accelerationStructureIndirectBuild on all platforms - Vulkan Hardware Database by Sascha Willems
No support still from NVidia on this feature? AMD and Intel ARC fully supported. Very powerful feature to have for fully gpu-driven raytraced engine architectures. Any plans for this to finally come to NVidia?
When I check my CPU vs GPU profiler, updating the TLAS adds to my CPU timers, i.e. it’s all calculated on the host CPU prior to internally staging it / updating the GPU copy of the TLAS, even though it’s a Vulkan command submitted through a command buffer (not all commands get executed on the GPU – see Sascha Williams’ recent comments/updates in his VK samples).
So in light of this, if you update the BLAS’ input vertex / index buffers on the GPU, you still need to update the TLAS to reflect those changes, and that would mean the driver needs to sync it back to the host CPU RAM, logically. I could be wrong about that. But I wouldn’t hold my breath hoping for an answer here, the Nvidia graphics Discord is probably a better place.