Arm Unveils Dedicated Neural Accelerators for GPUs, Ushering in a New Era of AI-Powered Graphics

At the SIGGRAPH 2025 conference, Arm made a landmark announcement, revealing its plan to integrate dedicated neural accelerators into its GPUs starting from 2026. Crucially, these advanced GPUs are expected to be available in consumer devices shipping by the end of next year.

Embracing a software-first strategy to fast-track development, Arm has already launched the world's first publicly available Neural Graphics Development Kit ahead of the hardware release. This initiative accelerates the integration of Arm's rendering features by making all necessary neural technologies—including model architectures, weights, and tools for retraining models—completely open and accessible to developers.

A key application showcased for these dedicated neural accelerator units is the new Arm Neural Super Sampling (NSS) technology. This next-generation graphics upscaling engine combines ASR (Advanced Super Resolution) with AI to deliver impressive results. It can upscale resolution from 540p to 1080p with a mere 4ms per-frame latency, all while maintaining image quality that is nearly indistinguishable from native rendering.

The benefits of NSS are significant. Arm states that the technology can reduce the GPU workload by up to 50% in gaming scenarios. This frees up the power budget, allowing for lower overall power consumption during gameplay, achieving higher frame rates, or enhancing other visual effects. Ultimately, it enables a superior balance between image quality and power efficiency.

Looking further ahead, Arm plans to continue its push into neural technology in 2026. The company is developing a Neural Frame Rate Boost technology, which aims to double frame rates without doubling the rendering load. Additionally, a Neural Super Sampling and Denoising technology is in the works to enable real-time path tracing on mobile devices using a lower number of rays per pixel.

Back to blog

Related Posts

  • _{region} ( _{language} )

    • _{area}

    _{region}
    _{language}