Microsoft has filed a patent for a system and method that aims to improve ray-tracing performance in scenarios where video RAM (VRAM) is constrained. The patent, titled "Systems and Methods for Ray Tracing Acceleration Structure Level of Detail Processing," describes a technique for managing the memory footprint of ray-tracing acceleration structures.
As ray-tracing acceleration structures compete for memory space with other data, such as geometry and texture data, Microsoft's proposed solution involves managing these structures with varying levels of detail for different geometry segments resident in the random access memory (RAM).
The key components of the system include a CPU-managed geometry node pool and corresponding level of detail (LOD) residency map, as well as a GPU-maintained LOD recording map. The residency map, maintained by the CPU, contains entries pointing to different LOD representations of sub-trees in the geometry node pool.
During rendering, the GPU determines the projected size of bounding volumes hit by primary rays and computes a desired LOD value for each sub-tree. This computed LOD is compared against the resident LOD from the residency map, and the maximum of the two is selected as the final LOD for traversal.
The GPU updates the recording map with the minimum LOD required for each sub-tree, which is then transferred back to the CPU after rendering completes. The CPU uses this historical usage data to evict unnecessary LODs from the geometry node pool and load in required LODs for the next frame, thereby minimizing the memory footprint.
By selectively managing the resident LODs and only fetching the required geometry data, Microsoft aims to reduce the storage and bandwidth requirements for ray-tracing acceleration structures, particularly in VRAM-constrained scenarios.
Are you interested in getting a new Windows 11 PC or upgrading your existing one?
We Deliver Nationwide
