ATI Technologies Patents a Way to Cut Wasted Ray Tracing Tests in 3D Scenes

By Patentlyze Team · Updated Jun 26, 2026

Ray tracing makes 3D graphics look stunning, but it can waste enormous processing power testing light rays against boxes that don't really need testing. ATI Technologies has filed a patent for a method that deliberately skips those tests for certain geometry, making the whole process leaner.

What ATI's 'always hit' ray tracing fix actually does

Imagine a librarian organizing millions of books into nested boxes, where each box tells a delivery person roughly where the books inside are. Ray tracing works similarly: your GPU fires millions of virtual light rays and checks which boxes they pass through before deciding what they actually hit. The problem is that when the box labels are imprecise, the GPU ends up checking far too many boxes, most of them dead ends.

ATI's patent describes a fix where some boxes get tagged as "always hit", meaning the GPU just skips the check entirely and dives straight inside. This sounds counterintuitive, but it actually helps: by removing those outlier boxes from the precision calculation, the remaining boxes get more accurate labels, which means fewer false alarms overall.

The net result is that your GPU spends less time on pointless intersection tests and more time on the pixels that actually matter, which should translate to faster frame rendering.

How the BVH node flagging and fixed-point bounds interact

Ray tracing relies on a data structure called a bounding volume hierarchy (BVH), essentially a tree of nested boxes that tells the GPU which parts of a 3D scene a light ray might touch. To save memory, the box coordinates are stored in fixed-point format (a compressed number representation with a fixed number of decimal places, rather than the floating-point numbers your CPU typically uses for math).

The problem: the precision of that fixed-point format is governed by the outermost bounds of the whole node. If a few very large boxes dominate those bounds, all the smaller boxes inside end up with less precision, meaning they appear bigger than they really are in the compressed representation. That causes a flood of false positive intersections (the GPU thinks a ray hit something when it didn't), wasting processing time.

ATI's method introduces an "always hit" flag on individual node references inside the BVH. When the GPU traverses the tree and encounters a node marked always-hit, it skips the intersection test entirely and just descends into that node's children. Because those flagged nodes are excluded from the fixed-point bounds calculation, the remaining nodes benefit from tighter, more accurate bounds.

The trade-off is deliberate: you accept a small number of guaranteed traversals into certain nodes in exchange for much better precision, and therefore far fewer false positives, across the rest of the scene.

What this means for GPU ray tracing performance

For AMD's Radeon GPU line (ATI Technologies is AMD's GPU subsidiary), this is the kind of low-level optimization that adds up quickly in complex scenes. Ray tracing is already one of the most compute-heavy workloads a GPU faces, and false positive intersection tests compound with scene complexity, the bigger and busier the game world, the worse the problem gets.

If this technique ships in future Radeon hardware or driver-level BVH builders, it could improve frame rates in ray-traced titles without requiring any changes from game developers. It also signals that AMD is continuing to invest in the compression and traversal efficiency of its ray tracing pipeline, an area where it has historically competed hard against Nvidia's RT Core architecture.

Editorial commentary on a publicly published patent application. Not legal advice.