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

Arm, the UK‑based semiconductor and software design powerhouse, staged its Arm Everywhere China showcase in Shanghai to announce a suite of fresh AI‑computing initiatives. Central to the reveal is a comprehensive AI‑native ecosystem that includes the new CSS for Mobile 2 compute platform and the Mali G2‑Ultra NX GPU – a graphics processor built around neural‑accelerator technology. Arm argues that its on‑device AI‑driven graphics pipeline can deliver higher visual fidelity while consuming far less power and generating less heat than competing solutions, positioning it as a more open alternative to Nvidia’s controversial DLSS 5 and as a closer match to developers’ original creative intent.

While Arm did not name specific products that will ship with the Mali G2‑Ultra NX, the predecessor Mali G1‑Ultra has already been integrated into smartphones from brands such as Xiaomi, Vivo and OnePlus, suggesting a similar market trajectory for the newer chip. Deyan Lazarov, a senior 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 design goal is to give mobile developers access to compute‑heavy features—most notably real‑time ray tracing—without the traditional performance penalties. "Ray tracing is finally gaining traction on mobile devices," Lazarov said.

"It was slow to take off, but we’re now seeing a growing number of titles that want to enable it." As reported earlier by GamesIndustry.biz, the new GPU supports Unreal Engine capabilities such as MegaLights, and Arm partnered with Sumo Digital to produce the demo title Neural Dawn, which serves as a technical showcase for the Mali G2‑Ultra NX’s capabilities. Lazarov emphasized that the chip targets games that are "mobile‑first but aspire to look like desktop‑class experiences." He also highlighted cross‑platform titles that aim to deliver a seamless visual experience whether the player is on a home console, a laptop, or a train‑bound smartphone. "When you move from a console at home to a phone on the go, you shouldn’t notice a dramatic drop in quality," he added. Among the early adopters are Tencent’s Arena Breakout Infinite, NetEase’s Where Winds Meet, and Infold Games’ Infinity Nikki.

Lazarov pointed out a fundamental hardware difference: mobile devices operate under tighter thermal and power constraints than consoles or PCs. "A phone can’t dump heat the way a console can, and it runs on a battery all the time," he explained. Arm’s answer is to blend neural processing with the traditional rendering pipeline. By offloading portions of the workload to dedicated neuro‑accelerators, the GPU’s overall power draw drops, which in turn improves system‑level efficiency.

"This new efficiency removes many of the trade‑offs developers normally have to make," Lazarov said. "They can reinvest the saved budget into higher visual fidelity, smoother frame rates, longer battery life, or any combination thereof." To foster developer adoption, Arm is taking an open‑source 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 allow the GPU to render a fraction of the pixels and let the neuro‑accelerators reconstruct the remainder, or even generate entire frames without traditional rasterisation. Lazarov illustrated the process with two example frames: in the first, the GPU renders only 25 % of the image while the neuro‑accelerators fill in the remaining 75 % via NSS; in the second, the neuro‑accelerators produce every pixel, meaning the GPU does no conventional rendering at all. Nvidia’s DLSS 5 also uses AI to enhance images, but its approach has drawn criticism for being a closed, generative system that can diverge from a developer’s artistic vision. Arm deliberately differentiates itself.

"Our networks are not generative; they are convolutional and simply learn how things should look based on real‑world data," Lazarov explained. "We train them on authentic content and then publish the models publicly on GitHub and Hugging Face." Because the models are open, developers can fine‑tune them with data from their own games.

If a title has a unique art style, the team can capture representative frames, retrain the network, and ensure that upscaling or frame‑generation respects that distinctive look. This process preserves intellectual‑property control and eliminates the “black‑box” uncertainty associated with proprietary solutions. "DLSS is a closed, proprietary system—you can’t see what it’s been trained on or retrain it yourself," Lazarov noted.

"Our approach is take‑it‑and‑make‑it‑your‑own." Arm also stresses that its technology was engineered with mobile constraints in mind—power, bandwidth, and thermal budgets are baked into the network design. By contrast, DLSS would quickly exceed a phone’s limits even if the software were ported.

Lazarov cited a 70 % reduction in DRAM traffic for the Mali G2‑Ultra NX, a major source of power consumption on handheld devices. Less memory access translates directly into longer battery life and cooler operation. Performance numbers illustrate the leap. Running Neural Dawn on the previous generation GPU capped at about 40 fps under ideal conditions and fell to 15 fps in typical use.

The G2‑Ultra NX comfortably sustains 60 fps, delivering a fluid experience without sacrificing visual quality. From a tooling perspective, Arm currently offers Unreal Engine plug‑ins and plans to extend support to Unity and other engines. A recent partnership with Unity China signals that Unity integration is on the roadmap, though the immediate focus remains on Unreal.

For studios using custom engines, Arm provides an SDK that streamlines integration without requiring bespoke plug‑ins. Lazarov highlighted the speed with which existing titles can adopt the technology. "If you have a shipped game and you want to add our AI‑native graphics, you can do it in as little as three months and release an updated version," he said.

This rapid turnaround was demonstrated by Neural Dawn, which a team of fewer than 20 developers built in under 18 months, leveraging Arm’s tools to achieve high‑quality graphics without the usual development overhead. In summary, Arm’s mobile‑first AI reconstruction strategy offers an open, developer‑friendly alternative to Nvidia’s DLSS 5. By combining neural‑accelerated upscaling, denoising, and frame‑rate generation with a focus on power efficiency and transparency, Arm aims to empower game creators to deliver console‑level visuals on smartphones while maintaining control over their artistic direction and respecting the strict constraints of mobile hardware.