Arm has introduced its latest mobile platform, unveiling new CPU and GPU designs alongside redefined interconnect and memory management functionality, all tailored for the upcoming wave of AI-powered smartphones. Over recent years, Arm has shifted towards more integrated solutions, culminating in the Lumex compute subsystem (CSS), representing a significant evolution in this strategy.

In this iteration, Arm continues to optimize performance and power efficiency, boasting a 15% average improvement in CPU and a 20% enhancement in GPU, along with a 15% reduction in power consumption. Central to Lumex’s strategy is the integration of SME2 Scalable Matrix Extensions in the CPU cluster, promoted by Arm as the premier approach for AI acceleration, complemented by system-level optimizations to enhance scalability for devices running AI models.

Stefan Rosinger, Senior Director for CPUs, emphasizes that SME2 offers exceptional AI acceleration, delivering enhanced performance with minimal power use. Arm anticipates Lumex’s integration into smartphones and other devices imminently, with designs centered around 3nm manufacturing principles. Chips developed by Arm’s licensees are expected to reach clock speeds exceeding 4 GHz.

Arm’s streamlined approach designates all Lumex CPU cores as C1, distinguishing core models merely by performance tiers. This granular classification extends to GPU designs, as Lumex introduces the Mali G1, differentiated by shader core count.

Predicated on Arm’s new platform, the Lumex configuration initiates a fresh phase of chip design flexibility with four core types: C1-Ultra (maximum performance), C1-Premium, C1-Pro, and the power-efficient C1-Nano.

In summary, as Arm reinforces CPU-based AI, the choice between embracing SME2 or integrated NPUs in smartphones remains pivotal for licensees, potentially shaping future trends in mobile AI processing.