Now that we do not allow arbitrary large buffers (to avoid crashing the GPU), the initial boundary needs to be large enough for smooth decoding.