Arm, the British chip and software design powerhouse, recently staged its Arm Everywhere China showcase in Shanghai, where it introduced the latest advances it is bringing to the AI‑computing arena. The centerpiece of the announcement was a comprehensive AI‑native ecosystem that includes the new CSS for Mobile 2 compute platform and the Mali G2‑Ultra NX graphics processor. This GPU is built around neural‑compute principles, promising a noticeable uplift in visual performance while keeping heat output and power draw dramatically lower than conventional designs. While Arm did not disclose the exact products that will ship with the Mali G2‑Ultra NX, its predecessor, the Mali G1‑Ultra, already powers devices from brands such as Xiaomi, Vivo and OnePlus, hinting that the next‑generation silicon could soon appear in a similar range of smartphones and tablets.
Deyan Lazarov, a senior member of Arm’s GPU team within the Edge AI division, explained that development on the G2‑Ultra NX has spanned roughly two and a half years. The goal, according to Lazarov, is to give mobile developers a pathway to leverage heavyweight rendering techniques—most notably ray tracing—without the prohibitive thermal and energy costs that have traditionally confined such features to PCs and consoles. "Ray tracing has been gaining traction on mobile games," Lazarov said. "It was a slow burn at first, but now we see a growing number of titles experimenting with real‑time reflections and global illumination on phones.
Our chip is designed to make that transition smoother." The new GPU also integrates tightly with Unreal Engine, unlocking capabilities like MegaLights. Arm partnered with Sumo Digital to produce a demo title called *Neural Dawn*, which serves as a showcase for the Mali G2‑Ultra NX’s capabilities.
Other upcoming games slated to run on the chip include Tencent’s *Arena Breakout Infinite*, NetEase’s *Where Winds Meet*, and Infold Games’ *Infinity Nikki*. Lazarov emphasized that the chip targets three distinct use cases: (1) mobile‑first games that aspire to look like desktop experiences, (2) cross‑platform titles that need a consistent visual quality across consoles, PCs and phones, and (3) on‑the‑go gaming where a user might switch from a home console to a train‑bound smartphone without a noticeable dip in fidelity. He noted that mobile devices operate under very different thermal envelopes and power budgets compared with stationary hardware, making it impossible to simply port console‑level rendering pipelines without adaptation. To address these constraints, Arm has fused neural processing units (NPUs) with the conventional graphics pipeline.
By offloading portions of the rendering workload to specialized AI accelerators, the GPU can render a fraction of the scene and let the NPU reconstruct the remaining pixels. This hybrid approach reduces the raw GPU cycles required, curtails power consumption, and lessens memory traffic—especially DRAM accesses, which are a major source of energy drain on mobile SoCs. "When the GPU only has to draw a quarter of the frame and the neural engine fills in the rest, we see up to four times the performance and efficiency gains," Lazarov explained.
"In extreme cases, the entire frame can be generated by the AI engine, meaning the GPU does virtually no work for that frame." Arm’s AI‑driven suite consists of Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD), and Neural Frame‑Rate Upscaling (NFRU). These technologies collectively upscale lower‑resolution images, add extra rays for richer lighting, and synthesize intermediate frames to boost perceived frame rates. The result is a smoother, sharper visual experience without the traditional trade‑offs between fidelity, speed and battery life. When contrasted with Nvidia’s DLSS 5, Lazarov was quick to point out a philosophical divergence.
Nvidia’s solution relies on generative neural networks that can hallucinate details, a method that has sparked debate over whether the output stays true to the developer’s artistic intent. Arm, by contrast, employs convolutional networks that simply learn how existing content should look when upscaled or interpolated. These networks are trained on real‑world data and, crucially, are released openly on platforms like GitHub and Hugging Face.
"Our models don’t imagine new content; they recognize patterns and reconstruct them based on what they have seen," Lazarov said. "Because the networks are open source, developers can fine‑tune them with their own art assets, ensuring the AI respects the unique visual style of each game. This eliminates the risk of losing creative control, something you can’t do with a closed‑box system like DLSS." The openness also means developers can retrain the models on bespoke datasets, tailoring the upscaling behavior to quirky or stylized aesthetics.
This flexibility is a stark contrast to Nvidia’s proprietary, black‑box approach where the training data and internal workings are hidden, leaving developers with a take‑it‑or‑leave‑it proposition. Arm’s mobile‑first design philosophy further distinguishes its solution.
The neural networks are optimized for the limited power, bandwidth and thermal envelopes of smartphones. Attempting to run a DLSS‑style algorithm on a phone would instantly exceed those budgets, according to Lazaror. In terms of raw efficiency, Arm claims the Mali G2‑Ultra NX delivers up to a 70 % reduction in DRAM traffic compared with traditional pipelines, translating into lower overall power draw.
In a head‑to‑head test, the previous‑generation GPU could only sustain *Neural Dawn* at about 40 fps when pushed to its limits, and fell to roughly 15 fps under normal conditions. The new chip, however, comfortably hits the 60 fps target, delivering a fluid experience even with AI‑enhanced rendering active.
To support developers, Arm has released Unreal Engine plug‑ins and is working on extending support to Unity and other engines. While Unity integration is still on the roadmap, Arm has already announced a partnership with Unity China, signaling a commitment to broader engine coverage.
For studios using custom engines, Arm provides an SDK that offers a streamlined path to integrate its AI‑native graphics stack without needing a dedicated plug‑in. Lazarov highlighted that retrofitting an existing shipped title with Arm’s technology can be accomplished in as little as three months, allowing developers to quickly roll out visual upgrades, performance improvements, or battery‑saving enhancements. The *Neural Dawn* demo itself was built by a team of fewer than twenty engineers in under eighteen months, illustrating how Arm’s tools can accelerate development cycles. Overall, Arm’s announcement positions its Mali G2‑Ultra NX and associated AI‑reconstruction framework as a compelling alternative to the closed, generative approaches championed by Nvidia.
By offering open, trainable networks optimized for the strict constraints of mobile hardware, Arm aims to empower developers to deliver console‑grade graphics on handheld devices without sacrificing power efficiency or creative control.