AMD and Sony’s PS6 Chipset to Revolutionize Graphics Pipeline
▼ Summary
– Sony and AMD are collaborating on “Project Amethyst,” a co-engineering effort for a future console, which is still in early development but showing promising results.
– The project aims to move beyond traditional rasterization by focusing on more efficient machine-learning-based neural networks for graphics upscaling.
– Current GPU limitations require inefficient parallel processing of subproblems, which Project Amethyst seeks to address with new architectural improvements.
– Project Amethyst introduces “neural arrays” that allow compute units to share data and function as a unified AI engine for better scalability.
– This technology will enable more on-screen graphics to be enhanced by machine learning, improving real-time visual performance.
The next generation of gaming hardware is already taking shape, with AMD and Sony collaborating on Project Amethyst, a co-engineering initiative aimed at transforming graphics processing for future PlayStation consoles. In a recent video discussion, Sony’s Mark Cerny and AMD’s Jack Huynh revealed that this early-stage project focuses on moving beyond conventional rendering methods to deliver unprecedented visual performance through advanced machine learning integration.
Project Amethyst represents a shift away from traditional rasterization, which struggles to scale efficiently when relying solely on raw processing power. Instead, the architecture is being designed to optimize machine learning-based neural networks, similar to those used in AMD’s FSR and Sony’s PSSR upscaling systems. This approach promises more intelligent and efficient graphics handling, moving past the limitations of current GPU designs.
Huynh explained that existing upscaling techniques help achieve real-time 4K graphics but face inherent inefficiencies due to the GPU’s structure. Calculations must be divided into smaller subproblems and processed in parallel across separate compute units, which doesn’t always make the best use of available resources. Cerny noted that the fundamental nature of GPU architecture creates obstacles for seamless, high-resolution rendering.
To overcome these challenges, Project Amethyst introduces “neural arrays” that enable compute units to share data and function as a unified AI engine. While not every part of the GPU will be interconnected in this way, linking smaller groups of compute units allows shader engines to process larger portions of the screen simultaneously. This scalable design means machine learning will play a greater role in shaping what gamers see, with more on-screen elements being touched or enhanced by ML algorithms. Both Cerny and Huynh described initial simulation results as highly promising, signaling a strong foundation for the hardware that may one day power upcoming consoles.
(Source: Ars Technica)