ARM unveils NSS — neural resolution scaling technology for weak devices

ARM’s neural supersampling technology will be integrated into processors starting in 2026. NSS is closer to AMD FSR4 because it requires an artificial intelligence calculator.

ARM has presented the result of its long work on the development of resolution scaling technology for low-power devices. Earlier this year, the company announced ARM Accuracy Super Resolution (ASR), based on AMD FidelityFX Super Resolution 2.0. The new neural technology develops it with artificial intelligence-based acceleration.

Starting next year, the new ARM chips will begin to integrate NPU accelerators that will provide more than 10 TOPS/W for neural workloads. NSS replaces many fixed supersampling algorithms with a lightweight artificial intelligence model. It is capable of scaling images from 540p to 1080p in about 4ms per frame, reducing GPU load by up to 50% compared to native rendering.

NSS is based on ASR, which is already used in games such as Fortnite and Infinity Nikki. The new technology improves image clarity, stability, and handling of shadows, fine geometry, and particle effects. NSS runs on a graphics processor with shared memory, uses INT8 quantization to accumulate data through training, correction, and targeted pixel-by-pixel filtering without reactive masks.

Developers can access the Neural Graphics Development Kit, which includes an Unreal Engine plugin, Vulkan emulation, profiling tools, open source models on GitHub, and Hugging Faceand Arm’s machine learning extension for Vulkan. Partners such as Epic Games, Tencent, NetEase, Sumo Digital, and Enduring Games are already working with it.

Sources: ARM, VideoCards