Arm Details Its Mobile‑First AI Reconstruction Approach, Contrasting With Nvidia’s DLSS 5 Black‑Box Model

Arm, the British chip and software design powerhouse, recently staged its Arm Everywhere China showcase in Shanghai, unveiling a suite of new tools aimed at reshaping AI‑driven graphics on mobile devices. The centerpiece of the announcement is the Mali G2‑Ultra NX GPU, a processor built around neural‑accelerated rendering that promises desktop‑level visual fidelity while staying within the tight power, thermal and bandwidth limits of smartphones, tablets, laptops and even handheld consoles. The company’s broader vision is to create an end‑to‑end ecosystem for AI‑native computing.

This includes the CSS for Mobile 2 compute platform, which provides a software stack optimized for on‑device AI, and a collection of developer tools that make it easier to integrate neural‑enhanced graphics into existing pipelines. Arm positions its solution as an open alternative to Nvidia’s DLSS 5, which has attracted criticism for its closed, proprietary nature and for sometimes diverging from a developer’s artistic intent. While Arm did not disclose the exact products that will ship with the Mali G2‑Ultra NX, the predecessor Mali G1‑Ultra has already found its way into phones from Xiaomi, Vivo and OnePlus, suggesting that the new chip could appear in a similar range of high‑end Android devices. Deyan Lazarov, a senior engineer on Arm’s GPU team within the Edge AI business unit, explained that the team has spent roughly two and a half years refining the silicon and the accompanying software.

Their goal was to give mobile developers access to compute‑heavy techniques such as ray tracing without sacrificing battery life or causing the device to overheat. "We’re seeing a growing appetite for ray tracing in mobile titles," Lazarov said. "It was slow to take off at first, but more games are now looking to enable it, and our GPU is designed to make that feasible." One of the first demonstrations of the chip’s capabilities is a demo game called *Neural Dawn*, built in collaboration with Sumo Digital.

The title showcases how the Mali G2‑Ultra NX can leverage Unreal Engine features like MegaLights, delivering lighting and shading effects that would normally require a much more powerful desktop GPU. Other upcoming titles that will benefit from the new hardware include Tencent’s *Arena Breakout Infinite*, NetEase’s *Where Winds Meet* and Infold Games’ *Infinity Nikki*.

Lazarov emphasized that mobile devices face unique constraints compared with consoles or PCs. "A phone can’t dissipate heat the way a console can, and it runs on a battery all the time," he explained. "Our approach integrates neural processing directly into the traditional rendering pipeline, which offloads work from the GPU, reduces power draw, and ultimately frees up system resources for other tasks." The result, according to Arm, is a dramatic boost in efficiency.

The company cites up to four‑times performance gains, four‑times better energy efficiency, and a 70 % reduction in DRAM traffic—one of the biggest sources of power consumption on mobile chips. By cutting memory accesses, the GPU not only saves energy but also lowers latency, helping to maintain smooth frame rates even when complex AI‑based upscaling is in use. Arm’s AI‑enhanced graphics stack revolves around three core technologies: 1.

**Neural Super Sampling (NSS)** – an AI‑driven upscaling algorithm that reconstructs a higher‑resolution image from a lower‑resolution render. 2.

**Neural Super Sampling and Denoising (NSSD)** – combines upscaling with noise reduction to produce cleaner visuals. 3. **Neural Frame Rate Upscaling (NFRU)** – generates intermediate frames using AI, effectively increasing the perceived frame rate without additional GPU rendering work. In practice, a typical frame might be rendered at only a quarter of the native resolution by the GPU, with the remaining three‑quarters reconstructed by the on‑chip neural accelerators.

In some cases, an entire frame can be produced solely by the AI engines, eliminating traditional rasterization altogether. This flexibility allows developers to allocate GPU cycles to other tasks, such as more sophisticated lighting, physics simulations, or simply to conserve battery life. When comparing this approach to Nvidia’s DLSS 5, Lazarov drew a clear line between openness and opacity.

"Our networks are convolutional, not generative. They don’t invent new content; they learn how existing content should look and then apply that knowledge," he said. "All of our models are published publicly on GitHub and Hugging Face, so developers can inspect, fine‑tune, or even retrain them with data from their own games. That guarantees artistic control and eliminates any risk of IP loss." By contrast, DLSS is a closed‑source, black‑box solution.

Developers receive a pre‑trained model that cannot be altered, meaning they must accept whatever visual style the model produces or abandon the technology entirely. Lazarov described Arm’s model as a "take‑it‑and‑make‑it‑work‑for‑you" approach, where studios can capture data from their own art assets, retrain the network, and ensure the upscaled output matches their unique aesthetic. Another major differentiator is the mobile‑first design philosophy.

Arm’s neural networks are built to respect strict power, memory bandwidth and thermal budgets. "If you tried to run DLSS on a phone, you would instantly exceed those limits," Lazarov warned. "Our solution is engineered from the ground up to stay within mobile constraints, delivering high‑quality upscaling without blowing the battery or overheating the device." To help developers adopt the technology, Arm has released plugins for Unreal Engine and is planning support for Unity and other engines.

An SDK provides a streamlined integration path for custom engines as well. Lazarov noted that even legacy titles can benefit: "If you have an existing game on the market, you can integrate our AI graphics stack and ship an updated version in as little as three months." The performance leap is tangible. When pushed to its limits, the previous‑generation Mali GPU could only run *Neural Dawn* at about 40 fps under ideal conditions and dropped to roughly 15 fps in typical gameplay.

The new G2‑Ultra NX comfortably maintains 60 fps, delivering a smoother, more responsive experience while preserving battery life. Arm’s open, developer‑friendly stance, combined with a hardware architecture that embraces AI without compromising mobile constraints, signals a shift in how graphics will evolve on handheld platforms. By offering transparent, retrainable models and a clear path for integration, Arm aims to empower studios of all sizes to bring console‑grade visual fidelity to phones and tablets, all while keeping power consumption and thermal output in check. In summary, Arm’s Mali G2‑Ultra NX and its accompanying AI‑native graphics suite present a compelling alternative to proprietary solutions.

The open‑source convolutional networks, mobile‑first optimization, and robust toolchain give developers the freedom to enhance resolution, frame rate and visual quality on devices that were previously limited by hardware constraints. As more titles adopt these capabilities, gamers can expect richer, smoother experiences on the devices they already carry in their pockets.