Arm explains its mobile-first AI-reconstruction technology, which takes a different tack from the "black box" approach of DLSS 5

GamesIndustry.biz details Arm’s Shanghai unveiling of CSS for Mobile 2 and the Mali G2-Ultra NX GPU, part of a broader AI-native compute push for phones, tablets, and other battery-limited devices. Arm is pitching the tech as a developer-friendly alternative to closed AI upscalers, with open networks on GitHub and Hugging Face and tooling that lets studios retrain models on their own game data.

The core pitch is practical: use neural processing to reduce traditional GPU work, cut power draw, and preserve thermals while still pushing image quality. Arm says its Neural Super Sampling, Neural Super Sampling and Denoising, and Neural Frame Rate Upscaling can reconstruct large portions of a frame, cast extra rays, and generate frames in ways tuned for mobile constraints rather than desktop assumptions.

Arm says the Mali G2-Ultra NX has been in development for about two and a half years, and that it is meant to help mobile-first and cross-platform games look closer to desktop titles without forcing the usual quality-versus-battery tradeoff. The company also claims the approach can support ray tracing and Unreal Engine features such as MegaLights, with Neural Dawn serving as a showcase built with Sumo Digital.

Specific shipping devices for the new GPU have not been named, though Arm’s earlier Mali G1-Ultra appeared in Xiaomi, Vivo, and OnePlus hardware. The company says titles including Arena Breakout Infinite, Where Winds Meet, and Infinity Nikki are among the games expected to use the new chip’s capabilities.