This article details "Project Amethyst," a collaborative engineering effort between Sony and AMD to design the chipset for a future PlayStation console, presumably the PS6. Hardware architect Mark Cerny and AMD's Jack Huynh outline a new architecture, currently in the simulation phase, that aims to move beyond traditional graphics pipelines by focusing on efficiency.
The project introduces three core innovations to address current bottlenecks. First, "neural arrays" will allow GPU compute units to function collectively as a focused AI engine, dramatically improving the efficiency of machine learning tasks like upscaling. Second, dedicated "radiance cores" will offload the computationally intensive process of ray traversal, freeing up the main GPU for other shading tasks and making ray-tracing more efficient. Finally, a new "universal compression" system will compress all data sent to the GPU, not just textures, to overcome memory bandwidth limitations and achieve an effective bandwidth that exceeds its paper specifications.
Inside Project Amethyst: Sony and AMD's Blueprint for the PS6 Chipset
A Glimpse into the Future: Sony's Mark Cerny and AMD's Jack Huynh have revealed "Project Amethyst," the codename for the co-engineered chipset intended to power a future PlayStation console. The project is still in its early, simulation-only stages but aims to rethink the graphics pipeline for next-generation performance.
Rethinking Machine Learning: The new architecture is designed to run machine-learning-based neural networks (like those in FSR and PSSR upscaling) more efficiently. Instead of breaking tasks into inefficient parallel processes, new "neural arrays" will allow clusters of compute units to share data and work as a "single focused AI engine," enabling ML to enhance a much larger portion of the on-screen image.
Dedicated Ray-Tracing Hardware: To address the inefficiency of having shaders handle both ray-tracing and traditional shading, Project Amethyst will introduce a separate hardware block of "radiance cores." These cores will be dedicated solely to the computationally intense task of ray traversal (calculating ray paths), freeing up the main GPU for texture and material calculations.
Overcoming Bandwidth Limits: The project targets the significant bottleneck of memory bandwidth. Building on existing Delta Color Compression, the new chip will feature a "universal compression" system. This technology will compress all data sent to the GPU—not just specific textures—to create an "effective bandwidth" that is much higher than the hardware's official specifications.
Early Stages, Big Ambitions: While no hardware or demos exist yet, the project provides insight into the future of real-time graphics. The focus is on intelligent efficiency and specialized hardware to squeeze more performance out of silicon, rather than relying on brute-force power, in an era of slowing technological improvement.