Arm released "Arm Neural Technology" on SIGGRAPH today (13th), the technology was first created in the industry and will introduce dedicated neural accelerators to the Arm GPU launched in 2026.

"Arm Neural Technology" can increase the performance of GPUs for image rendering to a higher level; for the more complex mobile device content in the current mobile gaming domain, it reduces the load on GPUs by up to 50%. In addition, Arm has launched the world's first fully open neural image development suite, integrating AI rendering into existing workflows, allowing developers to start developing a year before the hardware is in trouble.

Cai Wunan, president of Arm Mobile Strategy, pointed out that mobile phones are currently the first wave of devices to be introduced to change the graphics processing methods of Android flagship products and game phones. We expect to see more of this type of products in 2026. Arm also hopes to add ecological development workers to use them through the opening development kit.

Geraint North, an academician of Arm's artificial intelligence and developer platform, said that this is the first time that Arm has implemented PC-level AI imaging performance in optimizing mobile devices, laying the foundation for future mobile device-side AI innovation. "Arm Category neural technology is not only suitable for the game field, but also brings a practical impact on applications such as the loading of neural cameras, providing developers with a powerful tool for large-scale real-world imaging technology in application scenarios such as image enlargement to path tracking.

In addition, Arm NSS is an AI-driven image amplification engine that makes full use of various resources within the development suite. This is based on Arm's Precision Super Resolution Technology (Arm ASR), which has been used in game studios such as "Fortress Heroes" and "Infinite Warm".

It is reported that Arm NSS technology works very similarly to NVIDIA's DLSS technology. Both use AI-driven neural networks to increase low-resolution shadow grids to higher resolutions, reducing GPU computing burden while maintaining high-quality visual effects. The difference between the two is that DLSS runs on NVIDIA Tensor Cores in a desktop GPU, while NSS uses a dedicated neural accelerator integrated into the Mali GPU and optimizes the mobile devices only.

Arm NSS can zoom in from 540p to 1080p at four milliseconds per frame, almost no difference from native painting. Compared to traditional full-frame rendering methods, developers can reduce GPU load by up to 50% through Arm NSS, which can be used to reduce overall power consumption of the game, improve the rate or enhance visual effects. This technology can also allow developers to use AI to preserve surface detail, lighting and motion clarity, thereby flexibly balancing visual fidelity and energy consumption according to game needs.

Arm said that all Arm-type neural technologies in the suite will be fully open, including the model architecture, powers and tools needed to retrain models in the game studio. Currently, the development kit has been supported by partners such as Enduring Games, Epic Games, Netease Games, Sumo Digital, Teng Games and Traverse Research.

Arm plans to further expand its neural technology application pathways in 2026, launch Neural Frame Rate Upscaling that doubles the current rate of AI without doubled rendering, and Neural Super Sampling and Denoising that uses AI to implement real-time path tracking on mobile devices, and each pixel only needs to emit less light. Both applications will be launched before the hardware industry.

Extended reading: "The United States needs Intel!" Former CEO shouted: Only Intel can save the American chip industry, not Telco Korean media claims that Samsung will use NVIDIA to test and supply 12-layer HBM3E? Officials refuse to respond positively