RT Cores

RT Cores
Name	RT Cores
Designer	Nvidia
Launched	2018
Part of	GeForce RTX 20 series and later
Type	Dedicated ray tracing acceleration hardware

RT Cores. They are specialized processing units integrated within Nvidia's GPUs, starting with the Turing architecture, designed to accelerate the complex calculations required for real-time ray tracing. This technology represents a significant shift in computer graphics, enabling more realistic lighting, shadows, and reflections in interactive applications like video games and professional visualization software. By offloading these intensive tasks from the traditional CUDA cores, RT Cores allow for a practical balance between visual fidelity and frame rate performance.

Overview and Purpose

The primary purpose of RT Cores is to make the computationally expensive technique of ray tracing feasible for real-time rendering. Before their introduction, ray tracing was predominantly confined to offline rendering for films and animation, used by studios like Pixar and Industrial Light & Magic. In a graphics pipeline, these dedicated cores handle specific algorithms, such as bounding volume hierarchy traversal and ray-triangle intersection tests, which are fundamental to simulating the path of light. This specialization allows game engines like Unreal Engine and Unity to incorporate cinematic-quality lighting effects into interactive experiences, bridging a gap previously separating the video game industry from Hollywood.

Hardware Architecture

Architecturally, RT Cores are fixed-function units embedded within the streaming multiprocessors of Nvidia GPUs. In the initial Turing design, each core contained components for accelerating bounding volume hierarchy operations and performing ray-triangle intersection calculations at high speed. This design was refined in subsequent architectures like Ampere and Ada Lovelace, increasing throughput and efficiency. The cores work in concert with other GPU components, including the Tensor Cores used for deep learning tasks like NVIDIA DLSS, and the standard CUDA cores responsible for general-purpose shader operations. This integration is a hallmark of Nvidia's approach to heterogeneous computing.

Ray Tracing Acceleration

RT Cores accelerate ray tracing by executing two key operations with extreme efficiency: traversing an acceleration data structure, typically a bounding volume hierarchy, and testing for intersections between rays and geometric primitives. When an application, such as a title using Microsoft's DirectX Raytracing API, sends a ray tracing workload, these cores rapidly determine which objects in a 3D scene a ray will hit. This process is critical for accurately calculating effects like accurate reflections visible in Cyberpunk 2077, soft shadows in Control (video game), and global illumination in Minecraft with RTX. The acceleration is so substantial that it enables these effects at interactive frame rates, a feat not possible with GPGPU approaches alone.

Performance and Applications

The performance impact of RT Cores is most evident in benchmarks and real-world applications that enable ray tracing features. In games like Battlefield V and Metro Exodus, enabling ray tracing can significantly increase visual realism, though it often necessitates the use of upscaling technologies like DLSS to maintain smooth performance. Beyond entertainment software, these cores are leveraged in professional sectors for computer-aided design with Autodesk software, scientific visualization, and architectural rendering in tools like Chaos's V-Ray. The NVIDIA Omniverse platform also utilizes this hardware for real-time simulation and collaborative design, demonstrating its utility across the technology industry.

Development and History

The development of RT Cores was driven by Nvidia's long-term research into real-time ray tracing, which included academic collaborations and the development of APIs like OptiX. Their commercial debut was in 2018 with the launch of the GeForce RTX 2080, based on the Turing architecture, which Nvidia marketed as a revolutionary step for computer graphics. This launch was followed by integration into the professional Quadro and data center Tesla product lines. Subsequent generations, including the Ampere-based GeForce 30 series and the Ada Lovelace-based GeForce 40 series, have continually evolved the technology, increasing ray tracing performance and enabling more advanced effects.

Comparison with Other Technologies

RT Cores are often compared to the ray tracing acceleration solutions from competitors like AMD and Intel. AMD's approach, first introduced with RDNA 2 architecture in products like the Radeon RX 6000 series, uses a more unified hardware design where ray intersection tasks are handled by enhanced compute units. Intel's Xe HPG architecture, found in the Arc Alchemist GPUs, includes similar dedicated hardware called Ray Tracing Units. While the underlying goal of accelerating the DXR and Vulkan ray tracing APIs is the same, the architectural implementations differ, leading to variations in performance across titles like Fortnite and Shadow of the Tomb Raider. The ecosystem support from Nvidia, including partnerships with Electronic Arts and Ubisoft, has also been a distinguishing factor in adoption.

Category:Nvidia Category:Graphics processing units Category:Computer hardware Category:3D computer graphics