During SIGGRAPH 2025, Arm announced the launch of its Neural Technology roadmap, aiming to enable desktop-level AI graphics processing capabilities in mobile devices using Arm-based GPUs, launched in 2026. By collaborating with a wider range of industry players through an open ecosystem, this will lower the barrier to entry for development and application development. This blueprint is also expected to significantly reduce power consumption during high-performance display computing on mobile devices, thereby driving growth in the mobile gaming market.
Cai Wunan, director of market strategy for Arm's Terminal Products Division, pointed out that a significant portion of consumers in China and the Asia-Pacific region's demand for mobile device performance stems from their need for smooth gaming. This is especially true for games with desktop-like graphics, such as miHoYo's "Genshin Impact" and "Honkai Impact 3," as well as battle royale games like "Fortnite." These games place increasingly high demands on GPU performance, while also posing a severe challenge to mobile device battery life.
Therefore, as AI gradually becomes an application technology for mobile devices and is used for camera shooting, content creation, or office productivity improvement, gaming has also become a key application project. Arm has also proposed ray tracing, variable rate shading (VRS), deferred vertex shading (DVS), andArm Accurate Super Resolution (ASR)Technologies such as NVIDIA GeForce GTX 1060 and NVIDIA GeForce GTX 1060 are used to enable the GPU graphics performance of mobile devices to continuously approach desktop levels.
Arm's Neural Super Sampling (NSS) technology, introduced this time, further integrates AI to transform low-resolution image input into high-quality images within 4 milliseconds, delivering higher image quality under the same conditions. Compared to traditional full-frame rendering, Arm claims NSS can reduce GPU workload by up to 50%, while also effectively reducing power consumption and device heat.
According to Arm, if compared with the performance of desktop graphics cards, the technology proposed this time is roughly equivalent to NVIDIA's DLSS 2 presentation effect, which means it can achieve ray tracing display performance close to that of the GeForce RTX30 series graphics cards. At the same time, the AI model occupies only 15 sets of parameters, so it is claimed to be able to perform calculations with lower power consumption and higher efficiency, and can also be used in general smartphone hardware configurations.
New Applications of Neural Imaging: Frame Rate Upscaling and Noise Removal
Arm also announced the launch of Neural Frame Rate Upscaling, which generates intermediate frames to increase the native 30fps content to 60fps, making the actual output image more stable.
Another feature, Neural Supersampling and Noise Reduction (NSSD), uses a small number of rays and AI to infer details in ray tracing scenes, thereby improving the high computational cost of ray tracing to present the entire image. It also allows ray tracing effects to be presented more efficiently and reduces overall device power consumption and heat generation.
Vulkan extension and cross-platform open architecture
To drive industry adoption of this technology, Arm is providing machine learning extensions for the widely used Vulkan API. These include support for using images as tensor sources, integrating neural network-like technologies directly into the rendering process, and establishing the VGF (Vulkan Graph Format) standard, enabling AI models to be more widely used across different Vulkan API-compatible mobile devices.
Developers can quickly integrate the proposed NSS technology into engines like Unreal Engine and Godot through the Arm Neural Graphics Development Kit. Arm has also made the AI model format and weight data open source through the Hugging Face platform, providing a PC simulation environment and complete example code for easy verification and deployment.
Locking in the mobile game market layout with an open strategy, but there are still certain conditions and restrictions
Arm announced that it will make its neural technology available in an open form, but there are still some restrictions, including that the GPU must include a neural accelerator design, which means that only Arm's new generation GPUs expected to be launched in 2026 can fully use it. Another condition is that it must be compatible with Vulkan technology.
As for Arm-based neural technology, it is currently only available for Android platform devices, so there are no plans to use it for devices such as Windows on Arm.
However, developers can optimize model architecture, training tools, and runtime data according to their needs while retaining underlying control. Many Chinese game developers use specially modified versions of Unreal Engine, while some, such as Tencent and NetEase, use their own engines. This allows for library integration, allowing them to use Arm-based neural technology to enhance game performance.
