Announced at SIGGRAPH, Arm has introduced neural technology that brings dedicated neural accelerators to Arm GPUs beginning in 2026. The company calls this an industry first, and it is positioned to change how mobile devices render graphics by moving more intelligence on device. The basic idea is simple: pair neural compute with traditional GPU pipelines to accelerate rendering tasks and to unlock new, smarter features without depending on the cloud.
Arm says the neural accelerators target today´s most intensive mobile content, starting with mobile gaming, and can reduce GPU workload by up to 50 percent for those scenarios. That reduction is meant to translate into higher frame rates, richer visuals and less battery drain, while also enabling frame-to-frame enhancements and other perceptual improvements that were previously too costly on mobile hardware. The announcement frames this as a foundation for broader on-device artificial intelligence innovation across gaming, photography, productivity and other workloads.
To give developers an early start, Arm is also launching the world´s first publicly available neural graphics development kit. The kit is designed to slot into existing workflows so studios and creators can prototype and integrate AI-powered rendering now, roughly a year before hardware ships. Crucially, Arm has committed to openness: the kit and the neural technology will be fully open, with model architecture, weights and the tools studios need to retrain models all made available.
Partners already shown supporting the development kit include Enduring Games, Epic Games with Unreal Engine, NetEase Games, Sumo Digital, Tencent Games and Traverse Research. That early ecosystem backing aims to accelerate practical experimentation and production readiness, giving developers time to adapt engines, pipelines and assets to neural-assisted rendering. For studios and tool makers, the combination of open models and an advance development kit means exploration can begin now, while silicon and shipping products follow in 2026.
