The reason for this is that we don't sue SSE optimization for 32bit platforms because of T36316. Things to look into: - Nail the root of the issue of that report - Implement non-SSE traversal code for QBVH