Arm unveils CSS for Mobile 2 with neural graphics
Arm has outlined its next mobile compute platform, CSS for Mobile 2, which pairs the C2-Ultra CPU with the Mali G2-Ultra NX GPU. The graphics unit introduces dedicated Neural Accelerators to handle upscaling, frame generation, and ray reconstruction, positioning the architecture as a mobile alternative to desktop neural rendering features like DLSS.
The core of the announcement is the Mali G2-Ultra NX, which Arm describes as its first AI-native Mali GPU. The architecture integrates dedicated Neural Accelerators (NX) directly alongside the shader execution hardware. This hardware is specifically targeted at neural graphics workloads, including upscaling, frame generation, and ray reconstruction.
Arm lists three hardware-accelerated neural graphics features for this platform. Neural Super Sampling (NSS) is designed to upscale resolution from 540p to 1080p. Neural Frame Rate Upscaling (NFRU) aims to double the output frame rate from 30 FPS to 60 FPS. A third feature, Neural Super Sampling and Denoising (NSSD), combines super sampling with ray reconstruction to improve image quality in ray-traced scenarios.
According to internal comparisons provided by Arm, AI-native rendering on the G2-Ultra NX can deliver up to a 4x FPS uplift and a 4x efficiency improvement. The company also claims a 70% reduction in DRAM traffic when using these neural features. For traditional rendering paths, Arm reports a 24% benchmark uplift and 14% higher performance in non-AI gaming workloads. Additionally, the company lists a 70% reduction in ray-tracing workload, suggesting significant offloading to the neural hardware.
Beyond the neural accelerators, the GPU introduces a new Execution Engine. Arm describes this as its biggest ISA re-architecture in seven generations, with specific optimizations for Unreal Engine 5 Nanite and Lumen. The platform also includes a third-generation Ray Tracing Unit, intended to support full-scene lighting and shadows.
On the CPU side, the new Arm C2-Ultra includes SME2 support to accelerate AI workloads. Arm claims 15% higher single-thread performance and 12% higher multi-thread performance compared with its previous generation. The company also cites 15% faster web browsing and 12% faster application launches. In AI model tests, two SME2-enabled C2-Ultra cores are claimed to provide up to 1.7x higher performance across Arm’s latest benchmarks.
Arm has not provided specific commercial SoCs or launch devices in the preview material. The company states that CSS for Mobile 2 is intended as a deployable platform for partners building future mobile hardware, with the C2-Ultra CPU and Mali G2-Ultra NX forming its main compute components. The proposal positions Arm’s mobile strategy increasingly around on-device AI and neural graphics, mirroring trends seen in desktop GPU architectures.
Гика
