Click the blue text above
Follow us to stay updated
On September 10, Arm Holdings Limited (hereinafter referred to as Arm) officially launched the new Arm® Lumex™ Computing Subsystem (CSS) platform, designed to provide robust edge AI support for flagship smartphones and next-generation personal computers. This fully integrated hardware, software, and physical implementation platform not only significantly enhances AI performance and energy efficiency but also aims to help partners shorten product development cycles and quickly bring innovative experiences to market.
Chris Bergey, Senior Vice President and General Manager of Arm’s Client Line of Business, emphasized that artificial intelligence (AI) is no longer just a technical feature; it has become the foundational support for the next generation of mobile and consumer technologies. Leveraging the Arm Lumex CSS platform, Arm continues to enhance edge AI experiences to meet the growing demands and expectations of users. He further revealed that Arm is extending its second-generation Scalable Matrix Extension (SME2) technology to every CPU platform, with expectations that by 2030, SME and SME2 technologies will add over 10 billion TOPS of computing power to more than 3 billion devices worldwide, resulting in an exponential leap in edge AI performance.
1Core Architecture and Performance Breakthroughs of the Arm Lumex CSS Platform
The Lumex CSS platform represents Arm’s latest achievement in the field of mobile computing. Its core components include the next-generation Armv9.3 CPU cluster, Arm Mali™ G1-Ultra GPU, flexible and efficient Arm C1-DSU, physical implementations optimized for the 3nm process node, and deep integration across the software stack, providing developers using the Arm KleidiAI software library with a seamless AI acceleration experience.
Among these, the most notable is the CPU cluster equipped with SME2 technology, which includes the Arm C1-Ultra and C1-Pro designed for flagship devices, the C1-Premium aimed at the upper mid-range market, and the energy-efficient C1-Nano. This multi-tiered design allows partners to flexibly balance performance, power consumption, and chip area based on different product positioning.
In practical applications, SME2 technology has shown outstanding performance. It can deliver up to five times the performance improvement for AI workloads, with a 4.7 times reduction in voice task latency and a 2.8 times increase in audio generation speed. For example, in real-time applications like smart yoga coaching, the text-to-speech generation speed has improved by 2.4 times, allowing users to receive instant feedback without worrying about device battery life.
Additionally, Arm’s collaboration with Alipay and vivo demonstrates that SME2 technology helps reduce response times for edge large language model inference by 40%, paving the way for real-time experiences in generative AI.
In addition to the significant upgrades to the CPU, the new generation Mali G1-Ultra GPU has also achieved remarkable breakthroughs in graphics and AI processing. Its second-generation ray tracing technology (RTUv2) doubles ray tracing performance, bringing desktop-level visual experiences to mobile gaming. At the same time, the Mali G1-Ultra has achieved a 20% performance improvement in AI inference tasks, further enhancing the responsiveness of real-time applications.
2Empowering Developers and Ecosystem Collaboration
Another major advantage of the Arm Lumex CSS platform is its high developer-friendliness. Through theKleidiAIsoftware library, developers can automatically leverage SME2’s acceleration capabilities without modifying code. Currently, KleidiAI has been integrated into mainstream AI frameworks such as PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime. This deep software integration significantly lowers the development threshold and accelerates the deployment of edge AI applications.
Cross-platform portability is also an important feature of Arm Lumex. Many applications under Google, including Gmail, YouTube, and Google Photos, fully support SME2, ensuring that devices equipped with Lumex can provide a seamless experience from the moment they hit the market. Additionally, AI features optimized for the Android system can be seamlessly migrated to Windows and other operating systems based on Arm architecture, further expanding application scenarios.
The support of ecosystem partners fully reflects the platform value of Arm Lumex. Weng Xindan, Technical Director of Alipay, stated that Alipay has completed LLM inference validation based on Arm SME2 technology on vivo’s new generation flagship smartphone. The results showed performance improvements of over 40% and 25% during the prefill and decode stages, respectively. This marks a significant breakthrough in CPU backend capabilities.
Vivo’s Senior Vice President and Chief Technology Officer, Shi Yujian, mentioned that vivo is the first terminal brand to establish a joint laboratory with Arm. As a technical partner of Arm, vivo not only collaborates with chip platform manufacturers but also pioneers the implementation of Arm’s advanced computing technology in smartphones, providing consumers with upgraded experiences. More importantly, both parties can bring the needs from real user scenarios into the technology’s origin, jointly initiating micro-architecture-level feature optimizations, thus leading the evolution of future chip technologies.
3Redefining Edge AI Experience
The launch of the Arm Lumex CSS platform profoundly reflects the development trend of edge AI: intelligence is migrating from the cloud to the device. Compared to cloud-first AI, edge AI has significant advantages in latency, cost, and privacy. SME2 technology enables complex AI tasks—such as neural camera noise reduction to achieve 120fps at 1080P resolution and 30fps at 4K resolution. Users can enjoy a smoother and more secure intelligent experience without sacrificing data privacy.
This shift not only affects smartphones but will also cover PCs, wearable devices, and other emerging AI terminals. Arm is expanding high-performance AI capabilities to smaller devices through efficient cores like the C1-Nano, truly realizing the vision of “AI everywhere”.
4Driving Industry Win-Win, Co-Creating the Future
SME2 has already been widely adopted by leading ecosystem partners, including Alibaba, Alipay, Samsung System LSI, Tencent, and vivo.
Many globally leading technology companies have also highly praised Arm technology.GoogleAndroid platform engineer Iliyan Malchev pointed out that thanks to the enhanced hardware from SME2 technology, more advanced AI models like Gemma 3 can run directly on various devices. With the continued expansion of SME2, mobile developers will be able to seamlessly deploy next-generation AI features across the entire ecosystem, ultimately allowing end users to enjoy low-latency AI experiences on their smartphones at any time.
Meta’s AI Collaboration Director Sy Choudhury also expressed excitement about the integration of Arm Kleidi with the ExecuTorch tool under the PyTorch framework, which will allow Meta’s applications to run seamlessly on the next-generation technology platform, significantly accelerating the experience for end users.
5Opening a New Starting Point for Intelligence
The Arm Lumex CSS platform is not only a product of technological iteration but also a foundational infrastructure for the AI era. It provides OEM manufacturers and developers with powerful tools through fully optimized hardware, open software support, and flexible configuration options, helping them achieve more personalized, efficient, and privacy-secure AI experiences at the edge. As devices equipped with Lumex are set to hit the market in the near future, mobile innovation is ushering in a new starting point, and a future driven by edge intelligence is already upon us.
Tap “View” to receive the latest news every day!