News › performance
By Zayden R., August 31, 2026
Mesa 26.3 introduces a faster merge sort for Vulkan drivers, improving ray-tracing performance. Engineers should expect better efficiency with reduced vRAM roundtrips, particularly beneficial for small BVHs.
The Mesa 26.3 release brings a noteworthy improvement to Vulkan drivers, courtesy of Valve's Natalie Vock. A new custom merge sort for Morton code keys has been integrated, promising enhanced performance for ray-tracing tasks. This change is particularly significant for engineers focused on graphics optimization.
Morton code keys, crucial for fast BVH construction, are now sorted more efficiently. The custom merge-sort developed by Vock is tailored to outperform the previous sorting function, especially for smaller datasets. This optimization reduces vRAM roundtrips, enhancing GPU parallelism and efficiency.
Previously, the Radeon RADV, Intel ANV, Turnip Adreno, and Lavapipe Mesa Vulkan drivers relied on a less efficient sorting method. Vock's merge-sort, however, operates within a single dispatch without intermediate barriers, tightly integrating with the rest of the AS build code. This integration significantly cuts down on the need for vRAM roundtrips, which can be a bottleneck in graphics processing.
After three months of rigorous testing and debugging, this new sorting method has been merged into Mesa 26.3. Engineers can now expect improved performance, particularly in handling small BVHs. While the merge-sort may not be as effective for larger problem sizes, its benefits for smaller tasks are clear, making it a valuable update for graphics engineers.
For those interested in the technical specifics, the merge request details Vock's work and the testing process. This optimization marks a meaningful step forward in open-source Vulkan driver development, reflecting the ongoing commitment to improving Linux graphics performance.
The Linux Camp teaches these topics as hands-on labs on real virtual machines, verified as you type.