AMD Patents a Faster Way to Wrap 3D Objects for Ray Tracing
Every ray-traced image you see depends on millions of invisible boxes drawn around 3D objects. AMD has filed a patent for a smarter way to align those boxes, which could mean fewer wasted calculations and faster renders.
How AMD's bounding box trick speeds up ray tracing
Imagine you're packing oddly shaped sculptures into cardboard boxes for shipping. If you tilt a sculpture slightly, it fits into a much smaller box. But if you always keep every box axis-aligned (perfectly upright, flat against the wall), you waste a lot of space with air gaps around diagonal objects. Graphics cards face exactly this problem when rendering realistic light.
To calculate how light bounces around a 3D scene, your GPU wraps every object in an invisible bounding box and organizes those boxes into a tree-like structure. When a box is tilted to match the actual shape of the object inside, the GPU can skip more irrelevant objects and trace light rays faster. The tricky part is deciding which orientation to use, and AMD's patent has a specific answer: find the biggest box in a group, and let all its neighbors inherit its tilt.
The result is a more consistent family of boxes that share a common orientation, which makes the tree structure easier and faster for the GPU to search through when rendering each frame.
How the BVH builder picks and shares box orientations
Ray tracing hardware uses a data structure called a Bounding Volume Hierarchy (BVH), a tree where every branch and leaf is a box wrapping part of the scene. The GPU fires rays into the scene and quickly eliminates entire subtrees of boxes that the ray misses, rather than testing every triangle individually.
Traditional BVHs use axis-aligned bounding boxes (AABBs), meaning every box is perfectly square with the world's X, Y, and Z axes. These are fast to compute but wasteful for diagonal objects. Oriented bounding boxes (OBBs) can be tilted to hug an object more tightly, reducing empty space and cutting down on false-positive ray tests.
AMD's patented method decides how to orient each OBB by looking at sibling nodes sharing the same parent in the tree:
- Collect all child nodes under a common parent.
- Find the child node with the greatest total surface area (the biggest, most influential box in the group).
- Apply that node's orientation to the other siblings and, optionally, to the parent node itself.
- Render the image using the resulting, orientation-unified BVH.
By propagating the dominant child's orientation upward and sideways, the method keeps nearby boxes aligned with each other. That consistency is important because it reduces the overhead of switching between different coordinate frames as the GPU walks down the tree.
What this means for AMD's GPU ray tracing pipeline
Ray tracing is now standard on AMD's RDNA GPUs and in every modern game engine, but BVH quality is still one of the main bottlenecks. A poorly constructed BVH means more ray-box intersection tests per frame, which directly translates to lower frame rates or lower image quality within a fixed time budget. Better OBB construction is a low-level fix with broad benefits: every ray-traced scene benefits, from games to professional visualization software.
For AMD specifically, this filing signals continued investment in the software-hardware co-design of their ray tracing stack. Nvidia has held a performance edge in ray tracing for several generations, partly due to BVH build quality. Patents like this suggest AMD is working on the algorithmic layer, not just adding more transistors.
This is genuinely useful infrastructure work. BVH construction quality has a direct, measurable impact on ray tracing performance, and AMD publishing a specific algorithm for orienting boxes based on surface area dominance is the kind of detail that shows up in real driver or compiler improvements. It won't make headlines at a product launch, but it's exactly the sort of foundational work that narrows the gap with Nvidia's ray tracing advantage.
Get one Big Tech patent every Sunday
Plain English, intelligent commentary, no hype. Free.
Editorial commentary on a publicly published patent application. Not legal advice.