← All resources

BLAS Sharing for ray tracing cluster-based continuous level of detail

Auto-discovered

Documentation Advanced

Visit resource ↗

Description

BLAS Sharing for ray tracing cluster-based continuous level of detail is a technical documentation page in NVIDIA nvpro-samples vk_lod_clusters describing how to share bottom-level acceleration structures across LoD instances.

It explains instance LoD range classification with conservative min and max estimates, geometry histograms that pick a sharing instance, four instance categories for reuse versus per-instance builds, and how sharing relates to separate BLAS caching and merging docs. Example LEGO-brick scene stats show large cuts in BLAS builds, memory, and frame time.

It is for advanced Vulkan ray-tracing engineers working on cluster-based continuous LoD. Familiarity with BLAS and TLAS and the sample LoD builder README is expected.

The page is free Markdown documentation in the sample repository with diagrams, GLSL kernel pointers, and result tables.

It is useful when many instances of the same geometry force too many per-frame BLAS builds under continuous LoD.