Arm CSS for Mobile 2 Targets Agentic AI and Neural Graphics

Arm has unveiled CSS for Mobile 2, a compute platform engineered specifically for the demands of agentic AI and neural-enhanced graphics on mobile devices. The platform represents a shift from treating AI as isolated features to supporting persistent AI agents that coordinate models, maintain context, and act autonomously within smartphone power constraints.

Neural accelerators integrated into Mali GPU

The centerpiece GPU, Mali G2-Ultra NX, marks the first time Arm has embedded dedicated neural accelerators directly into its Mali architecture. By placing neural and traditional graphics processing in the same pipeline, the design enables AI-enhanced rendering techniques—reconstruction, enhancement, and detail generation—to run where graphics workloads already execute.

Arm reports the GPU delivers up to 4x higher performance per watt for neural graphics workloads compared to previous generations, while maintaining a 14% performance improvement on conventional game content. The architecture also introduces an updated execution engine and next-generation Ray Tracing Unit designed for complex modern game engine demands.

Game developers are already integrating the technology. NetEase plans to release Where Winds Meet with neural super sampling this year, while Tencent Games and Unity China have incorporated Arm Neural Technology into their respective development engines.

CPU cluster doubles down on SME2 for agent orchestration

The C2 CPU cluster pairs Arm's highest-performance C2-Ultra core with efficiency-focused C2-Pro cores, now equipped with two SME2 units—double the previous generation. This configuration positions the CPU as the orchestration layer for agentic AI, managing context, scheduling workloads, and coordinating across the broader compute system.

The doubled SME2 capability enables a 70% speedup on recent small language models. Compared to the C1-Ultra, the C2-Ultra delivers up to 1.7x higher AI performance and 15% better single-thread performance while consuming up to 38% less power at equivalent performance levels.

SME2 has gained adoption across the Android and iOS ecosystems, with implementations from Alipay, Google AI Edge Gallery, OPPO, and vivo.

Why it matters

The architecture signals a fundamental shift in mobile computing: AI is no longer a discrete accelerator problem but a system-level challenge. As AI agents move from executing single inferences to managing persistent workflows—reasoning, planning, acting—the entire compute stack must adapt. By integrating neural capabilities into both GPU and CPU paths, Arm is betting that future mobile experiences will blur the line between traditional compute and AI processing, requiring hardware designed from the ground up for that convergence.

Developer ecosystem and availability

Arm is supporting the platform with KleidiAI, the Neural Graphics Development Kit, and a new AI Portal that provides a centralized access point for optimized models and tools. The company positions CSS for Mobile 2 as a complete system solution rather than individual components, targeting silicon partners seeking faster paths to differentiated products.

These details were first reported by Arm in its official announcement.