Arm Details Its Mobile‑First AI Reconstruction Approach, Offering an Alternative to DLSS 5’s Black‑Box Model

Arm, the British chip and software design powerhouse, staged its Arm Everywhere China showcase in Shanghai to announce a suite of fresh AI‑computing offerings aimed squarely at the mobile market. Central to the announcement is a comprehensive AI‑native ecosystem that includes the new CSS for Mobile 2 compute platform and the Mali G2‑Ultra NX graphics processor. The G2‑Ultra NX is a GPU built around neural‑accelerator cores, promising higher visual fidelity while keeping heat and power draw far lower than conventional designs. Arm positions this on‑device AI solution as a more transparent and developer‑friendly counterpart to Nvidia’s controversial DLSS 5, aligning more closely with the original intent of game creators.

While Arm did not name specific products that will ship with the Mali G2‑Ultra NX, the predecessor Mali G1‑Ultra already powers devices from brands such as Xiaomi, Vivo and OnePlus, suggesting a similar rollout path for the new chip. Deyan Lazarov, a member of Arm’s GPU team within the Edge AI division, explained that the G2‑Ultra NX has been in development for roughly two and a half years.

Its purpose is to give mobile developers access to compute‑heavy features—most notably ray tracing—without the prohibitive energy costs that have traditionally kept such techniques off smartphones. "We’re seeing a real uptick in ray‑tracing adoption on phones," Lazarov noted. "It was slow to start, but more titles are now experimenting with it, and the G2‑Ultra NX is designed to support that shift." The GPU also integrates tightly with Unreal Engine, enabling developers to leverage technologies like MegaLights.

In partnership with Sumo Digital, Arm demonstrated the capabilities of the chip in the demo title Neural Dawn, which showcases real‑time ray‑traced lighting and AI‑enhanced upscaling on a mobile device. Lazarov described the target audience as "mobile‑first games that want desktop‑level visual quality" as well as cross‑platform titles that aim for a consistent experience across consoles, PCs and phones.

He emphasized that gamers should not notice a drop in quality when they move from a home console to a train‑riding session on a smartphone. Among the upcoming games slated to use the G2‑Ultra NX are Tencent’s Arena Breakout Infinite, NetEase’s Where Winds Meet, and Infinity Nikki from Infold Games.

Each of these titles will benefit from the chip’s ability to handle high‑resolution rendering while staying within the tight thermal and power envelopes of mobile hardware. "Phones can’t dissipate heat the way a console can, and they run on batteries all the time," Lazarov explained.

"Our solution blends neural processing with the traditional rendering pipeline, cutting GPU workload and power draw, which in turn lifts overall system efficiency." The efficiency gains translate into fewer compromises for developers. With AI‑assisted rendering handling a large portion of the visual workload, studios can reallocate saved cycles to improve texture quality, increase frame rates, extend battery life, or pursue any combination of those goals. Arm backs this vision with an open‑source software stack.

Over the past two years the company has built tools that let developers plug AI directly into their graphics pipelines. "We wanted an AI approach that is developer‑friendly and gives real, measurable efficiency gains," Lazarov said.

"Those gains can then be reinvested in visual fidelity, performance, battery life, or any mix you prefer." Key components of the stack include Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD) and Neural Frame Rate Upscaling (NFRU). Together they can upscale lower‑resolution images, generate additional rays for more realistic lighting, and synthesize extra frames to smooth motion. In a typical workflow, the GPU renders only a quarter of the final image while the neural accelerators reconstruct the remaining three‑quarters using NSS. In a more extreme mode, the entire frame can be generated by the AI engines, eliminating traditional rasterisation for that frame altogether.

Arm’s approach sharply contrasts with Nvidia’s DLSS 5, which relies on generative networks that can sometimes diverge from a developer’s artistic intent. Lazarov stressed that Arm’s networks are convolutional, not generative, meaning they do not invent new content but rather learn how existing content should look. All of Arm’s models—resolution upscalers, frame‑rate generators, etc.—are publicly available on GitHub and Hugging Face, allowing developers to fine‑tune them with their own art assets. This openness safeguards intellectual‑property concerns and lets studios tailor the AI to their unique visual style.

"DLSS is a closed, black‑box solution," Lazarov said. "You can’t see what it’s trained on or retrain it with your own data. Our model is open, you can retrain it, and you retain full control over the output." Another distinction is the mobile‑first design philosophy.

Arm’s networks are built to respect the strict power, bandwidth and memory limits of phones. Attempting to run DLSS on a smartphone would instantly exceed those budgets, whereas Arm’s solution fits comfortably within them. Power efficiency is a cornerstone of the G2‑Ultra NX. Lazarov highlighted a four‑fold boost in performance and efficiency, coupled with a 70 % reduction in DRAM traffic—a major source of power consumption on mobile devices.

Less external memory access translates directly into longer battery life. In real‑world tests, the previous‑generation GPU could only push Neural Dawn to about 40 fps under ideal conditions and dropped to roughly 15 fps in typical scenarios. The new G2‑Ultra NX comfortably sustains 60 fps, delivering a noticeably smoother experience.

To help developers adopt the technology, Arm currently offers Unreal Engine plug‑ins and plans to expand to Unity and other engines. An SDK provides a streamlined integration path for custom engines as well.

Lazarov noted that retrofitting an existing shipped title with Arm’s AI graphics can take as little as three months, enabling studios to roll out enhanced versions quickly. The demo title Neural Dawn was built by a team of fewer than 20 people in under 18 months, illustrating how Arm’s tools can accelerate development cycles while still delivering cutting‑edge visual quality.

In summary, Arm’s mobile‑first AI reconstruction platform offers an open, efficient, and highly adaptable alternative to the closed‑box DLSS 5 approach, promising richer graphics on smartphones without sacrificing power or creative control.