adminArm mobile AI graphics is designed to help phones deliver richer visuals without the heat and battery demands of console- or PC-class hardware. At its Arm Everywhere China event in Shanghai, the company outlined how its Mali G2-Ultra NX GPU combines neural processing with conventional graphics rendering to reconstruct images and generate additional frames.
Arm says the approach is built for mobile power, memory-bandwidth, and thermal limits. It also emphasizes that developers can inspect and adapt its AI models, rather than relying on a closed system. The Mali G2-Ultra NX is part of Arm’s wider AI-native computing plans, alongside its CSS for Mobile 2 platform.
Image credit: Arm
How Arm mobile AI graphics works
Deyan Lazarov, who works on Arm’s GPU product team in its Edge AI business unit, says development of the Mali G2-Ultra NX has taken two and a half years. One goal is to make demanding techniques such as ray tracing more practical in mobile games. Lazarov says interest in ray tracing among mobile game developers is growing.
Arm’s system combines neural processing with the traditional rendering pipeline. Its Neural Super Sampling (NSS), Neural Super Sampling and Denoising (NSSD), and Neural Frame Rate Upscaling (NFRU) technologies can reconstruct image detail, assist with ray processing, and create additional frames.
In one example described by Lazarov, the GPU renders a quarter of an image in one frame, while neural accelerators reconstruct the other three quarters using NSS. In a following frame, the traditional GPU rendering workload may be eliminated, with the image produced by the neural accelerators. The exact workload depends on how the technology is integrated into a game.
For developers, this can create room to prioritize visual detail, performance, battery life, or a balance of the three. Those trade-offs matter on phones, where heat cannot be dissipated as readily as it can on a PC or console and gameplay typically relies on battery power.
A different approach from DLSS 5
Arm distinguishes its reconstruction models from Nvidia’s DLSS 5 by describing them as non-generative convolutional networks. The company says its networks reconstruct images based on patterns learned from real content rather than inventing new visual details.
Arm says its resolution and frame-generation networks are publicly available on GitHub and Hugging Face. Developers can also capture data from their own games and retrain a network to better match a distinctive art style. This gives studios more control over how reconstructed frames look, according to Lazarov.
Image credit: Nvidia/Capcom
That openness is a key contrast with the proprietary, closed nature Arm attributes to DLSS 5. Arm’s mobile-first design is another distinction: its network is intended to operate within phone-specific limits for power use and memory bandwidth. Lazarov says a technology designed for more powerful hardware could exceed those budgets if run on a phone.
When evaluating an AI graphics solution, developers should consider not only image quality but also how much control they have over model behavior, what data is used for training, and the impact on device power and performance.
Performance, efficiency, and supported games
Arm reports that the Mali G2-Ultra NX delivers four times the performance and four times the efficiency of its predecessor in the stated comparisons. It also claims a 70% reduction in traffic to DRAM, an important factor because accessing external memory can contribute significantly to a mobile device’s power use.
Arm uses Neural Dawn, developed with Sumo Digital, to demonstrate the technology. Lazarov says the previous-generation GPU ran the game at 40 frames per second when pushed to its limit and at 15 fps under normal conditions. He says the Mali G2-Ultra NX can run it at 60 fps.
Neural Dawn, developed with Sumo Digital, showcases Arm’s technology. Image credit: Sumo Digital/Arm
Games named as potential users of the new GPU include Arena Breakout Infinite from Tencent, Where Winds Meet from NetEase, and Infinity Nikki from Infold Games. Arm has not confirmed which devices will use the Mali G2-Ultra NX. Its earlier Mali G1-Ultra chip appears in devices from Xiaomi, Vivo, and OnePlus, but that does not confirm which manufacturers will adopt the new GPU.
Developer tools and engine support
Arm currently offers plug-ins for Unreal Engine and says it plans to support additional game engines. Unity is among the engines under consideration, while Arm has also announced a partnership with Unity China. For teams using custom engines, the company provides an SDK intended to support direct integration.
Arm says existing games could be updated with its technology in as little as three months, depending on the project. Neural Dawn was made by a team of fewer than 20 people in less than 18 months, according to Lazarov. These figures illustrate Arm’s claims about development workflows, but actual integration time will vary with a game’s engine, rendering setup, and testing needs.
For studios considering AI reconstruction, a practical first step is to test the tools on representative gameplay scenes. Checking performance, visual consistency, battery use, and compatibility across target devices can help determine whether the technology delivers a worthwhile improvement.
What Arm mobile AI graphics could mean for players
Arm mobile AI graphics aims to narrow the visual gap between mobile games and larger-platform releases while respecting the limits of phone hardware. Its reconstruction methods, public models, and developer retraining options offer a different approach from closed, generative systems. The real-world result will depend on device support, game integration, and whether the promised efficiency gains hold up in finished titles.
