How Ray Tracing Cores Revolutionize Real-Time Graphics

Ray Tracing Core (RTC) is a dedicated hardware unit in modern GPUs (primarily NVIDIA’s RTX series and AMD’s RDNA 2/3/4) designed to accelerate real-time ray tracing—a rendering technique that simulates the physical behavior of light by tracing the path of light rays through a 3D scene. Unlike traditional rasterization (which renders images by projecting 3D geometry onto a 2D plane), ray tracing delivers photorealistic visuals by calculating reflections, refractions, shadows, and global illumination with physical accuracy. Ray Tracing Cores offload the computationally expensive ray-triangle intersection tests (the core of ray tracing) from general-purpose GPU cores (CUDA/Stream Processors), enabling real-time ray tracing in video games, architectural visualization, and film production.

1. Why Ray Tracing Cores Are Necessary

Ray tracing’s realism comes at a steep computational cost, primarily due to ray-triangle intersection tests:

A single frame in a 3D scene may require billions of rays to be cast (primary rays for camera view, secondary rays for reflections/shadows, tertiary rays for global illumination).
Each ray must check for intersections with millions of triangular polygons (the building blocks of 3D models) in the scene. A software-based (CPU/GPU core) intersection test is slow—even optimized algorithms (e.g., BVH traversal) take hundreds of clock cycles per ray.

Ray Tracing Cores are hardwired to execute intersection tests in hardware, reducing the time per test to just a few clock cycles. This hardware acceleration is the key to making real-time ray tracing (60+ FPS at 1080p/4K) feasible, whereas software-based ray tracing would only achieve a few FPS even on high-end GPUs.

2. Core Function of Ray Tracing Cores

Ray Tracing Cores specialize in two critical steps of the ray tracing pipeline:

2.1 Bounding Volume Hierarchy (BVH) Traversal

Before checking for ray-triangle intersections, the 3D scene is organized into a BVH—a tree-like data structure that partitions the scene into nested bounding volumes (e.g., axis-aligned bounding boxes, AABBs). This reduces the number of intersection tests by quickly discarding volumes the ray does not pass through.

Ray Tracing Cores accelerate BVH traversal by:

Executing parallel traversal of the BVH tree for multiple rays simultaneously.
Using dedicated logic to compare ray bounds with BVH nodes, skipping non-intersecting branches of the tree.

2.2 Ray-Triangle Intersection Testing

Once a ray reaches a leaf node of the BVH (containing triangular polygons), the Ray Tracing Core performs the ray-triangle intersection test—a mathematical calculation to determine if and where the ray intersects a triangle.

Key optimizations in hardware-based intersection testing:

Fixed-Function Logic: The core uses hardwired arithmetic units to compute the intersection formula (based on the Möller-Trumbore algorithm) in a single clock cycle, far faster than a software implementation on a CUDA core.
Batch Processing: Processes multiple rays and triangles in parallel (e.g., 8 rays per core per cycle), maximizing throughput.
Precision Optimization: Uses floating-point precision tailored to graphics (FP32 for coordinates, FP16 for intermediate calculations) to balance accuracy and speed.

End-to-End Ray Tracing Pipeline with RTCs

Ray Generation: The GPU’s shader cores generate primary rays (from the camera) and secondary rays (reflections, shadows) for the scene.
BVH Traversal: Ray Tracing Cores traverse the scene’s BVH to narrow down potential triangle intersections for each ray.
Intersection Testing: RTCs perform hardware-accelerated ray-triangle intersection tests to find the closest intersection point.
Shading: The GPU’s shader cores calculate the color/lighting at the intersection point (using materials, textures, and light sources).
Recursion: For secondary rays (e.g., reflections), the process repeats until the ray exits the scene or a recursion limit is reached.

3. Ray Tracing Core Architectures

Ray Tracing Cores are implemented differently by NVIDIA and AMD, the two leading GPU manufacturers:

3.1 NVIDIA RT Cores (RTX Series)

NVIDIA introduced the first Ray Tracing Cores with the Turing architecture (2018) (RTX 20 series), with successive generations bringing significant improvements:

Turing (RTX 2080): First-gen RT Cores support ray-triangle intersection and BVH traversal, delivering up to 10 GigaRays/second (GR/s) of ray tracing throughput.
Ampere (RTX 3080): Second-gen RT Cores add support for ray-AABB intersection (faster BVH traversal) and double the throughput to 20 GR/s.
Ada Lovelace (RTX 4090): Third-gen RT Cores introduce Opacity Micromaps and Displacement Micro-Meshes (optimizations for complex geometry) and boost throughput to 191 GR/s (FP32) / 382 GR/s (FP16).
Blackwell (RTX 5090): Fourth-gen RT Cores integrate AI acceleration (via Tensor Cores) for denoising and further increase throughput to over 500 GR/s.

NVIDIA RT Cores are integrated into the GPU’s Streaming Multiprocessors (SMs):

Turing/Ampere: 1 RT Core per SM.
Ada Lovelace: 2 RT Cores per SM (for higher parallelism).

3.2 AMD Ray Accelerators (RDNA 2/3/4)

AMD introduced ray tracing hardware with the RDNA 2 architecture (2020) (Radeon RX 6000 series), branded as Ray Accelerators:

RDNA 2 (RX 6900 XT): Each Compute Unit (CU) has a single Ray Accelerator that handles BVH traversal and ray-triangle intersection, delivering ~10 GR/s of throughput.
RDNA 3 (RX 7900 XTX): Second-gen Ray Accelerators improve BVH traversal efficiency and double the throughput to ~20 GR/s; they also support ray traversal for volumetric effects (e.g., smoke, fog).
RDNA 4 (RX 8000 series): Third-gen Ray Accelerators add support for AI-driven ray tracing optimizations and boost throughput to ~40 GR/s.

Unlike NVIDIA’s SM-integrated RT Cores, AMD’s Ray Accelerators are distributed across all CUs, enabling better load balancing for ray tracing workloads.

4. Key Performance Metrics

Ray Tracing Core performance is measured by ray tracing throughput (GigaRays/second, GR/s)—the number of billions of rays the cores can process per second:

GPU Model	Ray Tracing Core Generation	Throughput (GR/s)	Key Feature
NVIDIA RTX 2080 Ti	Turing (1st-gen)	11	Basic ray-triangle intersection
NVIDIA RTX 3090	Ampere (2nd-gen)	23	Faster BVH traversal
NVIDIA RTX 4090	Ada Lovelace (3rd-gen)	191	Opacity Micromaps, Micro-Meshes
AMD Radeon RX 6900 XT	RDNA 2 (1st-gen Ray Accelerator)	10	CU-distributed ray tracing
AMD Radeon RX 7900 XTX	RDNA 3 (2nd-gen)	20	Volumetric ray tracing

Throughput alone does not determine visual quality—factors like BVH construction efficiency, denoising (AI/traditional), and shader performance also play critical roles in real-time ray tracing.

5. Complementary Technologies for Ray Tracing

Ray Tracing Cores work with other GPU features to deliver real-time photorealism:

5.1 AI Denoising

Ray tracing often uses stochastic sampling (casting a limited number of rays per pixel) to reduce computational load, which results in noisy images. AI denoisers (e.g., NVIDIA DLSS 3, AMD FSR 3) use Tensor Cores/RDNA AI accelerators to remove noise and upscale low-resolution ray-traced images to 4K/8K—doubling or tripling frame rates while preserving quality.

5.2 Hybrid Rendering (Rasterization + Ray Tracing)

Most modern games use hybrid rendering: rasterization for primary scene geometry and Ray Tracing Cores for high-impact effects (e.g., reflections, shadows, global illumination). This balances performance and realism, as full ray tracing (path tracing) is still too computationally expensive for real-time 4K gaming.

5.3 BVH Optimization

GPUs use dynamic BVH construction to update the scene’s BVH in real time (e.g., for moving objects), and Ray Tracing Cores are optimized to traverse these dynamic BVHs efficiently. Advanced techniques like incremental BVH updates reduce the overhead of modifying the BVH for dynamic scenes.

6. Applications of Ray Tracing Cores

Ray Tracing Cores enable photorealistic rendering across industries:

Video Games: Real-time ray tracing for reflections (e.g., Cyberpunk 2077), shadows (e.g., Minecraft RTX), global illumination (e.g., Fortnite RTX), and ray-traced lighting (e.g., Resident Evil 4 Remake).
Architectural Visualization: Real-time ray tracing for photorealistic previews of buildings and interiors (e.g., Autodesk 3ds Max, SketchUp with V-Ray GPU).
Film & Animation: Accelerated offline ray tracing for movie CGI (e.g., Pixar’s RenderMan, NVIDIA Iray), reducing render times from hours to minutes per frame.
Automotive Design: Real-time ray tracing for virtual prototyping of car exteriors/interiors, enabling designers to evaluate lighting and materials interactively.
Virtual Reality (VR): Low-latency ray tracing for immersive VR experiences, with photorealistic lighting that enhances presence in virtual environments.

7. Limitations of Ray Tracing Cores

Hardware Exclusivity: Ray Tracing Cores are only available in modern high-end GPUs (NVIDIA RTX 20+/AMD RX 6000+), making real-time ray tracing inaccessible on older or budget GPUs.
Computational Overhead: Even with hardware acceleration, full path tracing (simulating all light rays) is still not feasible for real-time 4K/60 FPS gaming—requiring hybrid rendering or AI upscaling.
Scene Complexity Limits: Ray Tracing Cores struggle with extremely dense scenes (e.g., millions of polygons) or volumetric effects (e.g., dense fog), as BVH traversal and intersection tests become more expensive.
Power Consumption: Ray tracing is power-hungry—high-end GPUs (e.g., RTX 4090) draw 450W or more when running ray-traced workloads, limiting adoption in laptops and small form-factor PCs.