Arm builds neural acceleration into new Mali G2-Ultra NX mobile GPU

# Arm builds neural acceleration into new Mali G2-Ultra NX mobile GPU

*Event date: 2026-09-08*

Arm has introduced the Mali G2-Ultra NX, a mobile graphics processor designed to run neural rendering alongside conventional graphics and compute work. The company describes it as its first AI-native Mali GPU and is positioning the design as a way to bring more demanding game visuals to phones without exceeding their tight power, heat and memory limits.

The architecture places neural accelerators inside the shader cores and shares the GPU's memory system, caches and control structures. That integration is intended to reduce the movement of data between separate processing blocks. In practice, neural techniques can reconstruct image detail, generate intermediate frames and refine rendered output while lowering the amount of work required from other parts of the system.

Arm highlighted a demonstration called Neural Dawn, developed with Sumo Digital, that combines its Neural Frame Rate Upscaling and Neural Super Sampling technologies in a production-style game pipeline. The company reports up to four times better performance efficiency and as much as 70% less external-memory traffic than native rendering in that test. Those figures are vendor results from a specific demonstration rather than independent measurements of shipping devices.

Beyond neural processing, the G2-Ultra NX introduces what Arm calls the largest change to the Mali instruction-set architecture in seven generations. The execution engine provides up to twice as many registers per warp. Arm reports benchmark performance gains of up to 24% and non-AI gaming gains of 14% compared with its previous generation, while neural frame-rate upscaling is designed to support longer sessions at frame rates as high as 120 frames per second.

The GPU also carries Arm's third generation of hardware ray tracing. The company says it cut DRAM traffic by up to 13% in leading ray-tracing benchmarks. Support for Opacity Micromaps is meant to process detailed transparent geometry, such as foliage and layered fabric, more efficiently. In one Arm demonstration, that feature increased frame rate by 30% and reduced the ray-tracing workload by up to 70%.

Hardware adoption will also depend on software support. Arm began releasing neural-graphics technology and an open development kit ahead of the processor, giving engine and content developers time to experiment. Its current development kit includes neural-scaling and frame-generation components, Vulkan machine-learning extensions, Unreal Engine plug-ins, tools for custom engines, and profiling and model-optimization utilities.

Tencent Games and Unity China are among the named ecosystem participants. Tencent's Magic Dawn technology and Unity China's Tuanjie Engine are integrating Arm's neural tools, while the developers of Where Winds Meet plan to use the technology through their Messiah Engine.

Arm has not identified consumer devices, prices or release dates in the supplied announcement. The technical claims therefore describe the processor's intended capabilities and Arm's own test results, not guaranteed performance in a particular phone. Device cooling, memory configuration, game-engine implementation and chip design will determine how much of the promised improvement reaches players.