Sophia Windsor
[ Gearbest Technology News]Today, Arm officially launched a new generation of computing subsystem platform – Arm Lumex CSS. This platform is committed to providing more powerful AI computing power support for flagship smartphones and next-generation PC devices, bringing a rich end-side AI experience including intelligent voice assistants, real-time translation, personalized services, etc. It is reported that the Lumex platform has achieved significant performance breakthroughs in many aspects, especially in areas such as AI reasoning, energy efficiency management and graphics processing.
In terms of performance, the Lumex CSS platform relies on the Armv9.3 architecture CPU equipped with the second-generation Scalable Matrix Extension (SME2) technology to achieve up to five times improvement in AI performance. In actual application scenarios, the delay of voice tasks is reduced by 4.7 times, and the audio generation speed is increased by 2.8 times, which greatly improves the smoothness and response experience of end users. The platform also achieved six consecutive years of double-digit instructions per clock cycle (IPC) growth in a flagship device for the first time, demonstrating Arm’s continued leadership in processor microarchitecture design.
In addition to the major upgrade of the CPU, Arm also released a new generation of MaliG1-Ultra GPU, whose ray tracing performance is twice that of the previous generation, significantly improving the image quality and realism of mobile games. At the same time, the GPU also achieved performance gains of up to 20% in AI inference tasks, further enhancing the platform's comprehensive performance in complex visual and AI loads.
It is particularly noteworthy that Lumex is not a single IP combination, but a complete set of hardware and software collaboration solutions. Through the KleidiAI software library, developers can seamlessly call SME2 hardware acceleration capabilities in mainstream AI frameworks (such as PyTorch ExecuTorch, Google LiteRT, Alibaba MNN and Microsoft ONNX Runtime), and achieve performance improvements without modifying the code. This feature significantly lowers the threshold for AI application development and deployment, laying the foundation for the rapid maturity of the ecosystem.
In terms of actual implementation, Arm has also cooperated with a number of industry leaders. For example, in the large language model end-side inference project jointly promoted with Alipay and vivo, the response delay was successfully reduced by 40% with the help of SME2 technology, significantly improving the user experience. The Alibaba MNN team also pointed out that by deeply integrating SME2, it can support low-latency quantitative inference of billion-parameter-level large models on smartphones, demonstrating the key value of this technology in the next generation of mobile AI infrastructure.
In addition to performance, Lumex also offers a variety of options in terms of design flexibility. Partners can choose to directly adopt Arm's pre-integrated physical implementation solutions to shorten time to market, or they can customize configurations based on register transfer level (RTL) design to adapt to different markets and power consumption requirements. The platform covers multiple CPU configurations from the flagship C1-Ultra, the area-efficient C1-Premium, to the energy-efficient C1-Pro and ultra-low-power C1-Nano, covering a variety of product forms such as smartphones, PCs, and wearable devices.
Chris Bergey, general manager of Arm's terminal business unit, emphasized that AI is no longer an additional feature, but the core of the next generation of mobile technology. The launch of the Lumex platform is precisely in response to the comprehensive demand for higher computing power, lower latency, and stronger privacy protection capabilities on the device side. He predicts that by 2030, SME and SME2 technology will bring more than 10 billion TOPS of additional computing power to more than 3 billion devices.
From an industry perspective, the release of Arm Lumex further consolidates its core position in the mobile computing ecosystem. Google, Samsung, MediaTek, Tencent and other companies have stated that they are advancing relevant integration and optimization work. Especially in cutting-edge scenarios such as terminal-side large models, generative AI, and real-time rendering, Lumex is expected to become a key platform to promote terminal experience changes.
Overall, Arm Lumex CSS is not only a hardware iteration, but also marks that on-device AI is moving from experimental functions to system-level support. Through architectural innovation and ecological collaboration, it has laid a solid foundation for the next generation of consumer electronic devices to be “smarter, more efficient, and more personalized.” As terminals equipped with this platform are gradually launched, an AI era that truly belongs to the device itself is coming.
magicCubeFunc.write_ad(“dingcai_top_0”);
Login with Google