New Patent: Sony To Take PlayStation 5 To The Next Level With AI Upscaling

What has been working for a long time on the PC under the heading Deep Learning Super Sampling (DLSS), Sony now seems to want to implement for the PlayStation 5. The idea: A neural network can enable unprecedented details in upscaling.

AI makes upscaling possible with more details and at higher frame rates

If a console is to display content on 4K, content is created in a lower resolution and then extrapolated accordingly. This upscaling can lead to loss of detail or loss of frame rate depending on the technology used. As Golem reports, a patent from Sony suggests that an upscaling method is being used that uses artificial intelligence to significantly improve the results.

The patent does not yet conclusively show how Sony could implement the new method on the PlayStation 5. The PlayStation 4 Pro already had hardware specially developed for this purpose on board with the Object-ID-Buffer. It is conceivable that a similar approach will be followed with the PlayStation 5.

In principle, Sony uses the same process that Nvidia uses for its DLSS feature, which is exclusively provided on RTX cards. For this purpose, a neural network with the high-resolution source material, ground truth, is trained with the aim of enabling graphics processors to carry out a loss-free projection of lower resolutions. “The INT8 format is usually used for this, which also supports the graphics units of the chips on Playstation 5 and Xbox Series X”.

Upscaling across the industry

AI upscaling is a method that many companies have been researching and developing for several years. Microsoft, for example, had already presented an interface for DirectX 12 two years ago under the heading DirectML, with which AI-based upscaling in Forza Horizon 3 could be demonstrated. The Xbox Series X brings DirectML support.

With the Sony patent, it seems clear that all next-gen consoles in one form or another will rely on AI support for upscaling. It will be interesting to see how the technology works in this area. Only with version 2.0 did DLSS achieve a status that reconciles high frame rates and good detail reconstruction.