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

Arm, the British chip and software design powerhouse, staged its Arm Everywhere China showcase in Shanghai to announce the latest strides it is making in AI‑driven computing. The centerpiece of the event was a vision of a complete AI‑native ecosystem that spans hardware, software and tools, anchored by the new AI‑centric compute platform CSS for Mobile 2 and the freshly unveiled Arm Mali G2‑Ultra NX GPU. This GPU is built around neural processing units and promises to lift graphics performance while dramatically cutting heat output and power draw. While Arm did not name the specific devices that will ship with the Mali G2‑Ultra NX, it noted that 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 product team within the Edge AI division, explained that the G2‑Ultra NX has been under 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 thermal and battery penalties that traditionally accompany such workloads. "We’re seeing a clear uptick in interest for ray tracing on mobile titles," Lazarov said.

"It was a slow burn at first, but now more and more games are asking for it." The new GPU, according to a GamesIndustry.biz report earlier this year, enables developers to tap into Unreal Engine capabilities like MegaLights. Arm even partnered with Sumo Digital to produce a demo game called *Neural Dawn*, which serves as a practical showcase of the Mali G2‑Ultra NX’s capabilities. Lazarov emphasized that the chip is meant for games that are "mobile‑first but want desktop‑grade visuals." He added that many titles are built to be cross‑platform from the ground up, meaning players expect a seamless experience whether they are on a console at home or a phone on a train.

"The hardware has to adapt to very different thermal envelopes and power budgets," he noted. Unlike consoles or PCs, phones cannot dissipate large amounts of heat and must run continuously off a battery. To address those constraints, Arm has woven neural processing directly into the traditional graphics pipeline.

By offloading portions of the rendering workload to dedicated neural accelerators, the GPU can cut its own power consumption and reduce traffic to external DRAM—one of the biggest drains on mobile energy budgets. Lazarov explained that this synergy translates into "new efficiency that lifts many of the compromises developers normally have to make." Arm’s strategy for encouraging adoption is rooted in openness.

Over the past two years the company has been collaborating with the broader ecosystem to create developer‑friendly AI tools. "Our aim was to devise an AI workflow that developers can actually control and benefit from," Lazarov said. The result is a suite of technologies including Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD) and Neural Frame Rate Upscaling (NFRU).

These techniques can upscale images, add extra rays, and even generate new frames using AI‑based reconstruction. In a practical example, Lazarov described a frame where the GPU renders only 25 % of the pixels while the remaining three‑quarters are reconstructed by the neural accelerators via NSS.

In a second scenario, the entire frame is produced by the neural hardware with zero traditional rasterization work, demonstrating the potential for dramatic workload reduction. Nvidia’s DLSS 5 also uses AI to enhance image detail, but it has attracted criticism for being a closed, generative system that can diverge from a developer’s artistic intent. Arm’s approach is fundamentally different. "Our networks are convolutional, not generative.

They don’t imagine new content; they simply know how existing content should look," Lazarov clarified. The networks have been trained on real‑world data and are publicly available on GitHub and Hugging Face, allowing developers to fine‑tune them with their own art styles.

This openness eliminates concerns about intellectual‑property loss and gives studios the freedom to retrain the models on bespoke datasets. 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’s open‑source stance lets developers pull a network, adapt it to a quirky visual style, and retain full control over the upscaling and frame‑generation process. Another key distinction is that Arm engineered its solution with mobile constraints in mind.

Power, bandwidth and memory budgets are baked into the network design, making it viable on smartphones where DLSS would simply exceed the limits. "If you tried to run DLSS on a phone, you’d instantly blow past your power envelope," Lazarov warned.

Performance numbers back up the claims. The Mali G2‑Ultra NX delivers roughly four times the performance and efficiency of its predecessor while slashing DRAM traffic by about 70 %. Less memory traffic means lower power draw, a crucial factor for battery‑powered devices. In a head‑to‑head test, the older GPU could only push *Neural Dawn* to 40 fps at its limit and dropped to 15 fps under normal conditions.

The new chip comfortably hits 60 fps, delivering a buttery‑smooth experience. On the tooling side, Arm currently offers plug‑ins for Unreal Engine and has a roadmap that includes Unity support—already hinted at through a recent partnership with Unity China.

For studios using custom engines, Arm provides an SDK that streamlines integration without requiring a dedicated plug‑in. Lazarov highlighted that even legacy titles can benefit from the AI‑native graphics stack. "If you have an existing shipped game and you integrate our solution, you could have a new version with these enhancements in as little as three months," he said.

The *Neural Dawn* demo itself was built by a team of fewer than 20 people in under 18 months, underscoring how Arm’s technology can accelerate development cycles and reduce the resources needed to achieve high‑quality visuals. Overall, Arm’s mobile‑first, open, and power‑aware AI reconstruction approach aims to give developers the flexibility to prioritize visual fidelity, performance, battery life—or any combination thereof—without the trade‑offs that have traditionally hampered mobile gaming.