00:37 <cheako> Perhaps the ppl here would know, I was watching Travis Vroman and he seemed to be saying updates to a UBO from host needed to be followed by a write to descriptor? I don't think that's correct.

00:41 <airlied> would think you might need a barrier

01:12 <cheako> I'm not sure, access would be controlled by a fence or timeline.

01:23 <cheako> I got it he had two ubo and one image. Three descriptors, but he had three buffers so each frame would have it's own. So the write is to point to that present image ubo buffer.

01:24 <cheako> Can you write to a descriptor set that's part of a running queue?

01:25 <zmike> with EXT_descriptor_indexing and update-after-bind sometimes

12:30 <cheako> That led me to this article: https://zeux.io/2020/02/27/writing-an-efficient-vulkan-renderer/ where `It’s the responsibility of the application to manage the descriptor sets to make sure that CPU doesn’t update a descriptor set that’s in use by the GPU, ...`

12:31 <cheako> I don't know if that means "write", but that's how I understood things.