We'll handle them like f64 vertex buffers, by falling back to copying vertex data to the command buffer through translate.