Under the physical power consumption constraints of mobile devices, traditional rendering methods that rely on brute-force computing power are hitting a bottleneck. The demand for high-quality graphics driven by the premiumization of mobile games, coupled with the contradiction between phone heat dissipation and battery life, is forcing chip and engine manufacturers to seek new technical solutions.
On July 28, Arm and Unity China signed a strategic Memorandum of Understanding (MoU), announcing the native integration of Arm Neural Technology into Unity China's Tuanjie Engine.
This collaboration aims to bridge the underlying hardware and software engines, leveraging AI predictive rendering to lower the barriers for developers in AI graphics development and cross-platform adaptation, with plans to expand the partnership to scenarios such as automotive smart cabins and industrial edge computing in the future.
Built-in Shader Accelerators
The mobile game industry currently faces three major development challenges: the imbalance between graphics quality and performance (pursuing high definition often leads to frame drops, overheating, surging power consumption, and poor device adaptation), high development costs and long cycles (high technical integration barriers and steep learning curves), and fragmented cross-platform experiences when extending to desktop and console platforms.
As the industry's first innovative solution to integrate dedicated AI graphics acceleration hardware at the underlying level of Arm mobile GPUs, Arm Neural Technology integrates dedicated neural accelerators into the latest 2026 generation of Arm® GPUs, and plans to integrate multiple neural accelerators inside the shader cores of next-generation GPUs.
James McNiven, Vice President of Edge AI and Endpoint Computing at Arm, stated that this design differs from the past practice of deploying AI modules as standalone chips. The built-in accelerators can share the memory system with the GPU and work in close coordination with the Shader Cores and execution engines, thereby reducing latency caused by data transmission and delivering higher energy efficiency.
Figure | James McNiven, Vice President of Edge AI and Endpoint Computing at Arm (left), and Zhang Junbo, CEO of Unity China (right), during a media interview
At the hardware level, it is possible to pair each shader core with a neural accelerator and adjust the quantity based on hardware specifications. Previously, Arm has opened early access to the ML extensions for Vulkan (VK_ARM_tensors and VK_ARM_data_graph) and open-sourced the relevant code repositories.
Data shows that this technology can reduce the GPU computing workload of mobile games by up to 50%, featuring open compatibility, easy integration and adaptation, and deep performance tuning on real devices. The freed-up computing power can be freely allocated by developers to balance power consumption, boost frame rates, or enhance visual effects.
Native Integration of Arm Neural Technology to Resolve Cross-Hardware Adaptation Issues
For game developers, cross-hardware adaptation across multi-chip platforms has always been a high-cost engineering challenge.
Zhang Junbo, CEO of Unity China, pointed out that in actual development, the ratio of creative planners and artists (liberal arts background) to engineers responsible for code implementation (science background) is often as high as 1:10 or more, resulting in extremely high engineering conversion costs. In this context, while some game companies could choose to directly call Arm's underlying APIs for self-integration, the costs would be even higher.
To address hardware fragmentation and lower the barriers to game development, the Tuanjie Engine chose to natively integrate Arm Neural Technology directly at the underlying level—including Neural Super Sampling (NSS), Neural Frame Rate Upscaling (NFRU), and Neural Super Sampling and Denoising (NSSD).
These features are designed as optional capabilities, and the game will automatically determine whether to enable them based on the device's hardware capabilities during runtime. Developers do not need to package separately for specific chips; they only need to release a single installation package to automatically adapt to different device models, achieving out-of-the-box usability.
Zhang Junbo revealed that in the future, Unity China plans to support more granular profiles in Tuanjie Engine 2.0 to generate differentiated installation packages for different hardware.
Super Resolution and Frame Interpolation: The Technical Logic of AI Rendering
Traditional rendering logic requires the engine to calculate what the next frame should present based on the current state, often requiring the computation of 60 frames per second or more, which places extremely high demands on computing performance. The intervention of AI technology essentially transforms "hard computing" into "prediction and completion."
- Super Resolution Technology: Allows the engine to render only at a lower resolution of 480p, and then upscale it to 1K, 2K, or even 4K using super resolution technology. Since the computational overhead depends on the number of pixels to be processed, using neural networks to fill in the details can achieve near-native resolution effects at a lower computational cost.
- Frame Interpolation Technology: When computing power is limited and precise frame-by-frame calculation is impossible, the engine can calculate only a portion of the frames (e.g., 30 frames), with the remaining frames generated and filled in by AI, thereby achieving a smooth visual experience of 60 frames or higher with limited computing power.
James McNiven emphasized that the integration of neural accelerators does not mean it is a replacement for traditional technology routes; Arm still needs to steadily advance the performance of traditional rendering. Developers still need to make trade-offs among performance, graphics quality, and power consumption, but neural technology provides them with a more flexible technical path.
Industry Implementation and Expansion
China's mobile games have already joined the core ranks of the global gaming market, becoming a core force driving the development of the global industry. Data released by Sensor Tower shows that in June 2026, among the Top 100 global mobile game publishers by revenue, 38 were Chinese companies, with a combined revenue share of 41.6%. Meanwhile, the market has a direct demand for the updating of underlying graphics technologies.
On the ecosystem front, as of 2025, 82% of the world's top 100 best-selling mobile games were built using Unity. Hardware devices equipped with Arm's new-generation GPU technology are expected to be launched one after another this year, meaning this technology is about to enter a substantive stage of commercial implementation.
From a longer-term perspective, AI is bringing earth-shaking changes to game productivity. Zhang Junbo believes that in the future, there may emerge "world models" trained specifically for the games themselves, training game logic into a foundational model, and ultimately using this world model to directly restore and generate the visuals and content players see in the game. This technological integration will not only bring better graphics on mobile devices, but its collaborative achievements will also be further expanded to new scenarios in the AI era, such as automotive smart cabins and industrial edge computing.