September 10, 2025
Story
Arm recently announced the release of its new Lumex , advanced compute subsystem ( CSS) platform, which the company claims is specifically designed for AI experiences on premium smartphones and next-generation PCs.
In a blog post announcing the new platform, Chris Bergey, SVP and GM of the Client Line of Business at , Arm, stated that Lumex combines the company’s best practices with Scalable Matrix Extension version 2 ( SME2 ),  , GPUs, and system IP to enable the ecosystem to accelerate the development of AI devices and deliver experiences from desktop class mobile gaming to smart assistants and personalized applications.
He even predicted that SME and SME2 will contribute more than 10 billion TOPS of determine to more than 3 billion devices by 2030, leading to an exponential increase in on-device AI capacity.  ,
Finger partners can choose how they build Lumex according to their needs, according to the release, allowing them to use time-to-market and time-to-performance advantages. Partner may also set up the platform RTL to match their specific layers and dry the cores themselves.  ,
The software combines  ,
- Next-generation SME2-enabled Armv9.3 CPU grouping, which powers lineup devices and includes , C1-Ultra, and , C1-Pro.
- New and , designed specifically for the sub-flagship industry and offering best-in-class region efficiency
- With the addition of innovative graphics and gaming, as well as a boost to AI performance, the is new and improved.
- C1-DSU , the most adaptable and power-aware DynamIQ Shared Unit ( DSU) Arm to date,
- optimized natural solutions for 3nm nodes ,
- Heavy integration across the application stack enables smooth AI acceleration for developers using and  .
With SME2-Enabled Computers, accelerated It everyday.
SME2-enabled Arm C1 CPU grouping apparently increases AI performance dramatically for real-world, AI-driven things:
- Up to 5x improvement in AI performance ,
- 4. 7x lower overhead for speech-based workloads ,
- 2. 8x faster sound generation ,  ,  ,
This increase in CPU AI determine makes for real-time, on-device AI assumption capabilities, Arm said, giving users better, faster experiences across tasks like sound generation, computer vision, and situational assistants.  ,  ,
Artificial experiences will only work on the Lumex program for developers. Apps immediately benefit from SME2 momentum thanks to the integration of KleidiAI across main frameworks, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime, with no script changing required.
Lumex is designed to add new flexibility, to:  , for developers building cross-platform apps.
- Google Photos, YouTube, and Email are currently SME2-ready, ensuring seamless integration as Lumex-based devices become available.
- Cross-platform portability means that optimizations created for Android may be easily extended to Windows on Arm and different platforms.
- Partner companies like Alipay are currently showcasing on devices LLMs that are running smoothly with SME2 ,  , and SME2.
Leading tech companies like Apple, Samsung, and MediaTek are incorporating AI motion abilities to create more swift, effective on-device AI. Samsung and MediaTek are enhancing the sensitivity and effectiveness of real-time AI software like translation, summaries, and personal assistants using Google Gemini, while Apple is powering Apple Intelligence.  ,
More sophisticated AI models, like Gemma 3, can run directly on a range of devices thanks to SME2-enhanced equipment, according to the author. SME2 will enable cellular developers to easily deploy the newest AI feature generation across ecosystems as it grows. According to Iliyan Malchev, Distinguished Software Engineer, Android at Google, this will eventually benefit end-users who have access to low-latency activities that are commonly available on their phones.
This is great news for consumers and mobile technology as well as for those looking to gain a competitive edge, so check out the entire post for all the details.  ,