Arm, the UK‑based semiconductor and software design powerhouse, staged its Arm Everywhere China showcase in Shanghai to announce a fresh set of AI‑driven compute solutions aimed squarely at mobile devices. The centerpiece of the announcement is a new AI‑native compute platform called CSS for Mobile 2, paired with the Mali G2‑Ultra NX graphics processor – a GPU built around neural‑accelerator cores that promises higher visual fidelity while keeping heat and power draw to a minimum.
Arm positions its on‑device AI graphics pipeline as a more transparent alternative to Nvidia’s controversial DLSS 5. Rather than relying on a closed, generative model that can reinterpret a developer’s artistic intent, Arm’s solution uses convolution‑type networks that simply learn how existing content should look and then reproduce it faithfully. All of the trained models – for resolution up‑scaling, frame‑rate generation and denoising – are published openly on GitHub and Hugging Face, allowing developers to fine‑tune them with their own asset libraries and retain full control over intellectual property.
The Mali G2‑Ultra NX is the successor to the Mali G1‑Ultra, which already powers smartphones from brands such as Xiaomi, Vivo and OnePlus. While Arm has not disclosed specific upcoming devices, the new chip is engineered to give mobile developers access to compute‑heavy features that were previously the domain of consoles and high‑end PCs, most notably real‑time ray tracing. “We’ve seen a growing appetite for ray tracing in mobile games,” explains Deyan Lazarov, a member of Arm’s GPU product team within the Edge AI unit. “The technology was slow to catch on at first, but now more titles are asking for it, and the G2‑Ultra NX gives them a practical path.” One early showcase of the chip’s capabilities is the demo title *Neural Dawn*, created in partnership with Sumo Digital.
Built with Unreal Engine’s MegaLights technology, the game illustrates how the GPU can render a fraction of the frame traditionally and let the neural accelerators fill in the missing pixels. Lazarov describes the process: the first frame is rendered at 25 % resolution, with the remaining three‑quarters reconstructed by the NSS (Neural Super Sampling) engine. In the second frame, the GPU does virtually no raster work; the neuro‑accelerators generate the entire image, effectively delivering AI‑reconstructed frames at full resolution. Arm’s approach differs from Nvidia’s DLSS in several key ways.
DLSS relies on a generative network that can invent details, which sometimes leads to visual artifacts that diverge from the creator’s vision. Arm’s convolutional networks, by contrast, are trained on real‑world content and simply predict the most likely appearance of each pixel, preserving artistic intent.
Because the models are open‑source, developers can retrain them with custom data sets that reflect a game’s unique art style, ensuring that up‑scaled frames match the original aesthetic without sacrificing IP control. The open‑source philosophy also extends to tooling. Arm supplies plugins for Unreal Engine today and is working on Unity support, with a broader SDK that can be integrated into custom engines. This SDK abstracts the heavy lifting of neural inference, offering a streamlined path for studios to adopt AI‑native graphics without rewriting large portions of their rendering pipelines.
Lazarov notes that an existing shipped title can be upgraded with Arm’s tech in as little as three months, giving developers a rapid route to improve visual quality, performance, or battery life. Performance metrics underline the efficiency gains. Arm claims the G2‑Ultra NX delivers up to four times the performance and four times the energy efficiency of its predecessor, while slashing DRAM traffic by roughly 70 %.
Reducing memory accesses is crucial on smartphones, where external memory reads are a major source of power consumption. In a head‑to‑head test, the older GPU could only sustain *Neural Dawn* at 40 fps under ideal conditions and dropped to 15 fps in typical gameplay, whereas the new chip maintains a steady 60 fps, even when the AI reconstruction pipeline is fully engaged. Beyond *Neural Dawn*, several upcoming mobile titles will leverage the chip, including Tencent’s *Arena Breakout Infinite*, NetEase’s *Where Winds Meet* and Infold Games’ *Infinity Nikki*.
These games illustrate the “mobile‑first, desktop‑look” philosophy Lazarov describes: developers want their mobile releases to rival the visual polish of PC or console experiences, yet they must operate within the stricter thermal and power envelopes of a handheld device. The Mali G2‑Ultra NX’s neural‑accelerated rendering reduces the GPU’s workload, freeing up power headroom and allowing developers to avoid the compromises that traditionally plague mobile graphics.
Arm’s broader vision is to create an AI‑native ecosystem that empowers creators across the entire device stack. By making the neural models open, providing cross‑engine plugins, and delivering a robust SDK, Arm hopes to lower the barrier to entry for AI‑enhanced graphics. Lazarov sums it up: “We want developers to take the efficiency gains we provide and decide how to spend the saved budget—whether that’s higher fidelity textures, smoother frame rates, longer battery life, or any combination thereof.” In summary, Arm’s mobile‑first AI reconstruction technology offers a transparent, developer‑friendly alternative to Nvidia’s black‑box DLSS 5.
Its open convolutional networks, integrated neural accelerators, and focus on power‑constrained environments aim to bring console‑level visual effects to smartphones without sacrificing battery life or thermal stability, marking a significant step toward truly AI‑native graphics on everyday devices.