AMD Launches Next-Gen RDNA GPU Architecture with Radiance Cores, Neural Arrays & Universal Compression for Enhanced Ray Tracing, Improved Upscaling, and Reduced Bandwidth Requirements

AMD Launches Next-Gen RDNA GPU Architecture with Radiance Cores, Neural Arrays & Universal Compression for Enhanced Ray Tracing, Improved Upscaling, and Reduced Bandwidth Requirements

AMD has unveiled three crucial features of its upcoming RDNA architecture: Neural Arrays, Radiance Cores, and Universal Compression, which promise to revolutionize gaming technology.

In a recent collaborative video, AMD and Sony presented a preview of innovative technologies and advancements that will be integrated into future graphics processing units (GPUs) and system-on-chips (SoCs) powered by AMD’s next-generation RDNA architecture. These developments are expected to significantly enhance gaming experiences, particularly in areas like upscaling and real-time ray/path tracing visual rendering.

AMD and Sony Interactive Entertainment logos with the text From Project Amethyst to the Future of Play and Performance | Immersion | Efficiency.

A few highlights from these exciting updates include:

  • Neural Arrays: A system of interconnected compute units that work collectively to perform advanced data processing, functioning cohesively as a singular AI engine.
  • Radiance Cores: Advanced hardware dedicated to ray traversal, designed to enhance performance in real-time ray tracing and path tracing applications.
  • Universal Compression: An innovative solution aimed at compressing data to significantly lower memory bandwidth requirements, thus improving overall GPU efficiency.
Next-Generation AMD Gaming Technologies slide detailing Neural Arrays, Radiance Cores, and Universal Compression.

Radiance Cores: Enhancing Real-Time Rendering

The newly introduced Radiance Cores will represent a significant leap in the RDNA GPU architecture. These cores are engineered with dedicated ray traversal technology, allowing for unprecedented speed and efficiency in real-time ray tracing and path tracing scenarios. Following the substantial enhancements made in ray tracing with the RDNA 4 architecture, the next generation is poised to set new performance standards with these specialized cores.

AMD spotlight on Radiance Cores with text on ray tracing and path tracing, titled Immersion, during AMD Gaming Technologies Preview.

Neural Arrays: A Leap in AI Processing

Next, we explore Neural Arrays, which consist of multiple compute units that are interconnected to facilitate collaborative data processing. Rather than functioning in isolation, these units will operate in unison as an advanced AI engine. This innovation is expected to significantly improve neural rendering, enhancing future iterations of FidelityFX Super Resolution (FSR) and PlayStation Rendering (PSR) upscaling techniques, resulting in superior image quality and scene rendering through frame generation algorithms.

AMD Gaming Technologies slide with Neural Arrays and Performance text beside a visual of compute units.

Universal Compression: Optimizing Memory Performance

Finally, AMD introduced Universal Compression, an innovative hardware block designed to enhance efficiency by evaluating and compressing GPU data. This compression algorithm will dramatically minimize memory bandwidth usage while maximizing performance. By optimizing memory handling, Universal Compression allows for quicker loading times for textures and models, which can greatly improve the overall gaming experience.

Universal Compression text with graphics illustrating data compression, from AMD Gaming Technologies Preview, Oct 2025.

While AMD has not specified a release timeline for these groundbreaking technologies, it is anticipated that the next-generation RDNA GPU architecture will first power discrete graphics solutions such as graphics cards and custom SoCs, which will be essential for the gaming platforms developed by Sony’s PlayStation.

Source&Images

Leave a Reply

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