance with enhanced privacy,” said CK Wang, GM of MediaTek’s IoT Business Group.

Ideal for fanless, battery-powered devices, the platforms support a range of multimedia capabilities including dual 2.5K or ultrawide 4K/5K displays, dual ISP up to 32MP, and advanced camera support with MIPI virtual channels. Connectivity includes built-in Wi-Fi 6/6E, and support for Wi-Fi 7 and 5G RedCap via external modules.

The Genio 720/520 platforms are also OSM-ready with reference designs to simplify development and speed up time-to-market. MediaTek partners plan to release OSM-based modules later this year.

Developers can build on Android, Ubuntu, or Yocto Linux, with a rich selection of I/O and display interfaces to support multi-screen or interactive experiences.Sampling begins in Q2 2025. For more information, click here.