Arm has launched its second-generation mobile terminal computing subsystem, CSS for Mobile 2, which features the Mali G2-Ultra GPU integrated with a dedicated neural network accelerator for the first time.

Author: Zheng Minfang

Source: Wall Street Journal

Arm launches CSS for Mobile 2 in Shanghai

On September 8, Arm launched the second-generation mobile computing subsystem, CSS for Mobile 2, at the Arm Everywhere China annual event in Shanghai. The most notable change in this new platform occurs in the GPU.

Arm has integrated a dedicated neural network accelerator into its Mali GPU for the first time. The newly released Mali G2-Ultra NX enables certain neural graphics workloads to be processed directly within the GPU. Arm is not alone in this effort; on September 2, Qualcomm unveiled Adreno Neural Fusion technology, which adds dedicated Matrix Cores for AI computation in the next-generation Adreno GPU.

Qualcomm and Arm have successively integrated dedicated neural computing capabilities into GPUs, reflecting a shift in the competitive logic of mobile GPUs. Previously, AI computations were primarily handled by standalone NPUs within the SoC, while GPUs focused mainly on graphics rendering. However, as technologies like ray tracing, AI super-resolution, and frame generation enter mobile gaming, AI is now directly integrated into the graphics processing pipeline itself. The Mali G2-Ultra NX is Arm’s response to this evolution.

This GPU supports three types of neural graphics technologies: NSS is primarily used for neural super-sampling, NFRU enhances frame rates by generating intermediate frames via AI, and NSSD combines super-sampling and denoising for more complex scenarios such as ray tracing.

According to Arm, under neural graphics workloads, the Mali G2-Ultra NX can achieve up to a 4x improvement in performance per watt. Traditional graphics performance has also been enhanced simultaneously. Arm's data shows that the new-generation GPU delivers up to a 14% performance improvement in non-AI gaming content; when combined with neural graphics technologies such as NFRU, it enables sustained mobile gaming experiences of up to 120 FPS.

Behind this is actually a new approach to computational power. Smartphones have limited power consumption and thermal capacity, making it difficult to indefinitely scale traditional computing with GPUs. Instead of rendering every frame in full, reconstructing partial frames and generating intermediate frames through neural networks can achieve higher quality and frame rates using less traditional computation. However, this ultimately depends on whether game engine manufacturers and others are willing to “pay the bill”—Arm has already begun building out this layer of the ecosystem.

Game developers are increasingly integrating.

Light and Shadow Reborn, developed in collaboration by Arm and Sumo Digital, is the first to demonstrate the real-world performance of the Mali G2-Ultra NX on neural graphics.

Tencent Games' Public Technology Line is integrating Arm's neural network technology into Magic Dawn, an AI-driven game middleware, and Unity China is natively incorporating these capabilities into the Unity Engine.

In terms of specific gameplay, NetEase's "Yanyun Shisixiang" is integrating Arm neural network technology into the Messiah Engine and plans to release an NSS-supported version to players this year. Zhou Keli, head of the game's engine team, stated that this game will be among the first to feature Arm neural network technology.

Tencent has demonstrated NSSD technology with Arm in "Dark Zone: Infinite," enhancing complex graphical scenes through neural super-sampling and denoising; Papergames is also advancing the integration of NSS technology in "Infinite Warmth."

Can Arm drive the Mali G2-Ultra NX into more Chinese manufacturers, as the market continues to pay close attention?