As described here we could use dedicated present queues alongside a compute queue for post-processing to start rendering the next frame, before the current frame gets post-processed and presented.