Arm Unveils Mobile-First AI Reconstruction Technology, Distinct from DLSS 5's 'Black Box' Approach

Arm, the British semiconductor and software design firm, has announced its latest advancements in AI computing at the Arm Everywhere China event in Shanghai. The company aims to establish a comprehensive AI-native computing ecosystem with its new AI-native compute platform CSS for Mobile 2 and the Arm Mali G2-Ultra NX GPU. This new chip is designed around neural computing, promising enhanced graphics performance with significantly reduced heat and power consumption.

Arm positions its on-device AI technology as a more open alternative to Nvidia's DLSS 5, enabling developers to maintain closer control over their creative vision. While specific devices featuring the Mali G2-Ultra NX GPU remain unconfirmed, previous Mali G1-Ultra chips have appeared in devices from Xiaomi, Vivo, and OnePlus.

Deyan Lazarov from Arm's GPU product team explained that the Mali G2-Ultra NX, developed over two and a half years, is intended to empower mobile developers to utilize compute-intensive technologies like ray tracing. He noted a growing interest in ray tracing for mobile titles, despite initial slower adoption.

As previously reported, the new GPU supports Unreal Engine features such as MegaLights. Arm collaborated with Sumo Digital to create "Neural Dawn," a game demonstrating the Mali G2-Ultra NX's capabilities. Lazarov stated the chip is designed for "mobile-first" titles that aim for desktop-like visual quality and for cross-platform games seeking a consistent player experience across devices.

Upcoming games expected to leverage the new chip include Tencent's "Arena Breakout Infinite," NetEase's "Where Winds Meet," and Infold Games' "Infinity Nikki."

Addressing Thermal and Power Constraints

Lazarov highlighted the unique challenges of mobile devices, such as limited thermal dissipation and power budgets compared to consoles and PCs. Arm's solution integrates neural processing with traditional rendering pipelines to reduce the GPU's workload and power draw, enhancing overall system efficiency.

This efficiency allows developers to overcome limitations and reduce the need for extensive trade-offs. Arm is promoting an open software approach and providing a suite of tools to encourage AI-native graphics adoption. Lazarov emphasized their collaboration with the ecosystem to create a developer-friendly AI integration that offers significant efficiency gains, allowing developers to reinvest this budget into visual fidelity, performance, or battery life.

Key technologies include Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD), and Neural Frame Rate Upscaling (NFRU). Arm's technology reconstructs images and generates frames using AI, effectively upscaling content and casting additional rays.

Lazarov illustrated this with an example: in one frame, the GPU renders only a quarter of the image, with the rest reconstructed by neural accelerators using NSS. In another scenario, the GPU performs no traditional rendering, with all pixels AI-reconstructed by the neural accelerators.

A Contrast to DLSS 5

Arm differentiates its technology from Nvidia's DLSS 5, which has faced criticism for potentially deviating from developers' original artistic intent. Lazarov clarified that Arm's networks are convolution-based and not generative; they don't "imagine" or "create" but rather "know what things need to look like" based on trained data.

Arm's networks are open and publicly available on GitHub and Hugging Face, allowing developers to retrain them using their own game data. This enables them to capture their game's unique art style and ensure that upscaling and frame generation are consistent with their vision, thereby preserving IP control.

"DLSS is not open. It's not like our network. It's proprietary, it's closed, it's a black box, you don't know what's happening, you don't know what it's trained on. You cannot retrain it on your own [data]. You integrate it, you're happy with the results, or you're not. It's a bit of a take it or leave it, whereas [ours] is take it and make it so that you like it."

This open approach contrasts sharply with DLSS 5's proprietary, "black box" nature. Furthermore, Arm's technology is designed with mobile constraints in mind, including power and bandwidth, making it unsuitable for phones if adapted from desktop solutions like DLSS.

Hyper-Efficient Performance

Lazarov highlighted the Mali G2-Ultra NX's focus on power efficiency, noting significant reductions in DRAM traffic—a major contributor to mobile power consumption. This leads to substantial power savings.

As an example, Lazarov cited "Neural Dawn," which ran at a mere 15 fps on last year's GPU under normal conditions (and 40 fps at maximum load). The new GPU, however, can easily achieve 60 fps.

"In the future we're going to be where the developers are, we're going to make it as easy as we can for them."

Arm provides plug-ins for Unreal Engine and plans to support other game engines, with Unity being a priority, especially following their partnership with Unity China. For custom game engines, Arm offers an SDK for direct integration.

Developers can integrate Arm's AI-native graphics into existing titles, with new versions potentially shipping in as little as three months. Lazarov also pointed out that "Neural Dawn" was developed by a small team in under 18 months, demonstrating how Arm's technology enables developers to quickly enhance existing games or launch groundbreaking new titles.