Arm, the British chip and software design powerhouse, staged its Arm Everywhere China showcase in Shanghai to announce a suite of fresh AI‑computing offerings aimed squarely at mobile devices. Central to the reveal is the AI‑native compute platform CSS for Mobile 2 and the new Mali G2‑Ultra NX GPU, a processor built around neural‑accelerator cores that promises higher graphics performance while keeping heat and power draw low. Arm positions its on‑device AI graphics pipeline as a more transparent and developer‑friendly alternative to Nvidia’s contentious DLSS 5, aligning more closely with the original intent of game creators. While Arm did not name specific products that will ship the Mali G2‑Ultra NX, the predecessor Mali G1‑Ultra already powers smartphones from Xiaomi, Vivo and OnePlus, suggesting a similar market trajectory.
Deyan Lazarov, a senior member of Arm’s GPU product team within the Edge AI division, explained that the G2‑Ultra NX has been in development for roughly two and a half years. Its purpose is to give mobile developers access to compute‑heavy features such as real‑time ray tracing, which has historically been slow to appear on phones but is now gaining traction as more titles experiment with the effect. According to Lazarov, the chip is aimed at "mobile‑first" games that still aspire to look like desktop experiences, as well as cross‑platform titles that need a consistent visual quality across consoles, PCs and handhelds. He cited upcoming games that will leverage the GPU, including Tencent’s Arena Breakout Infinite, NetEase’s Where Winds Meet and Infold Games’ Infinity Nikki.
The challenge, he noted, is that phones operate under stricter thermal and power budgets than consoles or PCs; they cannot dissipate heat as freely and must run continuously on battery power. Arm’s answer is to blend neural processing with the traditional graphics pipeline. By offloading portions of the rendering workload to dedicated neuro‑accelerators, the GPU can render fewer pixels directly and let AI reconstruct the remainder.
Lazarov illustrated this with a two‑frame example: in the first frame the GPU draws only a quarter of the image while the neuro‑accelerators fill in the other three‑quarters using Neural Super Sampling (NSS). In the second frame, the GPU does no conventional rendering at all; the entire frame is generated by the AI engines, dramatically cutting power consumption and freeing up resources for higher fidelity, smoother performance, or longer battery life.
Arm bundles this capability with an open‑source software stack that includes Neural Super Sampling, Neural Super Sampling and Denoising (NSSD) and Neural Frame‑Rate Upscaling (NFRU). Unlike Nvidia’s DLSS 5, which relies on proprietary, generative networks that can introduce artefacts and limit developer control, Arm’s models are convolution‑based, trained on real‑world content and released publicly on GitHub and Hugging Face.
This openness allows developers to fine‑tune the networks with their own art‑style datasets, ensuring that upscaled textures and inserted frames match the game’s unique visual language without risking intellectual‑property loss. Lazarov emphasized that Arm’s approach removes the “black‑box” nature of DLSS. Developers can download the exact network used for resolution upscaling or frame‑rate generation, retrain it on bespoke data, and integrate it back into their pipeline, guaranteeing that the AI behaves exactly as intended.
By contrast, DLSS offers a take‑it‑or‑leave‑it proposition: the model is closed, its training data undisclosed, and it cannot be customized for niche artistic directions. Because the Mali G2‑Ultra NX was engineered with mobile constraints in mind, its neural networks operate within strict power, bandwidth and memory limits. Lazarov pointed out that DLSS would simply exceed a phone’s budget, even if the software were ported, leading to unacceptable thermal spikes and battery drain.
Arm’s solution reduces DRAM traffic by roughly 70 %, a major source of power consumption on handhelds, while delivering up to four times the performance and efficiency of its predecessor. Performance gains are tangible: the previous‑generation GPU could only push the demo title Neural Dawn to about 40 fps at its limit and dropped to 15 fps under normal conditions. The new G2‑Ultra NX runs the same game smoothly at 60 fps, showcasing the practical impact of AI‑assisted rendering on real‑world workloads.
From a developer tools perspective, Arm currently offers Unreal Engine plug‑ins and a roadmap that includes Unity support, following a recent partnership with Unity China. For studios using custom engines, Arm provides a comprehensive SDK that streamlines integration without requiring bespoke plug‑ins.
Lazarov noted that retrofitting an existing shipped title with Arm’s AI graphics can be accomplished in as little as three months, allowing publishers to quickly deliver visual upgrades or performance improvements. The demo game Neural Dawn, built by a team of fewer than 20 developers in under 18 months, exemplifies how Arm’s technology can accelerate development cycles. By handling much of the heavy lifting in rendering and frame generation, the AI‑native pipeline lets small teams achieve console‑grade visuals on mobile hardware without the usual trade‑offs.
In summary, Arm’s mobile‑first AI reconstruction strategy diverges sharply from Nvidia’s DLSS 5 model. It relies on open, non‑generative convolutional networks that can be retrained for specific art styles, operates within the tight power and memory envelopes of smartphones, and offers developers a transparent, flexible toolkit to enhance graphics quality, frame rates or battery life as they see fit.
This philosophy aims to give mobile developers the freedom to create richer, more consistent gaming experiences across devices without sacrificing efficiency or control.