News Highlights
• The Arm Lumex CSS platform supports real-time edge AI use cases, covering smart assistants, voice translation, and personalized services; leveraging the new SME2 technology integrated into the Arm CPU, this platform can achieve up to five times the AI performance boost.
• Developers can leverage the performance advantages brought by SME2 technology through KleidiAI; currently, KleidiAI is integrated into all major mobile operating systems and AI frameworks, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime.
• For flagship devices, the Arm Lumex CSS platform has achieved unprecedented milestones—delivering double-digit instruction per clock (IPC) performance improvements for six consecutive years.
• The new Mali G1-Ultra, designed for gamers, offers double the ray tracing performance, redefining the mobile entertainment experience.
Arm Holdings today announced the launch of the new Arm® Lumex™ Compute Subsystem (CSS) platform, an advanced computing platform designed to accelerate AI experiences for flagship smartphones and next-generation personal computers. The Lumex CSS platform integrates the highest performance Arm CPU, GPU, and system IP featuring the second-generation Scalable Matrix Extension (SME2) technology, enabling ecosystem partners to bring AI devices to market faster while supporting a rich array of experiences such as desktop-level mobile gaming, real-time translation, smart assistants, and personalized applications.
Chris Bergey, Arm’s Senior Vice President and General Manager of the Client Line of Business, stated: “AI is no longer just a technical feature; it has become the foundational base for the next generation of mobile and consumer technology. With the Arm Lumex platform, we continuously enhance edge AI experiences to meet the growing demands and expectations of users. To this end, we are actively expanding SME2 technology across every CPU platform. By 2030, SME and SME2 technologies are expected to add over 10 billion TOPS of computing power to more than 3 billion devices, bringing an exponential leap in edge AI performance.”
Partners can flexibly choose how to utilize Arm Lumex to create their system-on-chips (SoCs). For example, they can directly adopt the platform delivered by Arm and leverage advanced physical implementation solutions customized for their needs, thus gaining the dual advantages of reduced time to market and rapid realization of performance value; alternatively, partners can configure the platform’s register transfer level (RTL) design according to their target market and complete the hardening of core modules themselves.
The new Arm Lumex platform includes the following core components
• Next-generation Armv9.3 CPU cluster featuring SME2 technology: including Arm C1-Ultra and Arm C1-Pro, supporting flagship devices
• Arm C1-Premium: designed for the upper mid-range market, offering top-notch area efficiency
• Arm Mali™ G1-Ultra GPU: equipped with next-generation ray tracing technology, enhancing overall AI performance while delivering advanced graphics and gaming experiences
• Arm C1-DSU: Arm’s most flexible, energy-efficient, and multi-power mode DynamIQ Shared Unit (DSU) to date
• Physical implementation optimized for the three-nanometer process node
• Deep integration across the software stack, providing a seamless AI acceleration experience for developers using the KleidiAI software library
CPUs with SME2 technology: Enabling AI acceleration across all scenarios
The Arm C1 CPU cluster with SME2 technology brings significant AI performance improvements for AI-driven tasks in real-world scenarios, including:
• AI performance boost of up to five times
• Voice workload latency reduced by 4.7 times
• Audio generation speed increased by 2.8 times
The leap in AI computing capability of the new Arm CPU makes real-time edge AI inference a reality, providing users with a smoother and faster experience in various interactive scenarios such as audio generation, computer vision, and contextual assistants.
In practical scenarios, SME2 technology elevates responsiveness and operational efficiency to new heights. For example, in the “Smart Yoga Coach” demo application, thanks to SME2 technology, the text-to-speech generation speed improved by 2.4 times, meaning users can receive instant feedback and guidance on their poses without worrying about battery life. Additionally, through a collaboration between Arm, Alipay, and Vivo, the interaction response time of large language models (LLMs) was reduced by up to 40%, fully demonstrating that SME2 can provide a faster real-time generative AI experience for end devices.
The value of SME2 technology lies not only in speed improvements but also in unleashing AI-driven functionalities that traditional CPUs cannot achieve. For instance, running a neural camera denoising function on a single core with SME2 can achieve frame rates exceeding 120 frames per second (fps) at 1080P resolution or 30 fps at 4K resolution. This allows smartphone users to capture sharper and clearer images even in the darkest scenes, resulting in smoother operational interactions and richer user experiences on everyday devices.
Unlike cloud-first AI, which faces challenges such as latency, cost, and privacy issues, Lumex brings intelligence to the device side, enabling faster, safer, and always-available smart experiences locally. SME2 has already been widely adopted by leading industry ecosystem partners, including Alibaba, Alipay, Samsung System LSI, Tencent, and Vivo.
Empowering Architecture Freedom Across Product Tiers
Arm Lumex provides partners with ample freedom to achieve a flexible balance between peak performance, sustained energy efficiency, and chip area across various products, from high-end smartphones and PCs to emerging AI-first devices.
Mali GPU: Delivering Desktop-Level Gaming Experience and Faster AI Inference
With the cumulative shipment of chips featuring Arm GPUs surpassing 12 billion units, Arm continues to maintain a core position in the mobile gaming experience domain. The new Arm Mali G1-Ultra GPU further breaks the performance boundaries of mobile gaming, providing mobile gamers with high-fidelity, console-level graphics. This breakthrough is attributed to the new second-generation ray tracing (Ray Tracing Unit v2, RTUv2) technology, which significantly enhances lighting, shadow, and reflection effects, doubling the ray tracing performance compared to the previous generation. In terms of AI workloads, the Mali G1-Ultra can boost inference performance by up to 20%, significantly enhancing the responsiveness of various real-time applications.
In various graphics benchmark tests, the Mali G1-Ultra achieved a 20% performance improvement over its predecessor, providing comprehensive performance enhancements for mainstream games such as “Escape from Tarkov: Infinite,” “Fortnite,” “Genshin Impact,” and “Honkai: Star Rail.” Meanwhile, the Mali G1-Premium and Mali G1-Pro GPUs also offer superior performance and energy efficiency for resource-constrained devices.
Developer-Friendly Edge AI Solutions
Developers can access an out-of-the-box AI development experience on the Arm Lumex platform. Through the integration of KleidiAI with major frameworks, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime, developers can automatically leverage SME2’s acceleration capabilities without modifying any code.
For developers needing to build cross-platform applications, Lumex brings new portability:
• Google applications such as Gmail, YouTube, and Google Photos now fully support SME2, allowing devices equipped with Lumex to run seamlessly upon launch.
• Cross-platform portability means that optimizations built for Android can seamlessly extend to Windows and other operating systems that use Arm architecture.
• Partners like Alipay have successfully validated that, leveraging SME2 technology, large language models (LLMs) can run efficiently on end devices.
Leading tech companies such as Apple, Samsung, and MediaTek are actively integrating AI acceleration capabilities to drive edge AI towards faster and more efficient directions. Among them, Apple is leveraging this capability to provide core support for “Apple Intelligence”; Samsung and MediaTek are enhancing the responsiveness and operational efficiency of real-time AI applications such as translation, summarization, and personal assistants using Google Gemini.
Arm Lumex: Platform-Level Intelligence for the AI Era
Arm Lumex is not only Arm’s most advanced CSS platform for the consumer computing market but also the cornerstone for unlocking a new era of intelligent AI experiences. Whether OEM manufacturers or developers, Lumex provides the necessary tools to help achieve personalized, privacy-preserving, and high-performance AI experiences in critical edge scenarios. As a platform designed for the AI era, Arm Lumex will become a new starting point for future mobile innovation.
Partner Testimonials
“Through deep integration with SME2, MNN can now provide low-latency quantization inference capabilities for large models with billions of parameters, such as Tongyi Qianwen, on smartphones. This achievement fully demonstrates the innovative strength of Arm and Alibaba in jointly promoting the development of next-generation scalable mobile AI technology.”
Jiang Xiaotang, Head of MNN at Alibaba Taotian Group Business Technology
“In close collaboration between Arm, Alipay, and Vivo, Alipay has completed the verification of large language model inference based on Arm SME2 technology on Vivo’s next-generation flagship smartphone. The results show performance improvements of over 40% and 25% during the prefill and decode phases, respectively. This marks a significant breakthrough in CPU backend capabilities. We are encouraged by the results achieved so far. We believe that with the continued evolution of xNN and deep integration with SME2, Alipay is expected to further unleash the potential of AI in more scenarios, building a smarter, safer, and more personalized service ecosystem.”
Weng Xindan, Head of Terminal Technology at Alipay
“Thanks to the enhanced hardware with SME2 technology, more advanced AI models like Gemma 3 can run directly on various devices. As SME2 continues to expand, 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.”
Iliyan Malchev, Distinguished Software Engineer at Google Android Platform
“AI is reshaping the way we interact with devices and the world around us, and the Arm ecosystem is driving key technological breakthroughs in this field. At Meta, we are excited about the integration of Arm Kleidi with the ExecuTorch tool under the PyTorch framework, which will allow our applications to run seamlessly on next-generation technology platforms, significantly accelerating the experience for end users.”
Sy Choudhury, Director of AI Partnerships at Meta
“At Samsung, we are pleased to continue our collaboration with Arm to develop the next-generation flagship mobile products based on the Arm Lumex CSS platform. This collaboration enables us to continuously push the boundaries of edge AI, delivering smarter, faster, and more efficient experiences for users.”
Nak Hee Seong, Vice President and Head of SOC IP Development Team at Samsung Electronics
“SME2 technology effectively helps large language models to be deployed on mobile devices, such as Tencent’s mixed Yuan model, providing strong acceleration for edge large language models, ultimately delivering a better user experience.”
Yang Xiaofeng, Expert at Tencent Machine Learning Platform
Friendly Reminder
MCers please note that due to adjustments in the recommendation mechanism of WeChat public accounts, if you find it difficult to see articles pushed by the Microcomputer public account recently, but do not want to miss the exciting review content of the microcomputer, you can take a moment to set Microcomputer asstarpublic account!
