Intel Patents a Cut in the Data Searched to Produce Lifelike Shadows and Reflections

By Patentlyze Team · Updated Jul 10, 2026

Ray tracing makes game lighting look incredible, but it's brutally expensive for GPUs. Intel's new patent targets one of the biggest hidden costs: the enormous lookup tables GPUs must search just to figure out where light rays hit objects.

What Intel's ray tracing shortcut actually does

Imagine a game renderer as a very fast postal system. Every ray of light it shoots into a scene has to check a giant address book to find which surfaces it might hit. The bigger that address book, the slower everything gets.

Intel's patent describes a smarter way to organize that address book. Instead of listing every triangle in a 3D scene individually, it groups them into small bundles called meshlets and transforms them into a shared coordinate system tied to the ray itself. That way, a lot of triangles can be ruled out quickly without even being tested.

The result is a smaller, leaner lookup structure that the GPU can search faster. For you as a player or creator, that could translate to more rays per frame, better lighting quality, or the same quality at lower power draw.

How Intel reorganizes triangles to skip bad intersection tests

Ray tracing works by casting millions of virtual light rays and checking what they hit. The core tool for doing this efficiently is a bounding volume hierarchy (BVH), a tree-shaped data structure that wraps groups of triangles in progressively smaller boxes. The GPU walks the tree to quickly narrow down which triangles a ray might actually touch.

The problem is that BVH structures can grow very large, and large structures mean more memory bandwidth and more traversal time. Intel's patent addresses this by storing triangles in meshlets (small, fixed-size clusters of triangles, typically used in modern geometry pipelines) directly inside the BVH's leaf nodes rather than referencing them indirectly.

The key step is a coordinate transform: the system converts triangle vertices into unit ray coordinate space, meaning a local coordinate system where the ray itself defines the axes. In this space, intersection math becomes simpler and a culling step (a fast filter that discards triangles the ray definitely cannot hit) can eliminate a large fraction of candidates before the expensive per-triangle test runs.

The remaining triangles, post-culling, are then intersected with the ray in that same compact space. Fewer tests, less data moved, similar or better results.

What this means for GPU ray tracing performance

For GPU architects, shaving intersection test overhead is one of the highest-leverage places to optimize. Every frame in a ray-traced scene involves billions of these checks, and reducing the size of the acceleration structure directly reduces memory traffic, which is often the real bottleneck on modern graphics hardware.

For the industry broadly, Intel is competing with Nvidia and AMD in dedicated ray tracing silicon. A more efficient BVH traversal method could translate into real frame-rate gains or lower-power ray tracing on integrated graphics, which is where Intel has the most to prove. Whether this ends up in a discrete GPU or a future Core Ultra integrated chip is an open question, but the direction is clear.

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