Arm has officially launched the CSS for Mobile 2 (Computing Subsystem), a comprehensive platform designed to power the next generation of AI-native Android devices. While previous reports had already highlighted the potential of AI-integrated mobile GPUs, the full reveal of the subsystem provides concrete performance targets and architectural details for the upcoming 2027 flagship hardware.

Add AlexTech.ai asPreferred Source on Google

C2-Ultra: Iterative Gains and AI Boost

The centerpiece of the CPU cluster is the C2-Ultra core. According to FPGA simulations, Arm claims a 15% peak performance uplift in single-thread tasks and an average increase of 12% across traditional benchmarks compared to the C1-Ultra.

Architecturally, the C2-Ultra maintains a 10-wide decode and similar ALU configurations to previous generations, but focuses on critical efficiency bottlenecks. The performance gains stem from an expanded execution window and improved speculation through a refined branch predictor, reducing the time the CPU spends waiting for data. Furthermore, the integration of SME2 (Scalable Matrix Extension 2) allows for up to 1.7x higher AI model performance within the CPU cluster.

Mali G2-Ultra NX and Neural Graphics

The Mali G2-Ultra NX GPU represents a fundamental shift toward "AI-native" graphics. Unlike traditional mobile GPUs, it embeds neural accelerators directly into the shader cores. This allows AI workloads to run concurrently with standard graphics and compute tasks, utilizing a shared memory system and coherent caches to minimize data movement.

Key capabilities include:

  • Neural Super Sampling (NSS): Reconstructing high-resolution images from lower-resolution renders to reduce raw GPU load.
  • Ray Tracing: The third-generation ray tracing unit (RTUv3) reportedly reduces the computational cost of ray tracing workloads by up to 70%.
  • Early Adoption: The G2-Ultra architecture has already begun appearing in the wild, notably powering Xiaomi's XRING O3 chip launched in late August.

Beyond Mobile: CSS N4

While the focus remains on smartphones, Arm also introduced the Neoverse CSS N4 IP, targeting the datacenter space. This indicates a broader strategy to unify AI-native computing across both edge devices and cloud infrastructure, ensuring that the "Agentic AI" era is supported by a consistent architectural foundation from the server to the pocket.