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

Arm, the British chip and software design powerhouse, hosted its Arm Everywhere China showcase in Shanghai to announce a suite of fresh AI‑computing innovations. The centerpiece of the event was a comprehensive AI‑native ecosystem that includes the upcoming AI‑centric compute platform CSS for Mobile 2 and the brand‑new Arm Mali G2‑Ultra NX graphics processor. This GPU is built around neural‑computing blocks, promising higher graphics throughput while keeping thermal output and power draw modest.

Arm positions its on‑device AI‑driven graphics pipeline as a more transparent alternative to Nvidia’s contentious DLSS 5, aiming to align more closely with developers’ original creative intent. 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 manufacturers such as Xiaomi, Vivo and OnePlus. Deyan Lazarov, a 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 purpose is to give mobile developers access to heavyweight rendering techniques—most notably ray tracing—without sacrificing the efficiency required for handheld devices.

"We’ve seen a steady uptick in mobile games experimenting with ray tracing," Lazarov noted. "It moved slowly at first, but today more titles are actively exploring how to bring realistic lighting to phones." According to GamesIndustry.biz, the Mali G2‑Ultra NX enables developers to leverage Unreal Engine features such as MegaLights. Arm partnered with Sumo Digital to produce the demo title *Neural Dawn*, a showcase of the chip’s capabilities. Lazarov described the target audience as games that are "mobile‑first yet aspire to the visual fidelity of desktop experiences." He added that many modern titles are designed to be cross‑platform, allowing players to enjoy a consistent experience whether they are on a console at home or a phone on a commuter train.

Upcoming games slated to run on the new GPU include Tencent’s *Arena Breakout Infinite*, NetEase’s *Where Winds Meet* and Infold Games’ *Infinity Nikki*. Lazarov emphasized that smartphones operate under very different thermal and power constraints than consoles or PCs. "A phone can’t dissipate heat the way a console can, and it’s always running off a battery," he explained. Arm’s answer is to fuse neural processing with the traditional graphics pipeline.

By offloading portions of the rendering workload to dedicated neuro‑accelerators, the GPU can lower its power consumption and free up resources for other system tasks. "That translates into new efficiency gains, which in turn reduces the compromises developers normally have to make," Lazarov said. To spur adoption, Arm is taking an open‑software stance. Over the past two years the company has collaborated with ecosystem partners to create a developer‑friendly AI stack.

"Our goal was to devise an AI workflow that developers can actually use, delivering massive efficiency gains that they can reinvest in visual quality, performance, battery life, or any mix of those," Lazarov explained. Key components of Arm’s suite are Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD) and Neural Frame‑Rate Upscaling (NFRU). These technologies upscale images, add extra ray samples, and generate additional frames through AI‑driven reconstruction. Lazarov illustrated the process with a two‑frame example: in the first frame the GPU renders only 25 % of the pixels while the neuro‑accelerators reconstruct the remaining three‑quarters using NSS.

In the second frame, the GPU does virtually no traditional rendering; every pixel is produced by the AI engine. Nvidia’s DLSS 5 also uses AI to enrich image detail, but it has drawn criticism for its “black‑box” nature and for occasionally diverging from a developer’s artistic vision.

Arm draws a clear line between its approach and Nvidia’s. "Our networks are not generative; they’re convolutional, meaning they don’t invent new content—they simply know how existing content should look," Lazarov clarified. "We train them on real‑world data and make the models publicly available on GitHub and Hugging Face." Because the models are open, developers can fine‑tune them on their own art assets. "If your game has a distinctive style, you can capture a dataset from your title and retrain our network so the upscaling respects your unique aesthetic," Lazarov said.

This openness eliminates concerns about losing IP control, a risk inherent in proprietary solutions where the training data and algorithm are hidden. In contrast, DLSS is a closed system.

"You can’t see what’s inside the model, you can’t retrain it, and you’re stuck with a take‑it‑or‑leave‑it proposition," Lazarov argued. Arm’s solution, by contrast, invites developers to adapt the technology to their needs.

Another major differentiator is the mobile‑first design philosophy. Arm’s neural networks are built with strict limits on power, bandwidth and memory traffic. "If you tried to run DLSS on a phone, you’d instantly blow past the power budget," Lazarov warned.

Arm’s architecture, however, reduces DRAM traffic by up to 70 %, a significant factor because external memory accesses are a major source of power draw on mobile chips. Performance numbers back up the claims.

When pushed to its limits, the previous‑generation Mali GPU could only sustain *Neural Dawn* at about 40 fps in ideal conditions and dropped to roughly 15 fps in typical use. The G2‑Ultra NX, by contrast, comfortably delivers 60 fps for the same title, delivering up to four‑times the performance and efficiency. On the tooling side, Arm currently offers plug‑ins for Unreal Engine and is planning to expand to other engines, with Unity on the roadmap.

A recent partnership with Unity China signals that support for Unity will arrive soon. For studios using custom engines, Arm provides an SDK that streamlines integration without needing a dedicated plug‑in. Lazarov highlighted that existing games can also benefit from the AI‑native graphics stack. "If you have a shipped title and want to integrate our solution, you could have a new version ready in as little as three months," he said.

The *Neural Dawn* demo itself was built by a small team of fewer than twenty people in under 18 months, showcasing how Arm’s technology can accelerate development cycles and enable rapid visual upgrades. In summary, Arm’s mobile‑first AI reconstruction strategy offers an open, efficient and developer‑centric alternative to the more opaque, console‑focused solutions currently dominating the market.