NextFin news, Arm, a leading semiconductor and software design company, announced on this Wednesday in Cambridge, UK, the launch of its Lumex Compute Subsystem (CSS) platform aimed at enhancing on-device artificial intelligence (AI) and gaming performance for consumer devices such as flagship smartphones and next-generation personal computers.
The Lumex platform integrates Arm's highest performing CPUs enabled with Scalable Matrix Extension version 2 (SME2), new GPUs including the Mali G1-Ultra, and system IP optimized for 3nm process nodes. This combination enables real-time AI use cases like voice assistants, voice translation, and personalization with up to 5 times faster AI performance compared to previous generations.
Chris Bergey, Senior Vice President and General Manager of the Client Line of Business at Arm, highlighted that Lumex achieves unprecedented six years of double-digit Instructions Per Cycle (IPC) performance gains for flagship devices. The platform's SME2-enabled Armv9.3 CPU cluster delivers up to 5x uplift in AI performance, 4.7x lower latency for speech workloads, and 2.8x faster audio generation, enabling smoother and faster AI-driven interactions on devices.
The new Mali G1-Ultra GPU, designed for gamers, offers a 2x improvement in ray tracing performance and 20% faster AI inference, supporting high-fidelity, console-class graphics on mobile devices. It also delivers 20% better graphics performance across popular gaming titles such as Fortnite and Genshin Impact.
Lumex provides architectural flexibility for partners to tailor the platform for different product tiers, from flagship peak performance CPUs like the C1-Ultra to power-efficient cores for wearables. The platform also includes the most flexible and power-aware DynamIQ Shared Unit (DSU) and optimized physical implementations for advanced semiconductor manufacturing nodes.
Developers benefit from seamless AI acceleration through KleidiAI libraries integrated into major AI frameworks and mobile operating systems, including PyTorch ExecuTorch, Google LiteRT, Alibaba MNN, and Microsoft ONNX Runtime. This integration allows AI applications to leverage SME2 acceleration without code changes, facilitating faster time-to-market for AI-enabled apps.
Arm's ecosystem partners such as Alibaba, Alipay, Samsung LSI, Tencent, and vivo are already adopting SME2 technology to deliver faster, more efficient on-device AI experiences. The platform supports a wide range of AI workloads including large language model inference, computational photography, generative AI, and real-time translation.
Unlike cloud-based AI solutions, Lumex enables AI processing directly on devices, reducing latency, enhancing privacy, and improving energy efficiency. This approach meets growing consumer expectations for instant, private, and always-available AI capabilities on mobile and consumer devices.
The announcement was made on this Wednesday, September 10, 2025, at Arm's headquarters in Cambridge, UK, marking a significant step forward in the evolution of on-device AI and mobile gaming technology.
Explore more exclusive insights at nextfin.ai.
