4 ms·

Does this also happen during machine learning workloads?