Zink OpenGL-On-Vulkan Hits Another "Massively Improved Performance" Milestone
The Zink component to Mesa that provides a generic OpenGL implementation built atop the Vulkan API recently hit another "massively improved performance" milestone by Valve contractor Mike Blumenkrantz.
Mike began work on a suballocator for Zink that is based on the Gallium3D auxiliary/pipebuffer code originally started by the RadeonSI Gallium3D driver. After making significant changes to that code, Zink's new suballocator implementation is showing off significant performance improvements in just shy of 700 lines of new code.
Blumenkrantz summed up the impact in a new blog post as "massively improved performance" across many apps/games. As an extreme example, Tomb Raider (2013) went on Zink from running at a horrible 9 FPS to now at 91 FPS... Quite a big improvement and likely puts it closer to native OpenGL driver performance.
This suballocator code was recently merged to his zink-wip development tree. We'll see when this code manages to make it to mainline Mesa as unfortunately there is a ton of exciting improvements to this OpenGL-on-Vulkan implementation still residing within this work-in-progress branch.
Mike began work on a suballocator for Zink that is based on the Gallium3D auxiliary/pipebuffer code originally started by the RadeonSI Gallium3D driver. After making significant changes to that code, Zink's new suballocator implementation is showing off significant performance improvements in just shy of 700 lines of new code.
Blumenkrantz summed up the impact in a new blog post as "massively improved performance" across many apps/games. As an extreme example, Tomb Raider (2013) went on Zink from running at a horrible 9 FPS to now at 91 FPS... Quite a big improvement and likely puts it closer to native OpenGL driver performance.
This suballocator code was recently merged to his zink-wip development tree. We'll see when this code manages to make it to mainline Mesa as unfortunately there is a ton of exciting improvements to this OpenGL-on-Vulkan implementation still residing within this work-in-progress branch.
25 Comments