Arm Introduces Lumex CSS Platform with SME2 Support and GPU Mali G1-Ultra for local AI and mobile gaming

Arm has announced a platform Lumex CSS, designed to accelerate AI on flagship and sub-flagship devices. It is based on new CPU with the support Scalable Matrix Extension 2 (SME2), including C1-Ultra, C1-Premium and C1-Pro, providing up to 5x AI performance boost and a significant reduction in delays in speech tasks. The platform is also integrated with KleidiAI and supports popular frameworks: PyTorch ExecuTorch, Google LiteRT, ONNX and MNN.

The key graphical component became new GPU Mali G1-Ultra with an advanced RTUv2 ray tracing unit. It provides 2x increase in ray tracing and up to 20% improvement in AI tasks. Performance in popular games like Genshin Impact and Fortnite has increased up to 20% compared to the previous generation.

The Lumex architecture allows OEMs to flexibly configure the SoC for different device levels, from flagships to wearables. The C1-Premium model focuses on compactness, saving up to 35% of the die area, while the C1-Nano is designed for energy-efficient solutions. Arm notes that SME2 will achieve 10 billion TOPS of computing on over 3 billion devices by 2030.

In addition to AI and graphics, Lumex provides integration with Android and Windows-platforms. SME2 support has already been implemented in Google apps and partner projects, including Alipay, Samsung, vivo and Tencent. Lumex CSS is positioned as the foundation of the next generation of local AI, combining high performance, privacy and energy efficiency in a scalable architecture.