Features of News
- Arm Lumex CSS system activates real-time on-device AI use situations like assistants, speech language and personalization, with fresh SME2-enabled Arm CPUs delivering up to 5x faster AI performance ,
- Developers can now use KleidiAI, which is now integrated into all major smart Apps and AI systems, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime, to obtain SME2 performance.
- Finger Lumex CSS program achieves unmatched six years of double-digit IPC efficiency gains for flagship devices.
- New Mali G1-Ultra transcends portable entertainment and is built for enthusiasts, with 2x ray tracing uplift ,
AI is no longer a element; it is the basis of the upcoming smart and customer technologies. Users then anticipate fast, private, and accessible content that is personalized, available on device, and without any compromise, real-time assistance, seamless communication, and seamless communication.  , Meeting these expectations requires more than progressive upgrades, it demands a step change that brings performance, privacy and efficiency together in a flexible way.  ,
Introducing Finger Lumex ,
That’s why we’re introducing , our most advanced compute subsystem ( CSS) platform, purpose-built to accelerate AI experiences on flagship smartphones and next-gen PCs.  ,  ,
Lumex combines Scalable Matrix Extension version 2 ( SME2 ), , and system IP, enabling the ecosystem to launch AI devices more quickly, offer immersive experiences from desktop-class mobile gaming to smarter assistants, personalized applications, and smarter assistants.  ,  ,
By 2030, SME and may increase over 10 billion TOPS of determine to more than 3 billion devices, bringing an exponential step in on-device AI potential. We are enabling across every CPU system.  ,
Partners can decide exactly how they build Lumex into their Device – they can get the system as delivered and utilize cutting-edge real implementations tailored to their needs, reaping time to market and time to performance benefits. Partner can also harden the cores themselves and set up the platform RTL for their desired tiers.  ,
Lumex and our simplified naming conventions across the Arm portfolio were announced  ,  ,
The platform combines  ,
- Next-generation SME2-enabled Armv9.3 CPU cluster including C1-Ultra and C1-Pro, powering flagship devices ,
- New C1-Premium, designed specifically for the sub-flagship market and offering best in class area efficiency ,
- New Mali G1-Ultra GPU with next-generation ray tracing improves gaming and graphics performance, and, finally, improves AI performance.
- The most flexible and power-aware DynamIQ Shared Unit ( DSU) Arm has delivered to date: C1-DSU ,
- optimized physical implementations for 3nm nodes ,
- Deep integration across the software stack ensures seamless AI acceleration for developers who use KleidiAI libraries .
Accelerated AI Everywhere with SME2-Enabled CPUs
The SME2-enabled Arm C1 CPU cluster increases dramatically in AI performance for demanding, AI-driven tasks in the real world:  ,
- Up to 5x improvement in AI performance ,
- 4.7x lower latency for speech-based workloads ,
- 2. 8 times faster than previous audio generation  ,  , ,
This increase in CPU AI compute makes for real-time, on-device AI inference capabilities, giving users smoother, faster experiences across tasks like audio generation, computer vision, and contextual assistants.  ,  ,
What does this mean in practice-based contexts, then? SME2 can offer a whole new level of efficiency and responsiveness. For example, our Smart Yoga Tutor demo app saw a 2.4x boost in text-to-speech, meaning users get instant feedback on their poses, all without draining battery life. We were able to reduce the time it took for LLM to respond to user interaction by 40 %, demonstrating that SME2 is delivering faster real-time generative AI on devices.  ,
SME2 isn’t just about speed, it’s also unlocking AI-powered capabilities that traditional CPUs can’t match. For instance, a neural camera’s denoising now operates at speeds of over 120 fps in 1080p or 30 fps in 4K, all at the same core. That enables smartphone users to capture clear, sharp images even in the most obscure of settings, resulting in smoother interactions and richer experiences on common devices.  ,
Lumex brings intelligence directly to the device, where it is faster, safer, and always available, in contrast to cloud-first AI, which is limited by latency, cost, and privacy concerns. Leading ecosystem players like Alibaba, Alipay, Samsung LSI, Tencent, and vivo are embracing SME2.  ,
Architectural Freedom for Every Product Category ,
Lumex gives customers the freedom to strike a balance between top performance, sustained efficiency, and silicon area in everything from high-end smartphones and PCs to the emergence of AI-first form factors:  .
| CPU | Key benefit , | gains in performance and efficiency  , | Ideal use cases , |
| C1-Ultra | Flagship peak performance , | + 25 % single-thread performance , Year-over-year IPC growth of double digits, |
Large-model inference, computational photography, content creation, generative AI , |
| C1-Premium | C1-Ultra performance with higher area efficiency , | area is 35 % smaller than C1-Ultra , | Sub-flagship mobile segments, voice assistants, multitasking , |
| C1-Pro | sustained effectiveness , | + 16 % increased performance | Video playback, streaming inference , |
| C1-Nano | Extremely energy-efficient | + 26 % efficiency, using less area , | Wearables, smallest form factors , |
Making Mali GPU ,  , and enabling desktop-class gaming and faster AI inference,
Arm is at the forefront of mobile gaming experiences with over 12 billion Arm GPUs currently being shipped. The new Arm Mali G1-Ultra GPU continues to push the boundaries of mobile gaming, delivering high-fidelity, console-class graphics. A brand-new Ray Tracing Unit (RTUv2 ), which powers advanced lighting, shadows, and reflections, increases the performance of ray tracing by 2x over that of its predecessor. The G1-Ultra increases responsiveness across real-time applications by up to 20 % faster inference performance for AI workloads.  ,
With across-the-board improvements for top titles like Arena Breakout, Fortnite, Genshin Impact, and Honkai Star Rail, the Mali G1-Ultra delivers 20 % better performance across graphics benchmarks than the previous generation. For device constraints, the G1-Premium and G1-Pro GPUs offer superior performance and power-efficiency.  ,
Finally, Developer-Friendly AI for Mobile ,
AI-based experiences are simply implemented on the Lumex platform for developers. Through the KleidiAI integration across major frameworks including , Google LiteRT, and , apps automatically benefit from SME2 acceleration with no code changed required.  ,  ,  ,
Lumex introduces new portability for developers creating cross-platform apps:  .
- Google apps like Gmail, YouTube and Google Photos are already SME2-ready, ensuring seamless integration as Lumex-based devices hit the market ,
- Cross-platform portability means that optimizations created for Android can be seamlessly extended to Windows on Arm and other platforms.
- Partner companies like Alipay are already showcasing on devices LLMs that are efficient with SME2 ,  ,
Technology leaders – including Apple, Samsung, and MediaTek – are integrating AI acceleration capabilities for faster, more efficient on-device AI. Samsung and MediaTek are enhancing the responsiveness and effectiveness of real-time AI applications like personal assistants and translation using Google Gemini, while Apple is supporting Apple Intelligence.  ,
Arm Lumex: Platform-Level Intelligence for the AI Era ,
Arm Lumex serves as the foundation for the upcoming era of intelligent AI-enabled experiences, not just our most cutting-edge CSS platform for consumer computing. Lumex gives OEMs and developers the tools to create high-performance, personal AI that is both private and effective where it matters most. Built for the AI era, Lumex is where the future of mobile innovation begins.  ,
embedded content ]
Supporting Quotes:  ,
” Arm and Alibaba’s joint innovation in scalable, next-generation mobile AI is made possible by low-latency, quantized inference for billion-parameter models like Qwen on smartphones,” according to MNN, who has deep integration with SME2 and MNN. Xiaotang Jiang, Head of MNN, Alibaba , Taobao and Tmall Group,
” The validation of LLM inference using SME2 has been completed on vivo’s next generation flagship smartphone through the close collaboration of Arm, Alipay and vivo. We observe that the performance of prefill and decode can be improved by over 40 % and 25 %, respectively. We are very pleased with the outcomes achieved so far, and these results demonstrate significant progress in the CPU backend.
” SME2-enhanced hardware enables more advanced AI models, like Gemma 3, to run directly on a wide range of devices. SME2 will enable mobile developers to seamlessly deploy the newest AI feature generation across ecosystems as it grows. This will ultimately help end-users who have access to low-latency applications that are common on smartphones. Iliyan Malchev, Distinguished Software Engineer, Android at Google ,
” At Honor, our goal is to provide premium experiences to more users, particularly with our mid-range smartphones. We can deliver smooth performance, intelligent AI features, and outstanding power efficiency by utilizing the Arm Lumex CSS platform, which elevates the quality of our daily mobile experiences. Honor ,
The Arm ecosystem is driving significant developments in this area, and AI is changing how we interact with our devices and the world around us. We at Meta are excited to see how Arm Kleidi and PyTorch’s ExecuTorch can seamlessly run our applications on cutting-edge technology that makes end-user experiences faster. Sy Choudhury, Director, AI Partnerships, Meta ,
” At Samsung, we’re excited to keep working with Arm by using the Arm compute subsystem platform to create the next-generation of flagship mobile products,” said the company. By collaborating, we can push the boundaries of on-device AI, delivering users with more intelligent, efficient, and quick experiences. Nak Hee Seong, Vice President and Head of SOC IP Development Team at Samsung Electronics ,
SME2 addresses key performance bottlenecks and facilitates effective LLM deployment on mobile for improved user experiences, accelerating on-device large language models, like Tencent’s Hunyuan. Felix Yang, Tencent ’s Distinguished Expert, Machine Learning Platform,