Arm has introduced CSS for Mobile 2, a computing platform specifically designed to handle autonomous agents and advanced graphics rendering without compromising battery life.
This new architecture combines a novel GPU, which integrates several neural accelerators, with an optimized CPU cluster to coordinate complex workflows, delivering superior performance and significantly improved energy efficiency.
Arm: the new architecture for smartphone chips goes all-in on AI
<!– wp:pgcsimplygalleryblock/slider {"sliderAdminWidth":1068,"startPosIndex":0,"galleryType":"pgc_sgb_slider","galleryId":"49a74661","images":[{"id":688397,"title":"Arm-CSS-for-Mobile-2","url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-CSS-for-Mobile-2.jpg","link":"https://gizchina.it/?attachment_id=688397","alt":"Arm CSS for Mobile 2","description":"","caption":"Credits: Arm","mime":"image/jpeg","type":"image","width":1500,"height":844,"sizes":{"thumbnail":{"height":150,"width":150,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-CSS-for-Mobile-2-150×150.jpg","orientation":"landscape"},"medium":{"height":169,"width":300,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-CSS-for-Mobile-2-300×169.jpg","orientation":"landscape"},"large":{"height":675,"width":1200,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-CSS-for-Mobile-2-1200×675.jpg","orientation":"landscape"},"full":{"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-CSS-for-Mobile-2.jpg","height":844,"width":1500,"orientation":"landscape"}},"imageMeta":{"aperture":"0","credit":"","camera":"","caption":"","created_timestamp":"0","copyright":"","focal_length":"0","iso":"0","shutter_speed":"0","title":"","orientation":"0","keywords":[],"alt":""},"meta":false,"postlink":"https://gizchina.it/?attachment_id=688397"},{"id":688399,"title":"Arm-Mali-G2-Ultra-NX-GPU","url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-Mali-G2-Ultra-NX-GPU.jpeg","link":"https://gizchina.it/?attachment_id=688399","alt":"Arm CSS for Mobile 2","description":"","caption":"Credits: Arm","mime":"image/jpeg","type":"image","width":2048,"height":1152,"sizes":{"thumbnail":{"height":150,"width":150,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-Mali-G2-Ultra-NX-GPU-150×150.jpeg","orientation":"landscape"},"medium":{"height":169,"width":300,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-Mali-G2-Ultra-NX-GPU-300×169.jpeg","orientation":"landscape"},"large":{"height":675,"width":1200,"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-Mali-G2-Ultra-NX-GPU-1200×675.jpeg","orientation":"landscape"},"full":{"url":"https://gizchina.it/wp-content/uploads/2026/09/Arm-Mali-G2-Ultra-NX-GPU.jpeg","height":1152,"width":2048,"orientation":"landscape"}},"imageMeta":{"aperture":"0","credit":"","camera":"","caption":"","created_timestamp":"0","copyright":"","focal_length":"0","iso":"0","shutter_speed":"0","title":"","orientation":"0","keywords":[],"alt":""},"meta":false,"postlink":"https://gizchina.it/?attachment_id=688399"}]}The graphics engine represents one of the biggest leaps forward. The company unveiled the Mali G2-Ultra NX GPU, the brand’s first graphics chip to integrate neural accelerators directly into the rendering pipeline. This solution enables AI operations to be fused with traditional graphics processing.
The system, instead of calculating every single frame at maximum resolution, can render images at 540p and then scale them to 1080p using neural networks, or generate additional frames to go from 30 to 60 fps.
Data show up to 4x higher efficiency per watt in neural graphics operations and a 14% increase in standard gaming performance compared to previous models.
All of this is supported by a new Ray Tracing unit, designed to handle realistic lighting while keeping device temperatures under control and cutting DRAM data movement by 70%.
A processor designed for autonomous agents
Managing an AI that operates in the background, plans actions and understands the context of the user requires a remarkable coordination capability. This task is handled by the new C2 CPU cluster, which pairs the powerful cores C2-Ultra with the more efficient C2-Pro.
By integrating two SME2 units, the architecture guarantees a execution speed 70% higher with small-language models. Measurements show AI-related performance gains up to 1.7x and a 15% higher speed in single-thread operations compared to the previous C1-Ultra.
An essential detail concerns power consumption: these speeds are achieved using up to 38% less energy, a vital factor for smartphones operating with limited power.
An ecosystem ready for developers
Hardware of this caliber requires software optimized to unleash its potential. Several players in the gaming industry, such as Tencent Games, NetEase and Sumo Digital, are already testing the platform’s capabilities by integrating neural technology into their graphics engines.
To facilitate developers’ work, Arm has structured dedicated tools, including the KleidiAI libraries and the Neural Graphics Development Kit.
Moreover, through a single portal, developers can download pre-optimized models and test scripts, ensuring that applications can fully leverage the acceleration offered by CSS for Mobile 2 from the moment of market launch.
