Sony & AMD Tease ‘Project Amethyst’ — Neural Arrays, Radiance Cores, Universal Compression

Sony & AMD Tease “Project Amethyst”

Sony and AMD released a discussion video featuring Mark Cerny (lead architect of PS4/PS5) and Jack Huynh (SVP & GM, AMD Computing and Graphics Group) outlining early research collectively called Project Amethyst. The demo technologies—Neural Arrays, Radiance Cores, and Universal Compression—are intended to improve ray tracing, ML-based upscaling, and GPU efficiency for Sony’s next PlayStation and future AMD GPUs.

Key Technologies

Neural Arrays

Neural Arrays enable compute units to work together and share data across a larger tile of the screen, improving efficiency for machine-learning workloads such as upscaling (e.g., FSR/PSSR) and ray regeneration. AMD says this will deliver a “whole new level of machine learning performance.”

Radiance Cores

Radiance Cores are a new fixed-function hardware block designed to accelerate ray and path tracing math (similar in purpose to NVIDIA’s RT cores). Offloading ray/path tracing to Radiance Cores frees shader/texture units for other work, boosting overall efficiency.

Universal Compression

Building on Sony’s Delta Color Compression, Universal Compression aims to compress everything that flows through the GPU pipeline—reducing memory bandwidth needs and potentially lowering power consumption.

Implications

These technologies are still “early days,” but if they mature, they could make high-quality ray tracing and ML upscaling more affordable performance-wise—improving visuals and framerates on the next PlayStation and AMD cards. That could narrow the ray-tracing gap with NVIDIA and enable more games to use ray-traced global illumination and other advanced rendering techniques.

Sources & Video

Discussion

What do you think—will Project Amethyst be enough to push AMD and PlayStation ray tracing forward? Share your thoughts below.

Leave a Reply

Your email address will not be published. Required fields are marked *

Diese Seite verwendet Cookies, um die Nutzerfreundlichkeit zu verbessern. Mit der weiteren Verwendung stimmst du dem zu.

Datenschutzerklärung