During the 2026 World Artificial Intelligence Conference (WAIC), Rockchip debuted at booth H2-A325 under the theme “AIoT 2.0: Defining Next-Gen Smart Hardware,” showcasing its full-stack RK35xx host + RK1828 on-device AI coprocessor dual-chip collaborative solution, demonstrating comprehensive software-hardware ecosystem capabilities for on-device large models.
On the technical adaptation front, Rockchip completed deep底层 optimization for Gemma4, Qwen3.5, and Qwen3.5-Omni, enabling smooth local execution of mainstream 2B-7B models. Measured data shows Gemma4 first-token latency as low as 169.93ms, with outstanding multimodal interaction and agent execution performance. This performance level stands as a benchmark for mass production deployment in the on-device large model赛道.
The exhibition was organized around three pillars: large model底层 adaptation, full-scenario on-device AI agents, and mass-production terminal deployment. Leveraging the RK182X series coprocessor, Rockchip built a complete industrial chain from chip to model to terminal, demonstrating mature landing solutions including AI NAS, AI NVR, audio applications, industry applications, and in-vehicle AI Box, along with partially mass-produced products.
From an industry perspective, Rockchip’s on-device AI layout reflects the foresight of its “host SoC + AI coprocessor” dual-track strategy. Host chips (RK3588/RK3576, etc.) manage system operations, multimedia processing, and interface expansion, while coprocessors (RK1828, etc.) handle large-model inference, collaborating via high-speed interfaces. This architecture ensures system stability and compatibility while enabling modular AI capability upgrades.
For embedded solution providers, the ecosystem maturity demonstrated by Rockchip at WAIC means the technical risks of on-device AI solutions have been significantly reduced, with a clearer path to productization. Junyi Microelectronics’ core board products based on the RK3588/RK3576 + RK1828 dual-chip architecture can provide customers with one-stop support from hardware design to model deployment, accelerating time-to-market for on-device AI terminals.
Source: Rockchip Official Website