Arm Details Its Mobile‑First AI Reconstruction Approach, Diverging from the "Black Box" Model of DLSS 5

Arm, the British chip‑design and software firm, used its Arm Everywhere China showcase in Shanghai to announce a suite of new AI‑driven technologies aimed squarely at mobile devices. The company’s vision is to create a complete ecosystem for AI‑native computing, anchored by the next‑generation AI‑centric compute platform CSS for Mobile 2 and the freshly unveiled Arm Mali G2‑Ultra NX graphics processor. This GPU is built around neural‑compute blocks and promises higher graphics throughput while dramatically cutting heat and power draw compared with traditional mobile GPUs. While Arm did not name specific products that will ship with the Mali G2‑Ultra NX, its predecessor, the Mali G1‑Ultra, already powers smartphones from brands such as Xiaomi, Vivo and OnePlus.

Deyan Lazarov, a senior member of Arm’s GPU team within the Edge AI division, explained that the new chip has been under development for roughly two and a half years. Its primary goal is to give mobile developers access to compute‑heavy features—most notably real‑time ray tracing—without sacrificing the battery life or thermal envelope that handheld devices demand.

"We’re seeing a real uptick in ray‑tracing adoption on phones," Lazarov said. "It moved slowly at first, but now more titles are experimenting with it, and the hardware has finally caught up." The Mali G2‑Ultra NX is designed to let developers tap into advanced Unreal Engine capabilities such as MegaLights. Arm even partnered with Sumo Digital to produce a demo title, *Neural Dawn*, which showcases the chip’s ability to render complex lighting and geometry on a smartphone. Lazarov described the target audience as "mobile‑first games that want desktop‑grade visual fidelity" as well as cross‑platform experiences that run consistently across consoles, PCs and phones.

He highlighted the practical benefit: a player could enjoy the same level of detail on a home console and then hop onto a train with a phone, without a noticeable drop in quality. A handful of upcoming games are slated to leverage the new GPU, including Tencent’s *Arena Breakout Infinite*, NetEase’s *Where Winds Meet* and Infold Games’ *Infinity Nikki*. However, the real challenge for mobile graphics is the stark difference in thermal and power budgets compared with consoles or PCs. "A phone can’t dump heat the way a console can, and it runs off a battery all the time," Lazarov noted.

Arm’s answer is to blend neural processing directly into the traditional rendering pipeline. By offloading portions of the workload to dedicated neural accelerators, the GPU can render fewer pixels and let the AI reconstruct the missing information. This reduces both the raw GPU cycles and the overall power consumption, which in turn improves system‑level efficiency and frees up resources for other tasks. "The net effect is a new class of efficiency that eases many of the trade‑offs developers normally face," Lazarov explained.

"They can choose to reinvest the saved budget into higher visual fidelity, smoother frame rates, longer battery life, or any combination thereof." To make this vision attractive to developers, Arm is taking an open‑software stance. Over the past two years the company has built a toolbox that includes Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD) and Neural Frame Rate Upscaling (NFRU). These techniques let the AI upscale images, add extra rays, and even generate entirely new frames through reconstruction.

Lazarov illustrated the process with a two‑frame example. In the first frame, the GPU renders only a quarter of the final image; the remaining three‑quarters are filled in by the neural accelerator using NSS.

In the second frame, the GPU does no traditional rendering at all—the neural engine produces every pixel, delivering a fully AI‑reconstructed frame. Nvidia’s competing DLSS 5 technology also uses AI to enhance images, but it has attracted criticism for being a closed, generative system that can stray from a developer’s artistic intent. Arm draws a sharp distinction: "Our networks are not generative; they’re convolutional, meaning they don’t invent new content, they simply know how existing content should look. We train them on real‑world data and then publish the models openly on GitHub and Hugging Face." This openness lets developers fine‑tune the models with their own art assets.

If a game has a distinctive visual style, a studio can capture data from that style, retrain the public model, and ensure the upscaling respects the original aesthetic. "Because you do the retraining yourself, there’s no risk of losing IP control," Lazarov emphasized.

By contrast, DLSS is a proprietary black box: developers cannot see the training data, cannot retrain it, and must accept the results as‑is. Arm positions its solution as a collaborative, customizable alternative. Another key differentiator is the mobile‑first design philosophy. Arm’s networks are built with strict constraints on power, memory bandwidth and thermal headroom.

"If you tried to run DLSS on a phone, you’d instantly blow past the budget," Lazarov warned. Power efficiency is a central metric for the Mali G2‑Ultra NX. Arm claims up to four‑fold gains in both performance and efficiency, plus a 70 % reduction in DRAM traffic—a major source of power draw on mobile chips. Less external memory access translates directly into longer battery life.

In practical terms, the new GPU can run *Neural Dawn* at a smooth 60 fps, whereas the previous generation struggled to reach 40 fps under ideal conditions and dropped to 15 fps in typical usage. From a tooling perspective, Arm currently offers Unreal Engine plug‑ins and plans to extend support to Unity and other engines.

An SDK provides a streamlined integration path for custom engines as well. Lazarov noted that adding Arm’s AI‑native graphics to an existing shipped title could take as little as three months, enabling rapid upgrades for legacy games. The *Neural Dawn* demo itself was built by a small team—fewer than 20 people—in under 18 months, demonstrating how Arm’s technology can accelerate development cycles while delivering cutting‑edge visual quality. Overall, Arm’s approach promises a more open, efficient and mobile‑centric alternative to the black‑box AI upscaling models that dominate the high‑end PC and console markets.