Sony confirms new PS6 tech: Radiance Cores, Neural Arrays, and Universal Compression

Today, Sony and AMD outlined experimental new technologies in their multi-year Project Amethyst initiative that will be utilized in future PlayStation consoles, including Sony's next-gen PS6.

5

Sony and AMD have announced three new technologies for next-gen PlayStation hardware: Neural Arrays, Radiance Cores, and Universal Compression. All three features are built specifically for efficiency, optimization, and performance, with emphasis on offloading work from the CPU and GPU via dedicated blocks, smarter features, and drastically improving the rate of speed for data rendering. In short, Sony and AMD have created a trio of features that will supercharge the PS6 and maybe even Sony's rumored PlayStation handheld.

• Neural Arrays - Connects GPU Compute Units together for more performance through maximized efficiency
• Radiance Cores - Built onto chip, handles ray tracing features, frees up CPU and offers "significant speed boost."
• Universal Compression - Ambitious new tech that analyzes all data, and compresses all data where it can.

Perhaps the most immediately useful of the three is Universal Compression, which should have a net benefit to all games that are built on the new PlayStation systems, not just those that use ray tracing or machine learning features.

The PS5's compression tech, a derivative of RAD Tool's Kraken tech, allows it to deliver data assets at speeds of up to 5GB/sec and is part of a highly-customized I/O unit on the system's SoC.

Universal Compression should build off of this, as well as use new higher-speed storage, to deliver blazing fast load times as well as more streamlined gameplay.

Below we have a breakdown of each of the technologies, including quotes:

Neural Arrays

5

• Neural Arrays - Allows separate Compute Units within the GPU to work together instead of working alone. Connects CUs to shader engines. Works in real-time on GPU. Allows for larger Machine Learning (ML) models by increasing efficiency, reducing overhead. Enables faster, more capable performance for ML.

PS4 and PS5 architect Mark Cerny explains why Neural Arrays were developed:

"The neural networks found in technologies like PSSR and FSR are incredibly demanding on the GPU. They're both computationally expensive, and require speedy access to large amounts of memory.

"The nature of the GPU fights us here. It's made up of a large number of Compute Units, and problems are therefore broken up into bite-sized pieces to enable to the individual Compute Units to tackle them."

"Here's the idea: Instead of having a bunch of Compute Units all working on their own, we built a way for them to team up, to actually share data and process things together like a single, focused AI engine," said AMD SVP of Computing and Graphics Jack Huynh.

Mark Cerny explains that this will have a big impact on future collaborative tech moving forward, especially with the PS6's development:

"Neural Arrays will allow us to process a large chunk of the screen in one go."

"The efficiencies that come from that are going to be a game-changer as we develop the next-generation of upscaling and de-noising technology together [with AMD]."

Radiance Cores

5

• Radiance Cores - Adapted from Neural Radiance Cache from FSR Redstone. New dedicated hardware block for cinematic lighting. Handles ray and path tracing in real-time. Frees up CPU, will offer significant speed boost. New rendering approach for AMD.

"Radiance Cores takes full control of ray traversal, one of the most compute-heavy parts of the process. That frees up the CPU for geometry and simulation and lets the GPU focus on what it does best: Shading and lighting.

"The result is a cleaner, faster, and more efficient pipeline built for the next generation of ray traced games."

"There's a significant speed boost that comes from putting the traversal logic in hardware, and a further boost that comes from having that hardware operate independently from the shader cores," Cerny said.

"The challenge is that the current approach has reached its limit. To perform ray tracing today, a shader program has to juggle two very different responsibilities."

Universal Compression

5

• Universal Compression - Ambitious new tech that analyzes all data, and compresses all data where it can.

"Whether it's ML or ray tracing, they both hit the same bottleneck. Current GPU memory bandwidth limitations hinder the seamless adoption of next-gen rendering techniques, requiring significantly more bandwidth to handle 4K+ textures and raytracing denoising maps for smooth access streaming," said AMD's Jack Huynh.

"What we've built for future GPUs and SoCs takes the idea of data compression much further. We call it Universal Compression. It's a system that evaluates every piece of data headed to memory, not just textures, and compresses it whenever possible."

"Only the essential bytes are sent out, which dramatically reduces memory bandwidth usage. That means the GPU can deliver more detail, higher frame rates, and greater efficiency."