My understanding is yes, we commit to the log in batches, but abort conflicts at the transaction level. So only the conflicts have to suffer the retry loop, everything else is durably committed.