I refer to 2 documentation pages on PostgreSQL's site: #1 and #2. From these, we see the transaction log is flushed to disk 5 times per second (wal_writer_delay = 200ms), and we also see the RDBMS won't report a transaction as successful until it has been committed to the log file since synchronous_commit = on by default.
This would give us a theoretical peak of 5 reportable successfully writes per second if each is done in its own transaction, which is often the case. The reason being the log is not written to disk faster than that.
Yet we see speed in the thousands per second.
How does this work?
To be clear, I'm talking about each write being in its own transaction (the default usage). I know why bulk inserts don't apply to this question.