AMD Unveils Radiance Cores, Neural Arrays & Universal Compression For Next-Gen RDNA GPU Architecture: Faster RT, Better Upscaling, & Lower Bandwidth Needs

AMD has just announced three key features of its next-gen RDNA architecture: Neural Arrays, Radiance Cores & Universal Compression.

AMD & Sony Showcase Next-Gen Architecture Changes That Will Be Featured On Future RDNA GPUs & SoCs

Today, AMD and Sony shared a new video in which they offer a look into future technologies & graphics architecture updates that will be coming to next-gen GPUs and SoCs, based on AMD's next-gen RDNA graphics. The three technologies will play a major role in redefining the gaming landscape as the industry moves heavily into upscaling and ray/path tracing scenes.

Some of the highlights of these three updates include:

• Neural Arrays: A collection of compute units configured to share and process data to work together like a single AI engine.
• Radiance Cores: New dedicated ray traversal hardware to deliver high-performance real-time ray tracing and path tracing.
• Universal Compression: A new system that evaluates and compresses all available data within the GPU to dramatically reduce memory bandwidth usage.

First up, we have the new Radiance Cores, which will be a brand new core integrated within the next-gen RDNA GPU architecture. These cores offer dedicated ray traversal hardware, which is designed to deliver faster and higher-performance real-time raytracing and path tracing gaming. AMD already made some big updates on the raytracing side with its RDNA 4 architecture, and it looks like the next-gen is going to offer an even bigger uplift with these cores.

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. Now, we're not linking the entire GPU into one mega unit.

That'll be a cable management nightmare. But we are connecting CUs with each shader engine in a smart, efficient way, and that changes the game for neural rendering. Bigger ML models, less overhead, more efficiency, and way more scalability as work goes well.

With neural arrays, we're unlocking a whole new level of performance for ML. Not just faster, but more capable.

That means better FSR, better ray regeneration, and brand new ML power features we're just starting to imagine, all working in real time right on the GPU. And we're just getting started. As we look ahead, you also see dedicated innovations that bring cinematic rendering to an entirely new level.

Jack Huynh - AMD SVP and GM of Computing and Graphics Group

Neural arrays will allow us to process a large chunk of the screen in one go, and the efficiencies that come from that are going to be a game-changer as we begin to develop the next generation of upscaling and denoising technologies.

Mark Cerny - Lead Architect of PS5 and PS5 Pro

Next up, we have Neural Arrays, which are a collection of compute units, interconnected to share and process data to work together like a singular AI engine rather than each compute unit working alone. With Neural Arrays, AMD and Sony expect a major uplift in neural rendering, which will help future generations of FSR and PSR upscaling techniques, allowing better upscaled image quality and better scene rendering when using frame-gen algorithms.

And now with path tracing becoming more central to real-time graphics, the demands on GPU just continue to grow. That's why we've been pushing hard to go beyond the current approach to help developers bring even more realism and cinematic lighting into their game. But 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. One is ray traversal, digging through complex data structures to locate where the millions of rays being passed hit the millions of triangles in the scene geometry. When there are intersections, that same shader program has to also be doing its usual work of shading the scene using texture and lighting information.

Mark Cerny - Lead Architect of PS5 and PS5 Pro

And we spent the past two years rethinking the entire path tracing pipeline from hardware to software. Early this year at Computex, we introduced Neural Radiance Caching, a key part of FSR Redstone. Now, we're building on that with Radiance Cores, a new dedicated hardware block designed for unified light transport.It handles ray tracing and path tracing in real time, pushing light performance to a whole new level. Together, these form a brand new rendering approach for AMD.

Radiance Cores takes full control of ray transversal, one of the most compute-heavy parts of the process.And that feeds up the CPU for geometry and simulation, and lets the GPU focus on what does best, shading and lighting.

The result? A cleaner, faster, and more efficient pipeline built for the next generation of ray tracing. 

Jack Huynh - AMD SVP and GM of Computing and Graphics Group

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. On top of those performance increases, there's other features in the works too, such as flexible and efficient data structures for the geometry and ray trace. Overall, I'm really looking forward to the time when we can get Radiance Cores into the hands of game creators.

Mark Cerny - Lead Architect of PS5 and PS5 Pro

AMD is also working on its next-gen FSR Redstone update which is adding a lot of machine learning capabilities to the FSR suite. It is possible that Redstone will have some additions to make use of these future Neural Arrays for better upscaled image quality.

Lastly, AMD unveiled its Universal Compression solution for future RDNA GPUs. This compression algorithm is an efficiency-boosting hardware block that evaluates and compresses all available data within the GPU to dramatically reduce memory bandwidth use and deliver higher performance.

Memory compression algorithms, such as Universal Compression, not only reduce the need for higher system bandwidth but also help in loading textures and models faster.

AMD hasn't mentioned when we will get to see these technologies on the consumer front, but it is likely that AMD's next-gen RDNA GPU architecture, which will power discrete graphics solutions such as graphics cards and custom SoCs such as the ones used by Sony's PlayStation consoles, will be the first to utilize them in the coming year.

Follow Wccftech on Google to get more of our news coverage in your feeds.