Mobile AI Features Evolve: Training LLM Models Directly on Smartphones
Article By : Anthea Chuang, EE Times Taiwan
MediaTek's Dimensity 9400 chipset enhances smartphones with advanced Edge AI, immersive gaming, and superior imaging, while offering improved energy efficiency and performance for a smarter mobile experience...
What if users could train Large Language Models (LLMs) directly on their smartphones, incorporating their personal characteristics? Could this spark a new “golden age” for smartphones, more than a decade after their initial debut?
The shift of artificial intelligence (AI) from cloud systems to edge devices has accelerated the growth of Edge AI. The arrival of generative AI models like ChatGPT and LLMs has ignited a wave of new AI-powered applications. However, the same challenges persist: edge devices, including smartphones and PCs, are working to overcome issues related to computing power and energy consumption, enabling LLMs to function efficiently on mobile devices.
Since AI-powered PCs made their debut, smartphones with integrated AI features have increasingly captured consumer interest. But is simply supporting LLMs on smartphones enough to meet user expectations? Could the ability to train LLMs directly on phones—embedding them with individual traits—usher in a new era for smartphones, similar to the one that followed their first release over ten years ago?
Gallium Nitride (GaN) Power Solutions
MediaTek has unveiled its next-generation Dimensity 9400 flagship chipset, designed to enhance AI experiences on smartphones by improving both performance and efficiency. JC Hsu, Senior Vice President of MediaTek, explained that the Dimensity 9400 uses a second-generation big-core architecture, combining an Arm v9.2 CPU, GPU, and NPU. The chipset is purpose-built for Edge AI, immersive gaming, and superior imaging, positioning it as a 5G Agentic AI flagship product.
Energy Efficiency and Performance
Built on TSMC’s second-generation 3nm process, the Dimensity 9400 delivers 40% lower power consumption compared to its predecessor. Hsu detailed that the second-generation big-core CPU architecture integrates one Arm Cortex-X925 core running at up to 3.62GHz, three Cortex-X4 cores, and four Cortex-A720 cores. This results in a 35% increase in single-core performance and a 28% increase in multi-core performance over the Dimensity 9300. Additionally, the chipset includes MediaTek’s 8th-generation AI processor (NPU 890) and the Dimensity Agentic AI engine. The NPU supports device-side LoRA training and the generation of high-quality images, enabling the Dimensity 9300 to enhance generative AI performance and provide developers with Agentic AI capabilities. This allows AI applications to evolve into autonomous, reasoning-driven, and action-oriented experiences.
These advanced features enable users to train AI models directly on their smartphones. Agentic AI applications learn from user habits, proactively suggesting responses and improving overall user experiences. To further develop a rich AI ecosystem, Hsu emphasized MediaTek’s collaboration with developers to create a unified interface for connecting AI agents, third-party apps, and models. This initiative streamlines AI operations between edge devices and cloud services, while reducing product development cycles.
MediaTek Dimensity 9400 chipset delivers innovative Edge AI, immersive gaming, and exceptional imaging experiences for users.
(Source: MediaTek)
Enhanced Edge AI and More
Beyond enhancing Edge AI, the Dimensity 9400 also offers significant upgrades in gaming, photography, and wireless connectivity. Hsu highlighted the integration of a 12-core Arm Immortalis-G925 GPU and a PC-grade Dimensity OMM ray-tracing engine, delivering an immersive gaming experience with realistic lighting effects. The flagship ISP, Imagiq 1090, supports full-range HDR, enabling smooth zooming and clear tracking of moving subjects. For wireless communication, the chipset’s 5G modem, based on 3GPP Release 17, supports dual SIM and dual data functionalities. Its 4nm Wi-Fi/Bluetooth combo chip boosts Wi-Fi 7 multi-link operation (MLO) to 7.3Gbps and extends coverage by up to 30 meters.
The Dimensity 9400 chipset supports foldable smartphones, offering manufacturers more design flexibility while bringing innovative Edge AI, immersive gaming, and enhanced imaging to users.
Balancing Performance and Battery Life
With generative AI requiring significant power, the enhanced AI features of the Dimensity 9400 raise concerns about battery life. Hsu reassured that despite the major performance boosts, the new chipset’s advanced manufacturing techniques and second-generation big-core architecture lead to improved energy efficiency. For example, LLM prompt processing performance improves by 80%, while power consumption is reduced by 35%. GPU performance increases by 41%, while power consumption is cut by 44%. Additionally, optimized photography and video processing reduces power consumption by 14%, ensuring a balance between performance and battery life.