Sony confirms the possibility of using special GPUs on PS5 Pro to develop AI with AMD

Mark Cerny, chief architect of Sony Interactive Entertainment (SIE, Sony Interactive Entertainment), has revealed more hardware details for next-generation host PlayStation 5 Pro (PS5 Pro), including a performance upgraded AMD RDNA 2.x architecture GPU, and a machine learning/artificial intelligence program code-named "Amethylst" with AMD.

The focus of PS5 Pro performance improvement is still on three pillars, including a more powerful GPU, advanced ray tracing and AI-powered PSSR super-resolution capabilities.

The GPU adopts a customized AMD RDNA 2 architecture, which Sony named RDNA 2.x. It has advanced features that only imported by some RDNA 3 architectures, and has increased by 67% from 18 sets of WGPs (36 sets of CUs) with built-in GPUs on the original PS5 to 30 sets of WGPs (60 sets of CUs).

Sony stressed that the reason they don’t want to upgrade their architecture to RDNA 3 is because they think that PS5 and PS5 Pro are the same generation of products and do not want the game to launch different versions to match the two models. Existing games do not need to specifically adjust the rendering and shading methods, and can use GPU resources smoothly.

The memory is not listed in the exact specifications and configurations. It is only mentioned that the PS5 Pro still uses GDDR6, but the bandwidth has increased by 28% from 448 GB/s of the PS5 to 576 GB/s, and the memory capacity allocated to the game has increased by more than 1GB.

With the dual improvement of GPU size 67% and memory bandwidth 28% increase, the overall gaming performance of the PS5 Pro can be improved by about 45% compared to the PS5. Here, for example, the original time required for PS5 to render 1 frame was about 16 ms, and the PS5 Pro can only be completed in 11 ms.

In terms of ray tracing, the RDNA 2.x GPU specially made for PS5 Pro can support BVH8 acceleration structure, and the number of intersections between each cycle multilateral type and the ray border is increased from 1 to 2.

Sony also divided the original PS5 coloring code into two stacks and added and removed them to the PS5 Pro to integrate them, effectively simplifying the coloring stroke to improve performance.

Therefore, compared with PS5, the PS5 Pro has slightly improved its shadow and plane reflection performance, while the surface reflection and diffuse performance has significantly improved its surface reflection and diffuse performance.

PSSR is the abbreviation of PlayStation Spectral Super Resolution. It uses a convolutional neural network to expand the selected and dyed pictures with lower resolution into high resolution, or generate picture grids (complement frames) to improve picture fluency.

Compared with PCs, it mostly uses fixed rendering low resolution and fixed Upscaling magnification, PS5 Pro adopts a combination mode of dynamic modulation and magnification to provide a fixed and smooth game screen.

Finally, Mark Cerny pointed out that Sony will work closely with AMD to develop hardware architectures that are more suitable for machine learning and artificial intelligence, as well as more convolutional neural networks for gaming. This plan is not exclusive to Sony, but allows more developers to obtain resources from it.

The above is the detailed content of Sony confirms the possibility of using special GPUs on PS5 Pro to develop AI with AMD. For more information, please follow other related articles on the PHP Chinese website!