/PRZWT/At a time when the tide of technology is surging forward, artificial intelligence (AI) has long transcended the scope of pure technical functions and has become the core cornerstone of the next generation of mobile and consumer technologies. At present, users' needs for the device side are showing a trend of diversity and high requirements. They are eager to get instant and accurate real-time assistance directly from the device side, realize seamless communication, enjoy personalized content, and have strict requirements for privacy protection and performance stability. Obviously, traditional incremental upgrades are no longer able to meet these needs. The industry urgently needs a fundamental change to perfectly integrate performance, privacy and efficiency in a scalable way, and open a new technological chapter.
On September 10, 2025, Arm officially announced the launch of the new Arm ® Lumex ™ Compute Subsystem (CSS) platform at its new AI Technology Leadership Summit Arm Unlocked 2025 Shanghai Station, an advanced computing platform designed to accelerate its AI experience for flagship smartphones and next-generation PCs.
The Lumex CSS platform integrates the highest-performance Arm CPUs, GPUs, and system IP with the second-generation Scalable Matrix Extension (ᴗ 2) technology, enabling ecosystem partners to bring AI devices to market faster. It also supports rich experiences such as desktop mobile gaming, real-time translation, intelligent assistants, and personalized applications. Currently, Arm is actively expanding the ᴗ 2 technology to every CPU platform.
It is expected that by 2030, SME and ᴗ 2 technologies will add more than 10 billion TOPS of computing power to more than 3 billion devices, bringing an exponential leap in end-side AI performance.
The Arm C1 CPU cluster with ᴗ 2 brings significant AI performance improvements to AI-driven tasks in real-world scenarios: AI performance can be improved by up to five times, voice workload latency can be reduced by 4.7 times, and audio generation speed can be increased by 2.8 times, providing more powerful support for diverse AI-driven tasks.
Chris Bergey, senior vice president and general manager of Arm's end point business unit, said that Arm has fully enabled ᴗ 2 technology on all CPU platforms, bringing stronger AI performance, lower memory usage, and smoother end-side AI operation to developers. The launch of ᴗ 2 marks a new era of more developer-friendly mobile end AI development.
The value of ᴗ 2 is even more reflected in the actual experience of the user's daily application. Thanks to ᴗ 2, the response speed and efficiency of the "Smart Yoga Instructor" application have now been improved to a new level. The text-to-speech performance has been improved by 2.4 times, and users can get instant posture guidance feedback without worrying about battery life. In addition, through the tripartite cooperation of Arm, Alipay and vivo, the interaction response time of the Large Language Model (LLM) has been reduced by up to 40%, which fully proves that ᴗ 2 can bring a faster real-time generative AI experience to end point devices.
The Sche2 not only focuses on speed improvement, but also unleashes the unparalleled AI-driven power of traditional CPUs. For example, running neural camera noise reduction on a single core equipped with the Sche2 can achieve frame rates of over 120 frames per second (fps) at 1080P resolution, or 30fps at 4K resolution. This allows smartphone users to capture sharper, clearer images even in the darkest of scenes, resulting in smoother operation and richer experience on daily devices.
Unlike cloud-first AI, which is challenged by latency, cost, and privacy concerns, Lumex brings intelligence to the device, enabling faster, more secure, and ready-to-use intelligent experiences locally. ᴗ 2 has been widely adopted by industry-leading ecosystem partners, including Alibaba, Alipay, Samsung System LSI, Tencent, and vivo.
On the Lumex platform, developers can enjoy a ready-to-use AI development experience. Through KleidiAI's integration with major frameworks, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime, developers can automatically get the acceleration of ᴗ 2 without modifying any code. Google apps such as Gmail, YouTube, and Google Photos now fully support ᴗ 2, ensuring that Lumex-powered devices can run seamlessly once they are on the market.
Alipay and other partners have launched multiple end-to-end large language models (LLMs) based on the efficient operation of the SF2 platform.
The launch of Arm's Lumex CSS platform this time has undoubtedly injected strong impetus into the development of end-side AI. The new Arm CPU enabled with the ᴗ 2 technology not only makes the device achieve a qualitative leap in performance, but also sets the industry benchmark in privacy protection and energy efficiency optimization. It is foreseeable that the wide application of the ᴗ 2 technology will promote the smart phone towards a smarter and more efficient AI experience.
