We didn't have it wired up running as we were recording, but the model does run faster than realtime! We have it running in batch mode for our interactive demo right now, but it can do streaming in realtime with ~100ms latency.