In terms of the technical specifications, Sony’s next-gen PS5 console appears to be somewhat underwhelming. The single-precision (FP32) performance rating is a moderate 10.28 TFLOPs, just slightly more than the midrange Radeon RX 5700 XT which offers 9.75 TFLOPs while the Xbox Series X is also a tad bit ahead at 12 TFLOPs.
However, it would seem that Sony has a secret sauce prepared to make up for these shortcomings. As per a patent spotted on Free Patents Online, the PS5 may utilize an AI-based algorithm for upscaling the in-game resolution to 4K.
This will be especially useful in games using real-time ray-tracing as there’s no way you can get 60 FPS at native 4K with the tech enabled even on the latest hardware. The description of the upscaling technology sounds very similar to NVIDIA’s DLSS. Here’s the abstract:
An information processing device for acquiring a plurality of reference images obtained by imaging an object that is to be reproduced, acquiring a plurality of converted images obtained by enlarging or shrinking each of the plurality of reference images, executing machine learning using a plurality of images to be learned, as teaching data, that include the plurality of converted images, and generating pre-learned data that is used for generating a reproduction image that represents the appearance of the object.
What’s unclear is whether there will be any dedicated hardware required for this upscaling method. NVIDIA’s Turing GPUs have Tensor cores (one per SM) that help in reducing the overhead imposed by DLSS and from what we’ve heard AMD’s Navi 2x GPUs might leverage a similar set-up.
However considering that the PS5 GPU is an intermediate between RDNA 1 and 2, the console will likely see a software implementation of the same. It’ll be interesting to see how it holds up compared to NVIDIA’s proprietary DLSS 2.0 technology which uses dedicated hardware instead: