Arm has revealed its Lumex mobile platform, integrating new CPU and GPU designs along with rearchitected interconnects and memory management tailored for AI-driven smartphones. Emerging from its evolution towards more integrated solutions, Lumex showcases Arm’s philosophy of amalgamating performance and power efficiency. The platform delivers a 15% boost in CPU and 20% in GPU performance while curbing power usage by 15%.

Emphasizing Arm’s SME2 Scalable Matrix Extensions, Lumex aims to be the frontrunner in AI acceleration within CPU clusters. System-level optimizations cater to scalable AI model execution, while allowing OEMs to choose from four CPU core types—C1-Ultra, C1-Premium, C1-Pro, and C1-Nano—each tailored to specific performance and power efficiency needs. Premium devices are expected to leverage these cores for optimal operation.

Lumex also features the Mali G1 GPU, available in three variants: Ultra, Premium, and Pro, differentiated by shader cores ranging from one to over ten. The GPU’s high-tier model introduces a redesigned Ray Tracing Unit boosting performance by 40%, enhancing gaming and graphical processing.

Looking to future-proof AI processing capabilities, Lumex incorporates purpose-built System Interconnects (SI) and Memory Management Units (SMMUs), optimized for AI workloads. Arm’s software enhancements ensure developer frameworks like PyTorch and ONNX efficiently utilize SME2 accelerations.

While the industry’s direction on AI processing units varies, Arm champions CPU-based AI as the universally compatible choice for mobile devices, amid debates on integrating distinct NPUs. Analysts recognize the merit of Arm’s strategy amid a diverse chipset ecosystem, foreseeing a gradual standardization in AI processing architectures.