News Highlights:
- The ARM Lumex CSS platform unlocks real-time on-device AI use cases such as assistants, voice translation, and personalization.
- Developers can access SME2 performance with Kleidiai, integrated into all major mobile OS and AI frameworks, including Pytorch Executorch, Google Litert, Alibaba MNN, Microsoft OnNX Runtime, and more.
- For flagship devices, the ARM Lumex CSS platform delivers unprecedented 6 years of double-digit IPC performance improvements
- The new MaliG1-Ultra redefines mobile entertainment, built for gamers, with double raytracing uplift
AI is no longer a feature, it is the foundation of next-generation mobile and consumer technology. Users expect instant, private, real-time assistance, seamless communication, or personalized content available on the device, without compromising. To meet these expectations, we need more than a gradual upgrade, and step changes that bring together performance, privacy and efficiency in a scalable way.
Introducing Arm Lumex
That's why we're introducing ARM Lumex, the most advanced computing subsystem (CSS) platform dedicated to accelerating the AI experience with flagship smartphones and next-generation PCs.
Lumex uses scalable matrix extension version 2 (SME2), GPUs, and system IP to integrate top-performing CPUs, enabling the ecosystem to market AI devices faster and deliver experiences from desktop-class mobile games to real-time translation, smarter assistants and personalized applications.
We have SME2 enabled on all CPU platforms, and by 2030, SME and SME2 will add over 10 billion top computing to over 3 billion devices, providing an exponential leap of AI capabilities on the device.
Partners can choose exactly how Lumex is built into SOC. They can take the platform as it is offered, leverage cutting-edge physical implementations to suit their needs, buy time to the market and spend time on performance benefits. Alternatively, partners can configure platform RTL for targeted layers and strengthen the core itself.
Our simplified naming conventions across Lumex and the ARM portfolio were announced earlier this year.
Platforms combine:
- The flagship device for CPU clusters and power supplies including next-generation SME2 compatible ARMV9.3 C1-Ultra and C1-Pro
- New C1-Premium built for the sub-flagship market, providing efficiency in class areas
- New MaliG1-Ultra GPU with next-generation ray tracing, boosting AI performance in addition to advanced graphics and gaming
- The most flexible and power-aware DynamiQ Shared Unit (DSU) arm has been delivered so far: C1-DSU
- Optimized physical implementation of 3NM nodes
- Deep integration across the software stack that provides seamless AI acceleration for developers using the Kleidiai library
Accelerated AI everywhere using SME2-enabled CPU
ARM C1 CPU clusters with SME2 support provide dramatic AI performance improvements for real-world AI-driven tasks.
- Up to 5 times higher with AI performance
- 4.7x low latency for voice-based workloads
- 2.8X Fast Audio Generation
This leap with CPU AI Compute allows on-device AI inference in real-time, providing users with a smoother and faster experience across interactions such as audio generation, computer vision, and context assistants.
So, what does this mean in real-world use cases? SME2 can offer a whole new level of responsiveness and efficiency. For example, the Smart Yoga Tutor demo app gave a 2.4x boost to speeches from text. This means that users get instant feedback on pauses without eliminating battery life. Together with Alipay and Vivo, we have reduced the time taken to LLM responses to interactions with users by 40%, demonstrating that SME2 offers faster, real-time generation AI-on-device.
SME2 is not just about speed. In addition, the AI-equipped functions that cannot match traditional CPUs have been unlocked. For example, neural camera removal is performed at over 120fps at 1080p, 30fps at 4K, all on a single core. This allows smartphone users to capture sharper crystal clear images, even in the darkest scenes, allowing for smoother interactions and richer experiences on everyday devices.
Unlike cloud-first AI, which is constrained by latency, cost and privacy concerns, Lumex brings intelligence directly to your device, faster, safer and always available. SME2 has been accepted by major ecosystem players such as Alibaba, Alipay, Samsung LSI, Tencent and Vivo.
Architectural freedom for all product layers
Lumex offers partners the freedom to balance peak performance, sustainable efficiency and silicon area for products ranging from high-end smartphones and PCs to the new AI first form factor.
| CPU | Important Benefits | Increased performance and efficiency | Ideal use case |
| C1-Ultra | Flagship Peak Performance | +25% Single Thread Performance Double-digit IPCs are earned year-over-year |
Large model inference, calculation photos, content creation, generation AI |
| C1-PREMIUM | C1-Ultra performance with improved area efficiency | Area 35% smaller than C1-Ultra | Sub-flagship mobile segments, voice assistants, multitasking |
| C1-Pro | Sustainable efficiency | +16% sustainable performance | Video playback, streaming reasoning |
| C1-NANO | Very power efficient | +26% efficiency, use less area | Wearable, minimal form factor |
Enable faster AI inference for desktop-class games and Mari GPUs
With over 12 billion ARM GPUs shipped so far, ARM is at the heart of the mobile gaming experience. The new arm Mali G1-Ultra GPU continues to push the boundaries of mobile gaming and offers high fidelity console-class graphics. This is possible with the latest ray trace unit V2 (RTUV2), which powers advanced lighting, shadows and reflections, leading to double ridges in ray trace performance compared to its predecessor. For AI workloads, G1-Ultra allows for up to 20% inference performance, increasing responsiveness across real-time applications.
The Mali G1-Ultra offers 20% better performance across graphics benchmarks compared to the previous generation thanks to a complete improvement in key titles including Arena Breakout, Fortnite, Genshin Impact and Honkai Star Rail. The G1-PREMIUM and G1-PRO GPUs provide superior performance and power efficiency for constrained devices.
Finally, developer-friendly AI for mobile use
For developers, AI simply works on the Lumex platform. Through Kleidiai integration through key frameworks such as Pytorch Executorch, Google Litert, Alibaba MNN, and Microsoft OnNX Runtime, apps automatically benefit from SME2 acceleration without changing code.
For developers building cross-platform apps, Lumex brings new portability.
- Google apps such as Gmail, YouTube and Google Photos are already SME2-enabled and Lumex-based devices are on the market, ensuring seamless integration
- Cross-Platform Portability means optimizations built for Android can seamlessly extend to Windows on arms and other platforms
- Partners like Alipay have already featured on devices LLM running efficiently on SME2
Technology leaders, including Apple, Samsung and MediaTek, integrate the AI acceleration capabilities of faster, more efficient on-device AI. Apple powers Apple Intelligence. Samsung and MediaTek are improving the responsiveness and efficiency of real-time AI applications such as translations, summaries and personal assistants using Google Gemini.
ARM Lumex: Platform-Level Intelligence in the AI Era
ARM Lumex is more than the most advanced CSS platform for the consumer computing market, and is the foundation of the next era of intelligent AI-enabled experiences. Whether you're an OEM or a developer, Lumex offers the tools to deliver personal, private, and high-performance AI at the most important edge. Built for the AI era, Lumex is where the future of mobile innovation begins.
Quotation support:
“Deep integration with SME2 allows for low latency quantized inference for 1 billion parameter models like Qwen on smartphones. Xiaotang Jiang, MNN, Taobao, Tmall Group, Head of Alibaba
“Verification of LLM inference using SME2 was completed on Vivo's next-generation flagship smartphone through close collaborations between ARM, Alipay and Vivo. We can see that prefill and decoding performance can be improved by more than 40% each. Xindan Weng, Head of Client Engineering at Alipay
“SME2-enhanced hardware allows more advanced AI models like the Gemma 3 to run directly on a wide range of devices. As SME2 continues to expand, mobile developers will be able to seamlessly deploy next-generation AI capabilities across ecosystems. Iliyan Malchev, a well-known software engineer, Google's Android
“Honor is about bringing a premium experience to more users, especially through the top of the mid-range smartphone. By leveraging the ARMLumexCSS platform, we can provide smooth performance, intelligent AI capabilities and excellent power efficiency that improves the everyday mobile experience.” honor
“AI is changing the way we interact with our devices and the world around us, and the ARM ecosystem is driving the critical development of this space. In Meta, we are excited about the integration of ARM Kleidi and Pytorch's executorch, ensuring applications can be seamlessly executed with next-generation technologies that accelerate the end-user experience.” Director of Sy Choudhury, AI Partnerships, Meta
“We are pleased to continue our collaboration with ARM by leveraging ARM's computing subsystem platform to develop the next generation of flagship mobile products. This partnership will push the boundaries of on-vice AI and provide a smarter, faster and more efficient experience for our users.” Nak Hee Seong, Vice President and Head of SOC IP Development Team at Samsung Electronics
“SME2 accelerates large language models on devices like Tencent's Hunyuan by addressing major performance bottlenecks and enabling efficient LLM deployments on mobile to enhance the user experience.” Felix Yang, renowned expert, machine learning platform, Tencent
