Justification of 8‑wide BVH construction complexity versus BVH4 memory-traffic benefit

Determine whether the additional algorithmic complexity in constructing 8‑wide bounding volume hierarchies (BVH8) is compensated by their reduction in memory traffic compared to 4‑wide BVHs (BVH4) within the quantized BVH and triangle representation combined with ray stream tracing described in the paper, especially for large scenes where BVH4 appeared to reduce total memory traffic more strongly.

Background

The paper compares multiple configurations of BVHs (2‑wide, 4‑wide, and 8‑wide) under both single-ray and ray-stream traversal, with and without compression. Across large scenes, the BVH4 with ray stream tracing and compression often yields the lowest total memory traffic, with BVH8 close but not consistently superior.

In the Conclusion’s BVH8 vs. BVH4 discussion, the authors note that BVH8 introduces extra algorithmic complexity (e.g., wider SIMD opportunities and amortizing quantization data across more child boxes). However, their measurements do not establish a clear advantage that justifies this additional complexity over BVH4 from a memory-traffic perspective, leaving open whether BVH8’s complexity is warranted under these techniques and workloads.