A Deeper Look at PlayStation 5 Pro's Updated PSSR

This analysis examines the technical evolution of Sony’s PlayStation Spectral Super Resolution (PSSR) technology, as detailed in a recent presentation at the 2026 SIGGRAPH conference. The primary objective of the update was to address performance limitations and visual artifacts, such as temporal noise, that hindered the initial version of the upscaling system. By re-architecting the pipeline, Sony has shifted the workload to better align with the strengths of both AI models and traditional GPU processing.

The original PSSR implementation utilized a single Colour Prediction Network (CPN) to manage both image upscaling and temporal anti-aliasing. This approach proved inefficient, as the neural network was tasked with performing complex mathematical operations that are better suited for standard GPU execution. Furthermore, the reliance on 8-bit efficiency created challenges for high-precision HDR color values, and the system struggled with temporal stability when handling particles and translucent materials at high upscaling factors, such as 720p to 2160p.

In the updated version, the architecture has been streamlined to reduce the burden on the AI model. The reconstruction mathematics have been offloaded to the GPU, allowing the AI to focus exclusively on temporal and spatial data generalization. The system now employs a Kernel Predicting Network (KPN) rather than a CPN, which calculates blend weights to denoise local pixel data. This architectural refinement allows for more effective resource allocation, resulting in improved image quality, reduced temporal noise, and a more efficient training process. These changes enable the console to maintain high-fidelity output while freeing up GPU resources for other rendering tasks.