Arm's New Lumex CSS Platform, Driving Double-Digit Performance Gains (newsroom.arm.com)

Arm has unveiled its new Lumex CSS platform, delivering significant advancements in on-device AI capabilities for consumer devices. Central to Lumex is the SME2-enabled Armv9.3 CPU cluster, which offers up to 5x faster AI performance and dramatically reduces latency in speech and audio tasks, enabling real-time, private, and efficient AI inference without cloud dependence. This platform supports a diverse range of devices—from flagship smartphones with the powerful C1-Ultra cores offering double-digit IPC gains, to highly efficient C1-Premium and C1-Nano cores designed for sub-flagship and wearable markets. The new Mali G1-Ultra GPU complements these gains, doubling ray tracing performance and boosting AI inference speeds by 20%, empowering richer mobile gaming and AI workloads. Lumex’s tight software integration through KleidiAI allows developers to leverage SME2 acceleration seamlessly across major AI frameworks like PyTorch ExecuTorch, Google LiteRT, and Microsoft ONNX Runtime, enabling broad ecosystem support and faster time-to-market. Real-world tests with partners such as Alipay, vivo, and Tencent demonstrate up to 40% faster large language model responses and smooth AI-powered features like neural camera denoising at impressive frame rates. By combining scalable hardware configurability, cutting-edge 3nm physical designs, and robust ecosystem partnerships, Arm’s Lumex platform sets a new industry standard for delivering personal, private, and high-performance AI experiences directly on edge devices, heralding a transformative leap for mobile and consumer AI innovation.