Neural Super Sampling (NSS), a next-generation AI-powered upscaling solution from Arm is released for graphics and gaming developers to start experimenting today!

NSS is designed for real-time performance on future mobile devices with Arm Neural Technology. However, latency depends on implementation factors such as GPU configuration, resolution, and use case. In our Enchanted Castle demo video below, NSS reduced GPU workload by 50 percent. The model rendered at 540p and upscaled to 1080p in just 4ms in sustained performance setup.

Neural Super Sampling (NSS) is a parameter prediction model for real-time temporal super sampling developed by Arm, optimized for execution on Neural Accelerators (NX) in mobile GPUs. It enables high-resolution rendering at a lower compute cost by reconstructing high-quality output frames from low-resolution temporal inputs. NSS is particularly suited for mobile gaming, XR, and other power-constrained graphics use cases.

Check out the Neural Graphics Dataset: A collection of reference images and image sequences along with the corresponding motion, depth and other data required to train, validate and test neural super sampling algorithms.

The current version of the dataset includes a limited set of data for Neural Super Sampling to demonstrate how the NSS model development flow works. While this flow does not yet provide a comprehensive dataset for complete model (re)training, stay tuned for future releases of the Neural Graphics Model Gym where Arm will provide tools to capture and convert content for use in model training and retraining.

NSS has been integrated into Unreal Engine via two plugins, the NSS Plugin for Unreal® Engine and Unreal® NNE Plugin for ML extensions for Vulkan .

Source link